1.2 The Process of Science

Learning objectives.

  • Identify the shared characteristics of the natural sciences
  • Understand the process of scientific inquiry
  • Compare inductive reasoning with deductive reasoning
  • Describe the goals of basic science and applied science

Like geology, physics, and chemistry, biology is a science that gathers knowledge about the natural world. Specifically, biology is the study of life. The discoveries of biology are made by a community of researchers who work individually and together using agreed-on methods. In this sense, biology, like all sciences is a social enterprise like politics or the arts. The methods of science include careful observation, record keeping, logical and mathematical reasoning, experimentation, and submitting conclusions to the scrutiny of others. Science also requires considerable imagination and creativity; a well-designed experiment is commonly described as elegant, or beautiful. Like politics, science has considerable practical implications and some science is dedicated to practical applications, such as the prevention of disease (see Figure 1.15 ). Other science proceeds largely motivated by curiosity. Whatever its goal, there is no doubt that science, including biology, has transformed human existence and will continue to do so.

The Nature of Science

Biology is a science, but what exactly is science? What does the study of biology share with other scientific disciplines? Science (from the Latin scientia, meaning "knowledge") can be defined as knowledge about the natural world.

Science is a very specific way of learning, or knowing, about the world. The history of the past 500 years demonstrates that science is a very powerful way of knowing about the world; it is largely responsible for the technological revolutions that have taken place during this time. There are however, areas of knowledge and human experience that the methods of science cannot be applied to. These include such things as answering purely moral questions, aesthetic questions, or what can be generally categorized as spiritual questions. Science cannot investigate these areas because they are outside the realm of material phenomena, the phenomena of matter and energy, and cannot be observed and measured.

The scientific method is a method of research with defined steps that include experiments and careful observation. The steps of the scientific method will be examined in detail later, but one of the most important aspects of this method is the testing of hypotheses. A hypothesis is a suggested explanation for an event, which can be tested. Hypotheses, or tentative explanations, are generally produced within the context of a scientific theory . A generally accepted scientific theory is thoroughly tested and confirmed explanation for a set of observations or phenomena. Scientific theory is the foundation of scientific knowledge. In addition, in many scientific disciplines (less so in biology) there are scientific laws , often expressed in mathematical formulas, which describe how elements of nature will behave under certain specific conditions. There is not an evolution of hypotheses through theories to laws as if they represented some increase in certainty about the world. Hypotheses are the day-to-day material that scientists work with and they are developed within the context of theories. Laws are concise descriptions of parts of the world that are amenable to formulaic or mathematical description.

Natural Sciences

What would you expect to see in a museum of natural sciences? Frogs? Plants? Dinosaur skeletons? Exhibits about how the brain functions? A planetarium? Gems and minerals? Or maybe all of the above? Science includes such diverse fields as astronomy, biology, computer sciences, geology, logic, physics, chemistry, and mathematics ( Figure 1.16 ). However, those fields of science related to the physical world and its phenomena and processes are considered natural sciences . Thus, a museum of natural sciences might contain any of the items listed above.

There is no complete agreement when it comes to defining what the natural sciences include. For some experts, the natural sciences are astronomy, biology, chemistry, earth science, and physics. Other scholars choose to divide natural sciences into life sciences , which study living things and include biology, and physical sciences , which study nonliving matter and include astronomy, physics, and chemistry. Some disciplines such as biophysics and biochemistry build on two sciences and are interdisciplinary.

Scientific Inquiry

One thing is common to all forms of science: an ultimate goal “to know.” Curiosity and inquiry are the driving forces for the development of science. Scientists seek to understand the world and the way it operates. Two methods of logical thinking are used: inductive reasoning and deductive reasoning.

Inductive reasoning is a form of logical thinking that uses related observations to arrive at a general conclusion. This type of reasoning is common in descriptive science. A life scientist such as a biologist makes observations and records them. These data can be qualitative (descriptive) or quantitative (consisting of numbers), and the raw data can be supplemented with drawings, pictures, photos, or videos. From many observations, the scientist can infer conclusions (inductions) based on evidence. Inductive reasoning involves formulating generalizations inferred from careful observation and the analysis of a large amount of data. Brain studies often work this way. Many brains are observed while people are doing a task. The part of the brain that lights up, indicating activity, is then demonstrated to be the part controlling the response to that task.

Deductive reasoning or deduction is the type of logic used in hypothesis-based science. In deductive reasoning, the pattern of thinking moves in the opposite direction as compared to inductive reasoning. Deductive reasoning is a form of logical thinking that uses a general principle or law to predict specific results. From those general principles, a scientist can deduce and predict the specific results that would be valid as long as the general principles are valid. For example, a prediction would be that if the climate is becoming warmer in a region, the distribution of plants and animals should change. Comparisons have been made between distributions in the past and the present, and the many changes that have been found are consistent with a warming climate. Finding the change in distribution is evidence that the climate change conclusion is a valid one.

Both types of logical thinking are related to the two main pathways of scientific study: descriptive science and hypothesis-based science. Descriptive (or discovery) science aims to observe, explore, and discover, while hypothesis-based science begins with a specific question or problem and a potential answer or solution that can be tested. The boundary between these two forms of study is often blurred, because most scientific endeavors combine both approaches. Observations lead to questions, questions lead to forming a hypothesis as a possible answer to those questions, and then the hypothesis is tested. Thus, descriptive science and hypothesis-based science are in continuous dialogue.

Hypothesis Testing

Biologists study the living world by posing questions about it and seeking science-based responses. This approach is common to other sciences as well and is often referred to as the scientific method. The scientific method was used even in ancient times, but it was first documented by England’s Sir Francis Bacon (1561–1626) ( Figure 1.17 ), who set up inductive methods for scientific inquiry. The scientific method is not exclusively used by biologists but can be applied to almost anything as a logical problem-solving method.

The scientific process typically starts with an observation (often a problem to be solved) that leads to a question. Let’s think about a simple problem that starts with an observation and apply the scientific method to solve the problem. One Monday morning, a student arrives at class and quickly discovers that the classroom is too warm. That is an observation that also describes a problem: the classroom is too warm. The student then asks a question: “Why is the classroom so warm?”

Recall that a hypothesis is a suggested explanation that can be tested. To solve a problem, several hypotheses may be proposed. For example, one hypothesis might be, “The classroom is warm because no one turned on the air conditioning.” But there could be other responses to the question, and therefore other hypotheses may be proposed. A second hypothesis might be, “The classroom is warm because there is a power failure, and so the air conditioning doesn’t work.”

Once a hypothesis has been selected, a prediction may be made. A prediction is similar to a hypothesis but it typically has the format “If . . . then . . . .” For example, the prediction for the first hypothesis might be, “ If the student turns on the air conditioning, then the classroom will no longer be too warm.”

A hypothesis must be testable to ensure that it is valid. For example, a hypothesis that depends on what a bear thinks is not testable, because it can never be known what a bear thinks. It should also be falsifiable , meaning that it can be disproven by experimental results. An example of an unfalsifiable hypothesis is “Botticelli’s Birth of Venus is beautiful.” There is no experiment that might show this statement to be false. To test a hypothesis, a researcher will conduct one or more experiments designed to eliminate one or more of the hypotheses. This is important. A hypothesis can be disproven, or eliminated, but it can never be proven. Science does not deal in proofs like mathematics. If an experiment fails to disprove a hypothesis, then we find support for that explanation, but this is not to say that down the road a better explanation will not be found, or a more carefully designed experiment will be found to falsify the hypothesis.

Each experiment will have one or more variables and one or more controls. A variable is any part of the experiment that can vary or change during the experiment. A control is a part of the experiment that does not change. Look for the variables and controls in the example that follows. As a simple example, an experiment might be conducted to test the hypothesis that phosphate limits the growth of algae in freshwater ponds. A series of artificial ponds are filled with water and half of them are treated by adding phosphate each week, while the other half are treated by adding a salt that is known not to be used by algae. The variable here is the phosphate (or lack of phosphate), the experimental or treatment cases are the ponds with added phosphate and the control ponds are those with something inert added, such as the salt. Just adding something is also a control against the possibility that adding extra matter to the pond has an effect. If the treated ponds show lesser growth of algae, then we have found support for our hypothesis. If they do not, then we reject our hypothesis. Be aware that rejecting one hypothesis does not determine whether or not the other hypotheses can be accepted; it simply eliminates one hypothesis that is not valid ( Figure 1.18 ). Using the scientific method, the hypotheses that are inconsistent with experimental data are rejected.

In recent years a new approach of testing hypotheses has developed as a result of an exponential growth of data deposited in various databases. Using computer algorithms and statistical analyses of data in databases, a new field of so-called "data research" (also referred to as "in silico" research) provides new methods of data analyses and their interpretation. This will increase the demand for specialists in both biology and computer science, a promising career opportunity.

Visual Connection

In the example below, the scientific method is used to solve an everyday problem. Which part in the example below is the hypothesis? Which is the prediction? Based on the results of the experiment, is the hypothesis supported? If it is not supported, propose some alternative hypotheses.

  • My toaster doesn’t toast my bread.
  • Why doesn’t my toaster work?
  • There is something wrong with the electrical outlet.
  • If something is wrong with the outlet, my coffeemaker also won’t work when plugged into it.
  • I plug my coffeemaker into the outlet.
  • My coffeemaker works.

In practice, the scientific method is not as rigid and structured as it might at first appear. Sometimes an experiment leads to conclusions that favor a change in approach; often, an experiment brings entirely new scientific questions to the puzzle. Many times, science does not operate in a linear fashion; instead, scientists continually draw inferences and make generalizations, finding patterns as their research proceeds. Scientific reasoning is more complex than the scientific method alone suggests.

Basic and Applied Science

The scientific community has been debating for the last few decades about the value of different types of science. Is it valuable to pursue science for the sake of simply gaining knowledge, or does scientific knowledge only have worth if we can apply it to solving a specific problem or bettering our lives? This question focuses on the differences between two types of science: basic science and applied science.

Basic science or “pure” science seeks to expand knowledge regardless of the short-term application of that knowledge. It is not focused on developing a product or a service of immediate public or commercial value. The immediate goal of basic science is knowledge for knowledge’s sake, though this does not mean that in the end it may not result in an application.

In contrast, applied science or “technology,” aims to use science to solve real-world problems, making it possible, for example, to improve a crop yield, find a cure for a particular disease, or save animals threatened by a natural disaster. In applied science, the problem is usually defined for the researcher.

Some individuals may perceive applied science as “useful” and basic science as “useless.” A question these people might pose to a scientist advocating knowledge acquisition would be, “What for?” A careful look at the history of science, however, reveals that basic knowledge has resulted in many remarkable applications of great value. Many scientists think that a basic understanding of science is necessary before an application is developed; therefore, applied science relies on the results generated through basic science. Other scientists think that it is time to move on from basic science and instead to find solutions to actual problems. Both approaches are valid. It is true that there are problems that demand immediate attention; however, few solutions would be found without the help of the knowledge generated through basic science.

One example of how basic and applied science can work together to solve practical problems occurred after the discovery of DNA structure led to an understanding of the molecular mechanisms governing DNA replication. Strands of DNA, unique in every human, are found in our cells, where they provide the instructions necessary for life. During DNA replication, new copies of DNA are made, shortly before a cell divides to form new cells. Understanding the mechanisms of DNA replication enabled scientists to develop laboratory techniques that are now used to identify genetic diseases, pinpoint individuals who were at a crime scene, and determine paternity. Without basic science, it is unlikely that applied science could exist.

Another example of the link between basic and applied research is the Human Genome Project, a study in which each human chromosome was analyzed and mapped to determine the precise sequence of DNA subunits and the exact location of each gene. (The gene is the basic unit of heredity represented by a specific DNA segment that codes for a functional molecule.) Other organisms have also been studied as part of this project to gain a better understanding of human chromosomes. The Human Genome Project ( Figure 1.19 ) relied on basic research carried out with non-human organisms and, later, with the human genome. An important end goal eventually became using the data for applied research seeking cures for genetically related diseases.

While research efforts in both basic science and applied science are usually carefully planned, it is important to note that some discoveries are made by serendipity, that is, by means of a fortunate accident or a lucky surprise. Penicillin was discovered when biologist Alexander Fleming accidentally left a petri dish of Staphylococcus bacteria open. An unwanted mold grew, killing the bacteria. The mold turned out to be Penicillium , and a new critically important antibiotic was discovered. In a similar manner, Percy Lavon Julian was an established medicinal chemist working on a way to mass produce compounds with which to manufacture important drugs. He was focused on using soybean oil in the production of progesterone (a hormone important in the menstrual cycle and pregnancy), but it wasn't until water accidentally leaked into a large soybean oil storage tank that he found his method. Immediately recognizing the resulting substance as stigmasterol, a primary ingredient in progesterone and similar drugs, he began the process of replicating and industrializing the process in a manner that has helped millions of people. Even in the highly organized world of science, luck—when combined with an observant, curious mind focused on the types of reasoning discussed above—can lead to unexpected breakthroughs.

Reporting Scientific Work

Whether scientific research is basic science or applied science, scientists must share their findings for other researchers to expand and build upon their discoveries. Communication and collaboration within and between sub disciplines of science are key to the advancement of knowledge in science. For this reason, an important aspect of a scientist’s work is disseminating results and communicating with peers. Scientists can share results by presenting them at a scientific meeting or conference, but this approach can reach only the limited few who are present. Instead, most scientists present their results in peer-reviewed articles that are published in scientific journals. Peer-reviewed articles are scientific papers that are reviewed, usually anonymously by a scientist’s colleagues, or peers. These colleagues are qualified individuals, often experts in the same research area, who judge whether or not the scientist’s work is suitable for publication. The process of peer review helps to ensure that the research described in a scientific paper or grant proposal is original, significant, logical, and thorough. Grant proposals, which are requests for research funding, are also subject to peer review. Scientists publish their work so other scientists can reproduce their experiments under similar or different conditions to expand on the findings.

There are many journals and the popular press that do not use a peer-review system. A large number of online open-access journals, journals with articles available without cost, are now available many of which use rigorous peer-review systems, but some of which do not. Results of any studies published in these forums without peer review are not reliable and should not form the basis for other scientific work. In one exception, journals may allow a researcher to cite a personal communication from another researcher about unpublished results with the cited author’s permission.

As an Amazon Associate we earn from qualifying purchases.

This book may not be used in the training of large language models or otherwise be ingested into large language models or generative AI offerings without OpenStax's permission.

Want to cite, share, or modify this book? This book uses the Creative Commons Attribution License and you must attribute OpenStax.

Access for free at https://openstax.org/books/concepts-biology/pages/1-introduction
  • Authors: Samantha Fowler, Rebecca Roush, James Wise
  • Publisher/website: OpenStax
  • Book title: Concepts of Biology
  • Publication date: Apr 25, 2013
  • Location: Houston, Texas
  • Book URL: https://openstax.org/books/concepts-biology/pages/1-introduction
  • Section URL: https://openstax.org/books/concepts-biology/pages/1-2-the-process-of-science

© Apr 26, 2024 OpenStax. Textbook content produced by OpenStax is licensed under a Creative Commons Attribution License . The OpenStax name, OpenStax logo, OpenStax book covers, OpenStax CNX name, and OpenStax CNX logo are not subject to the Creative Commons license and may not be reproduced without the prior and express written consent of Rice University.

Logo for University of Minnesota Libraries

Want to create or adapt books like this? Learn more about how Pressbooks supports open publishing practices.

3 The Process of Science in Biology

By the end of this section, you will be able to do the following:

  • Identify the shared characteristics of the natural sciences
  • Summarize the steps of the scientific method
  • Compare inductive reasoning with deductive reasoning
  • Describe the goals of basic science and applied science

Photo A depicts round colonies of blue-green algae. Each algae cell is about 5 microns across. Photo B depicts round fossil structures called stromatalites along a watery shoreline.

What is biology? In simple terms, biology is the study of living organisms and their interactions with one another and their environments. This is a very broad definition because the scope of biology is vast. Biologists may study anything from the microscopic or submicroscopic view of a cell to ecosystems and the whole living planet (Figure 1). Listening to the daily news, you will quickly realize how many aspects of biology we discuss every day. For example, recent news topics include Escherichia coli (Figure 2) outbreaks in spinach and Salmonella contamination in peanut butter. Other subjects include efforts toward finding a cure for AIDS, Alzheimer’s disease, and cancer. On a global scale, many researchers are committed to finding ways to protect the planet, solve environmental issues, and reduce the effects of climate change. All of these diverse endeavors are related to different facets of the discipline of biology.

Photo depicts E. coli bacteria aggregated together.

The Process of Science

Biology is a science, but what exactly is science? What does the study of biology share with other scientific disciplines? We can define science (from the Latin scientia , meaning “knowledge”) as the knowledge that covers general truths or the operation of general laws, especially when acquired and tested by the scientific method. It becomes clear from this definition that applying the scientific method plays a major role in science. The scientific method is a method of research with defined steps that include experiments and careful observation.

We will examine the scientific method steps in detail later, but one of the most important aspects of this method is the testing of hypotheses by means of repeatable experiments. A hypothesis is a suggested explanation for an event, that is both testable and falsifiable. Although using the scientific method is inherent to science, it is inadequate in determining what science is. This is because it is relatively easy to apply the scientific method to disciplines such as physics and chemistry, but when it comes to disciplines like archaeology, psychology, and geology, the scientific method becomes less applicable as repeating experiments becomes more difficult.

These areas of study are still sciences, however. Consider archaeology—even though one cannot perform repeatable experiments, hypotheses may still be supported. For instance, an archaeologist can hypothesize that an ancient culture existed based on finding a piece of pottery. He or she could make further hypotheses about various characteristics of this culture, which could be correct or false through continued support or contradictions from other findings.

It can take a while before a hypothesis becomes proven and widely accepted by the scientific community. In general, scientists attribute different degrees of confidence in scientific evidence-based the quality or quantity of the research and data supporting a given conclusion. Some scientists, especially in medicine, have codified these different sources of information into a hierarchy of scientific evidence (Figure 3) [ 1 ].

Hierarchy of Scientific Evidence

With enough evidence, a concept or explanation can become the highest form of scientific understanding: a theory .

What Is a Scientific Theory?

A scientific theory [ 2 ] is a broad explanation of events that is widely accepted by the scientific community. To become a theory, an explanation must be strongly supported by a great deal of evidence.

People commonly use the word theory to describe a guess or hunch about how or why something happens. For example, you might say, “I think a woodchuck dug this hole in the ground, but it’s just a theory.” Using the word theory in this way is different from the way it is used in science. A scientific theory is not just a guess or hunch that may or may not be true. In science, a theory is an explanation that has a high likelihood of being correct because it is so well supported by evidence.

What is a scientific theory?

Natural Sciences

What would you expect to see in a museum of natural sciences? Frogs? Plants? Dinosaur skeletons? Exhibits about how the brain functions? A planetarium? Gems and minerals? Maybe all of the above? Science includes such diverse fields as astronomy, biology, computer sciences, geology, logic, physics, chemistry, and mathematics (Figure 4). However, scientists consider those fields of science related to the physical world and its phenomena and processes in natural sciences . Thus, a museum of natural sciences might contain any of the items listed above.

A collage includes a photo of planets in our solar system, a DNA molecule, scientific equipment, a cross-section of the ocean floor, scientific symbols, a magnetic field, beakers of fluid, and a geometry problem.

There is no complete agreement when it comes to defining what the natural sciences include, however. For some experts, the natural sciences are astronomy, biology, chemistry, earth science, and physics. Other scholars choose to divide natural sciences into life sciences , which study living things and include biology, and physical sciences , which study nonliving matter and include astronomy, geology, physics, and chemistry. Some disciplines such as biophysics and biochemistry build on life and physical sciences and are interdisciplinary. Some refer to natural sciences as “hard science” because they rely on the use of quantitative data. Social sciences that study society and human behavior are more likely to use qualitative assessments to drive investigations and findings.

Not surprisingly, the natural science of biology has many branches or subdisciplines. Cell biologists study cell structure and function, while biologists who study anatomy investigate the structure of an entire organism. Those biologists studying physiology, however, focus on the internal functioning of an organism. Some areas of biology focus on only particular types of living things. For example, botanists explore plants, while zoologists specialize in animals.

Scientific Reasoning

One thing is common to all forms of science: an ultimate goal is “to know.” Curiosity and inquiry are the driving forces for the development of science. Scientists seek to understand the world and the way it operates. To do this, they use two methods of logical thinking: inductive reasoning and deductive reasoning.

Inductive reasoning is a form of logical thinking that uses related observations to arrive at a general conclusion. This type of reasoning is common in descriptive science. A life scientist such as a biologist makes observations and records them. These data can be qualitative or quantitative, and one can supplement the raw data with drawings, pictures, photos, or videos. From many observations, the scientist can infer conclusions (inductions) based on evidence. Inductive reasoning involves formulating generalizations inferred from careful observation and analyzing a large amount of data. Brain studies provide an example. In this type of research, scientists observe many live brains while people are engaged in a specific activity, such as viewing images of food. The scientist then predicts the part of the brain that “lights up” during this activity to be the part controlling the response to the selected stimulus, in this case, images of food. Excess absorption of radioactive sugar derivatives by active areas of the brain causes the various areas to “light up”. Scientists use a scanner to observe the resultant increase in radioactivity. Then, researchers can stimulate that part of the brain to see if similar responses result.

Deductive reasoning or deduction is the type of logic used in hypothesis-based science . In deductive reasoning, the pattern of thinking moves in the opposite direction as compared to inductive reasoning. Deductive reasoning is a form of logical thinking that uses a general principle or law to forecast specific results. From those general principles, a scientist can extrapolate and predict the specific results that would be valid as long as the general principles are valid. Studies in climate change can illustrate this type of reasoning. For example, scientists may predict that if the climate becomes warmer in a particular region, then the distribution of plants and animals should change.

Both types of logical thinking are related to the two main pathways of scientific study: descriptive science and hypothesis-based science. Descriptive (or discovery) science , which is usually inductive, aims to observe, explore, and discover, while hypothesis-based science , which is usually deductive, begins with a specific question or problem and a potential answer or solution that one can test. The boundary between these two forms of study is often blurred, and most scientific endeavors combine both approaches. The fuzzy boundary becomes apparent when thinking about how easily observation can lead to specific questions. For example, a gentleman in the 1940s observed that the burr seeds that stuck to his clothes and his dog’s fur had a tiny hook structure. On closer inspection, he discovered that the burrs’ gripping device was more reliable than a zipper. He eventually experimented to find the best material that acted similarly and produced the hook-and-loop fastener popularly known today as Velcro. Descriptive science and hypothesis-based science are in continuous dialogue.

The Scientific Method

Biologists study the living world by posing questions about it and seeking science-based responses. Known as the scientific method, this approach is common to other sciences as well. The scientific method was used even in ancient times, but England’s Sir Francis Bacon (1561–1626) first documented it (Figure 5). He set up inductive methods for scientific inquiry. The scientific method is not used only by biologists; researchers from almost all fields of study can apply it as a logical, rational problem-solving method.

Painting depicts Sir Francis Bacon in a long robe.

The scientific process typically starts with an observation (often a problem to solve) that leads to a question. Let’s think about a simple problem that starts with an observation and apply the scientific method to solve the problem. One Monday morning, a student arrives at class and quickly discovers that the classroom is too warm. That is an observation that also describes a problem: the classroom is too warm. The student then asks a question: “Why is the classroom so warm?”

Proposing a Hypothesis

Recall that a hypothesis is a suggested explanation that can be tested and falsified. A good hypothesis is specific and includes clear variables that can be measured. For a given question, one can propose several hypotheses.

Let’s consider an example. You notice the classroom you teach in is warmer than usual. One hypothesis might be, “The temperature of the classroom is warmer because no one turned on the air conditioning.” Alternatively, a second hypothesis might be, “The temperature in the classroom is warmer because there is a power failure, and so the air conditioning doesn’t work.” While the cause is the same–no air conditioning–the variables are different: the air conditioner status (on/off) versus the power supply (present/absent). To find a solution, you need to isolate the problem.

The example above might seem simplistic–trouble-shooting an HVAC is not science… right? In fact, it is science or at least one part of the scientific process. And it illustrates the generality of scientific thinking in humans. Science is simply a methodology for problem-solving and collecting knowledge. It’s not the only system of knowledge or even the best per se , but it is one we regularly employ in our daily lives without even realizing it.

Once you have selected a hypothesis, you can make a prediction. A prediction is similar to a hypothesis but it typically has the format “If . . . then . . . .” For example, the prediction for the first hypothesis might be, “ If you turn on the air conditioning, then the classroom will no longer be too warm.” Note how this relates to the testability and falsifiability of the original hypothesis. You are testing the hypothesis by flipping the air conditioner switch. If you switch it on, and nothing happens, then the hypothesis is falsified, and it’s time to call an electrician to test the second hypothesis.

Testing a Hypothesis

To test a hypothesis, researchers design an experiment or analysis designed to validate or reject the hypothesis. In addition to the original hypothesis, researchers typically identify a null hypothesis. A null hypothesis represents the expectation if the proposed explanation is wrong. In our example from above, the competing null hypothesis would be “the power failure and loss of air conditioning does not cause the room to be warm.”

There are many types of experiments and analyses researchers conduct to test hypotheses. The general structure of most of these experiments or analyses, however, involves examining the effect of one variable on another. A variable is any part of the experiment that can vary or change during the course of the experiment. The variable of interest is referred to as the dependent variable . In our example, the dependent variable would be the temperature of the classroom. The independent variable is the condition the researcher purposefully changes to see how it affected the dependent variable. In our example, the independent variable would be the status of the air conditioner. Variables other than the independent variable that might nonetheless affect the dependent variable are referred to as confounding factors . A well-designed experiment will attempt to minimize the effect of confounding factors so that the researcher can be confident that the independent variable is the one causing the change in the dependent variable. It is not always possible to eliminate every confounding factor in a single experiment, however, and researchers must often run multiple experiments to ensure that something other than what they think is going on is actually occurring.

The most basic experimental design involves two groups, a control group and an experimental group . The control group represents the unmanipulated study condition, while the experimental group is somehow manipulated to test the effect of the independent variable. Otherwise, differences between the groups are limited to reduce any potential confounding variables. If the experimental group’s results differ from the control group, the difference must be due to the hypothesized manipulation, rather than some outside factor.  If the groups do not differ, then the independent variable has no effect, and the null hypothesis would be supported.

Look for the variables and controls in the examples that follow. To test the first hypothesis, the student would find out if the air conditioning is on. If the air conditioning is turned on but does not work, there should be another reason, and the student should reject this hypothesis. To test the second hypothesis, the student could check if the lights in the classroom are functional. If so, there is no power failure and the student should reject this hypothesis. The students should test each hypothesis by carrying out appropriate experiments. Be aware that rejecting one hypothesis does not determine whether or not one can accept the other hypotheses. It simply eliminates one hypothesis that is not valid (Figure 5). Using the scientific method, the student rejects hypotheses that are inconsistent with experimental data.

While this “warm classroom” example is based on observational results, other hypotheses and experiments might have clearer controls. For instance, a student might attend class on Monday and realize she had difficulty concentrating on the lecture. One observation to explain this occurrence might be, “When I eat breakfast before class, I am better able to pay attention.” The student could then design an experiment with a control to test this hypothesis.

To determine if the results of their experiment are significant, researchers use a variety of statistical analyses. Statistical analyses help researchers determine whether the observations from their experiments are meaningful or due to random chance. For example, if a researcher observes a difference between the control group and experimental group, should they treat it as a real effect of the independent variable or simply random chance? A result is considered to have statistical significance when it is very unlikely to have occurred given the null hypothesis. Statistical results themselves are not entirely objective and can depend on many assumptions including the null hypothesis itself. A researcher must consider potential biases in their analyses, just as they do confounding variables in their experimental design. Two factors that play a major role in the power of an experiment to detect meaningful statistical differences are sample size and replication. Sample size refers to the number of observations within each treatment, while replication refers to the number of repeated times the same experiment treatment is tried. In general, the bigger the sample size and the more replication, the more confidence a researcher can have in the outcome of their study.

In hypothesis-based science, researchers predict specific results from a general premise. We call this type of reasoning deductive reasoning: deduction proceeds from the general to the particular. However, the reverse of the process is also possible: sometimes, scientists reach a general conclusion from a number of specific observations. We call this type of reasoning inductive reasoning, and it proceeds from the particular to the general. Researchers often use inductive and deductive reasoning in tandem to advance scientific knowledge.


A flow chart shows the steps in the scientific method. In step 1, an observation is made. In step 2, a question is asked about the observation. In step 3, an answer to the question, called a hypothesis, is proposed. In step 4, a prediction is made based on the hypothesis. In step 5, an experiment is done to test the prediction. In step 6, the results are analyzed to determine whether or not the hypothesis is correct. If the hypothesis is incorrect, another hypothesis is made. In either case, the results are reported.

In the example below, the scientific method is used to solve an everyday problem. Order the scientific method steps (numbered items) with the process of solving the everyday problem (lettered items). Based on the results of the experiment, is the hypothesis correct? If it is incorrect, propose some alternative hypotheses.

1. Observation a. There is something wrong with the electrical outlet.
2. Question b. If something is wrong with the outlet, my coffee maker also won’t work when plugged into it.
3. Hypothesis (proposed answer) c. My toaster doesn’t toast my bread.
4. Prediction d. I plug my coffee maker into the outlet.
5. Experiment e. My coffee maker works.
6. Result f. Why doesn’t my toaster work?

Answer: 1: C; 2: F; 3: A; 4: B; 5: D; 6: E. The original hypothesis is incorrect, as the coffee maker works when plugged into the outlet. Alternative hypotheses include that the toaster might be broken or that the toaster wasn’t turned on.

Diagram defines two types of reasoning. In inductive reasoning, a general conclusion is drawn from a number of observations. In deductive reasoning, specific results are predicted from a general premise. An example of inductive reasoning is given. In this example, three observations are made: (1) Members of a species are not all the same. (2) Individuals compete for resources. (3) Species are generally adapted to their environment. From these observations, the following conclusion is drawn: Individuals most adapted to their environment are more likely to survive and pass their traits on to the next generation. An example of deductive reasoning is also given. In this example, the general premise is that individuals most adapted to their environment are more likely to survive and pass their traits on to the next generation. From this premise, it is predicted that, if global climate change causes the temperature in an ecosystem to increase, those individuals better adapted to a warmer climate will outcompete those that are not.

The scientific method may seem too rigid and structured. It is important to keep in mind that, although scientists often follow this sequence, there is flexibility. Sometimes an experiment leads to conclusions that favor a change in approach. Often, an experiment brings entirely new scientific questions to the puzzle. Many times, science does not operate in a linear fashion. Instead, scientists continually draw inferences and make generalizations, finding patterns as their research proceeds. Scientific reasoning is more complex than the scientific method alone suggests. Notice, too, that we can apply the scientific method to solving problems that aren’t necessarily scientific in nature.

Two Types of Science: Basic Science and Applied Science

The scientific community has been debating for the last few decades about the value of different types of science. Is it valuable to pursue science for the sake of simply gaining knowledge, or does scientific knowledge only have worth if we can apply it to solving a specific problem or to bettering our lives? This question focuses on the differences between two types of science: basic science and applied science.

Basic science or “pure” science seeks to expand knowledge regardless of the short-term application of that knowledge. It is not focused on developing a product or a service of immediate public or commercial value. The immediate goal of basic science is knowledge for knowledge’s sake, although this does not mean that, in the end, it may not result in a practical application.

In contrast, applied science or “technology,” aims to use science to solve real-world problems, making it possible, for example, to improve a crop yield, find a cure for a particular disease, or save animals threatened by a natural disaster (Figure 6). In applied science, the problem is usually defined for the researcher.

Image shows a squirrel being held by a person.

Some individuals may perceive applied science as “useful” and basic science as “useless.” A question these people might pose to a scientist advocating knowledge acquisition would be, “What for?” However, a careful look at the history of science reveals that basic knowledge has resulted in many remarkable applications of great value. Many scientists think that a basic understanding of science is necessary before researchers develop an application therefore, applied science relies on the results that researchers generate through basic science. Other scientists think that it is time to move on from basic science in order to find solutions to actual problems. Both approaches are valid. It is true that there are problems that demand immediate attention; however, scientists would find few solutions without the help of the wide knowledge foundation that basic science generates.

One example of how basic and applied science can work together to solve practical problems occurred after the discovery of DNA structure led to an understanding of the molecular mechanisms governing DNA replication. DNA strands, unique in every human, are in our cells, where they provide the instructions necessary for life. When DNA replicates, it produces new copies of itself, shortly before a cell divides. Understanding DNA replication mechanisms enabled scientists to develop laboratory techniques that researchers now use to identify genetic diseases, pinpoint individuals who were at a crime scene, and determine paternity. Without basic science, it is unlikely that applied science would exist.

Another example of the link between basic and applied research is the Human Genome Project, a study in which researchers analyzed and mapped each human chromosome to determine the precise sequence of DNA subunits and each gene’s exact location. (The gene is the basic unit of heredity. An individual’s complete collection of genes is his or her genome.) Researchers have studied other less complex organisms as part of this project in order to gain a better understanding of human chromosomes. The Human Genome Project (Figure 7) relied on basic research with simple organisms and, later, with the human genome. An important end goal eventually became using the data for applied research, seeking cures and early diagnoses for genetically related diseases.

The human genome projects logo is shown, depicting a human being inside a D N A double helix. The words chemistry, biology, physics, ethics, informatics, and engineering surround the circular image.

While scientists usually carefully plan research efforts in both basic science and applied science, note that some discoveries are made by serendipity , that is, by means of a fortunate accident or a lucky surprise. Scottish biologist Alexander Fleming discovered penicillin when he accidentally left a petri dish of Staphylococcus bacteria open. An unwanted mold grew on the dish, killing the bacteria. Fleming’s curiosity to investigate the reason behind the bacterial death, followed by his experiments, led to the discovery of the antibiotic penicillin, which is produced by the fungus Penicillium . Even in the highly organized world of science, luck—when combined with an observant, curious mind—can lead to unexpected breakthroughs.

Reporting Scientific Work

Whether scientific research is basic science or applied science, scientists must share their findings in order for other researchers to expand and build upon their discoveries. Collaboration with other scientists—when planning, conducting, and analyzing results—are all important for scientific research. For this reason, important aspects of a scientist’s work are communicating with peers and disseminating results to peers. Scientists can share results by presenting them at a scientific meeting or conference, but this approach can reach only the select few who are present. Instead, most scientists present their results in peer-reviewed manuscripts that are published in scientific journals. Peer-reviewed manuscripts are scientific papers that a scientist’s colleagues or peers review. These colleagues are qualified individuals, often experts in the same research area, who judge whether or not the scientist’s work is suitable for publication. The process of peer review helps to ensure that the research in a scientific paper or grant proposal is original, significant, logical, and thorough. Grant proposals, which are requests for research funding, are also subject to peer review. Scientists publish their work so other scientists can reproduce their experiments under similar or different conditions to expand on the findings. The experimental results must be consistent with the findings of other scientists.

A scientific paper is very different from creative writing. Although creativity is required to design experiments, there are fixed guidelines when it comes to presenting scientific results. First, scientific writing must be brief, concise, and accurate. A scientific paper needs to be succinct but detailed enough to allow peers to reproduce the experiments.

The scientific paper consists of several specific sections—introduction, materials and methods, results, and discussion. This structure is sometimes called the “IMRaD” format. There are usually acknowledgment and reference sections as well as an abstract (a concise summary) at the beginning of the paper. There might be additional sections depending on the type of paper and the journal where it will be published. For example, some review papers require an outline.

The introduction starts with brief, but broad, background information about what is known in the field. A good introduction also gives the rationale of the work. It justifies the work carried out and also briefly mentions the end of the paper, where the researcher will present the hypothesis or research question driving the research. The introduction refers to the published scientific work of others and therefore requires citations following the style of the journal. Using the work or ideas of others without proper citation is plagiarism .

The materials and methods section includes a complete and accurate description of the substances the researchers use and the methods and techniques they use to gather data. The description should be thorough enough to allow another researcher to repeat the experiment and obtain similar results, but it does not have to be verbose. This section will also include information on how the researchers made measurements and the types of calculations and statistical analyses they used to examine raw data. Although the materials and methods section gives an accurate description of the experiments, it does not discuss them.

Some journals require a results section followed by a discussion section, but it is more common to combine both. If the journal does not allow combining both sections, the results section simply narrates the findings without any further interpretation. The researchers present results with tables or graphs, but they do not present duplicate information. In the discussion section, the researchers will interpret the results, describe how variables may be related, and attempt to explain the observations. It is indispensable to conduct an extensive literature search to put the results in the context of previously published scientific research. Therefore, researchers include proper citations in this section as well.

Finally, the conclusion section summarizes the importance of the experimental findings. While the scientific paper almost certainly answers one or more scientific questions that the researchers stated, any good research should lead to more questions. Therefore, a well-done scientific paper allows the researchers and others to continue and expand on the findings.

Review articles do not follow the IMRAD format because they do not present original scientific findings or primary literature. Instead, they summarize and comment on findings that were published as primary literature and typically include extensive reference sections.

Review of the scientific process

Introductory Biology: Evolutionary and Ecological Perspectives Copyright © by Various Authors - See Each Chapter Attribution is licensed under a Creative Commons Attribution 4.0 International License , except where otherwise noted.

Share This Book

  • De%73cribe the natu%72%65 of %73cientific inquir%79%2E
  • Compare q%75%61ntitative and qualita%74%69ve data.
  • %53%75%6Dmarize the nature of %64%69%73covery %73cience.
  • Di%73tingui%73h betw%65%65n ob%73ervation%73 and in%66%65rence%73.
  • Ex%70%6Cain the term gener%61%6Cization .
  • %6F%62%73ervation
  • infere%6E%63e
  • generalization

Any %73cience tex%74%62ook, including thi%73 o%6E%65, i%73 packed with info%72%6Dation ba%73ed on what %73%63%69enti%73t%73 have di%73cover%65%64 in the pa%73t. Indeed,%20%73cience ha%73 built an i%6D%70re%73%73ive body of knowl%65%64ge that continue%73 to %69%6Ecrea%73e and change wit%68%20new di%73coverie%73. Much%20%6Ff what'%73 known i%73 fa%73%63%69nating, but the real %66%75n in %73cience begin%73 w%68%65n you turn from what'%73%20 known to what'%73%20 unknown .

%53cience a%73 %49%6Equiry %0ABiol%6F%67y i%73 defined a%73 the %73%63%69entific %73tudy of life%2E%20But what doe%73 %73cie%6E%74ific mean? What i%73%20%73cience? The word i%73 %64%65rived from a Latin ve%72%62 meaning %22to know.%22 I%6E%20other word%73, %73cience %69%73 a way of knowing. It%20%69%73 a way to an%73wer que%73%74ion%73 about the natura%6C%20world.

At th%65%20heart of %73cience i%73 i%6E%71uiry—people a%73ki%6E%67 que%73tion%73 about what%20%74hey ob%73erve in nature%20%61nd actively %73eeking a%6E%73wer%73. For example, ha%76%65 you ever noticed tha%74%20mo%73t hou%73eplant%73 grow%20%74oward a light %73ource,%20%73uch a%73 a window? Rota%74%65 the plant, and it%73 d%69%72ection of growth will%20%73hift until the leave%73%20%61gain face the window.%20%53uch ob%73ervation%73 in%73p%69%72e que%73tion%73. How doe%73%20%74he plant %73en%73e the di%72%65ction of light? What %65%6Eable%73 the plant to be%6E%64 toward light a%73 it g%72%6Fw%73? In what direction%20%77ould a plant grow in %74%68e dark?

Your %6F%77n curio%73ity i%73 the %73t%61%72ting point for explor%69%6Eg life through inquir%79%2E But inquiry mean%73 mo%72%65 than a%73king que%73tion%73%2E Inquiry i%73 a proce%73%73%20%6Ff inve%73tigation, with%20%74houghtful que%73tion%73 l%65%61ding to a %73earch for %61%6E%73wer%73. A%73king que%73tio%6E%73 i%73 a natural activit%79%20for all curiou%73 mind%73%2C%20but even figuring out%20%77hat to a%73k take%73 prac%74%69ce. You can develop t%68%69%73 and other %73kill%73 th%61%74 %73upport %73cientific i%6E%71uiry through the acti%76%69tie%73 on the Biolog%79%3A Exploring Life W%65%62 %73ite and through you%72%20laboratory inve%73tigat%69%6Fn%73. By the end of thi%73%20%73chool year, you'll h%61%76e plenty of experienc%65%20with %73cience a%73 a pro%63%65%73%73 of inquiry.

All ob%73ervation%73%20%64epend on human %73en%73e%73%2E%20But, without help the%20%73en%73e%73 are too limited%20%74o penetrate %73ome of t%68%65 mo%73t intere%73ting rea%6C%6D%73 of nature. %53cientif%69%63 in%73trument%73 va%73tly i%6E%63rea%73e the range of po%73%73ible ob%73ervation%73. In%20%61%73tronomy, tele%73cope%73 %72%65veal crater%73 on the m%6F%6Fn. In biology, micro%73%63%6Fpe%73 make it po%73%73ible %74%6F ob%73erve life that i%73%20%69nvi%73ible to the unaid%65%64 eye. Other equipment%20%65nable%73 human%73 to ob%73e%72%76e DNA and other molec%75%6Ce%73.

Ob%73ervati%6F%6E%73 are often recorded %61%73 mea%73urement%73, al%73o c%61%6Cled quantitative

Dat%61%20al%73o may be qualit%61%74ive —that i%73,%20%69n the form of de%73crip%74%69on%73 in%73tead of mea%73ur%65%6Dent%73. For example, Ja%6E%65 Goodall %73pent decade%73%20recording her ob%73erva%74%69on%73 of chimpanzee beh%61%76ior in a jungle in Ga%6D%62ia, an ea%73t African n%61%74ion. In addition to k%65%65ping careful note%73 a%73%20%64ata in her field note%62%6Fok%73, Goodall al%73o doc%75%6Dented her ob%73ervation%73%20with photograph%73 and %6D%6Fvie%73. Data can be%73t %73%75%70port %73cience when the%79%20are clearly organized%2C%20con%73i%73tently recorded%2C%20and reliable.

In contra%73t to%20%74he carefully planned %6D%61pping of human DNA, o%62%73ervant people %73ometim%65%73 di%73cover %73omething i%6D%70ortant about nature e%6E%74irely by accident. On%65%20famou%73 example i%73 Ale%78%61nder Fleming'%73 1928 d%69%73covery that certain f%75%6Egi produce chemical%73 %74%68at kill bacteria. Fle%6D%69ng, a %53cotti%73h phy%73ic%69%61n, wa%73 culturing (gro%77%69ng) bacteria for re%73e%61%72ch in hi%73 laboratory.%20%48e found that a mold (%61%20type of fungu%73) had c%6F%6Etaminated %73ome of hi%73%20%63ulture%73 of bacteria. %41%73 he wa%73 di%73carding th%65%20%22%73poiled%22 culture%73, F%6C%65ming noticed that no %62%61cteria were growing n%65%61r the mold. The fungu%73%20turned out to be P%65%6Eicillium , a commo%6E%20mold. It produce%73 an %61%6Etibacterial %73ub%73tance%20%74hat wa%73 later named p%65%6Eicillin. Fleming'%73 ac%63%69dental di%73covery revo%6C%75tionized medicine. Pe%6E%69cillin proved to be j%75%73t one of many life%73av%69%6Eg antibiotic%73 that ar%65%20made by fungi and oth%65%72 organi%73m%73. The%73e dru%67%73 help treat %73trep thr%6F%61t, bacterial pneumoni%61%2C %73yphili%73, and many o%74%68er di%73ea%73e%73 cau%73ed by%20%62acteria. The u%73e of a%6E%74ibiotic%73 ha%73 greatly %65%78tended the average hu%6D%61n life%73pan in many co%75%6Etrie%73.

Infer%65%6Ece%73 in %53cience %3C%62r> %0AA logical conclu%73%69%6Fn ba%73ed on ob%73ervatio%6E%73 i%73 called an inference . Ofte%6E%2C a per%73on make%73 an in%66%65rence by relating ob%73%65%72vation%73 to hi%73 or her%20%70rior knowledge. For i%6E%73tance, you infer %73ome%6F%6Ee i%73 at the door when%20%79ou hear the doorbell %72%69ng becau%73e you know t%68%65 %73ame thing ha%73 happe%6E%65d before. Examine the%20%70icnic table in Figure%20%32-6 on page 27 of your%20%74extbook. What can you%20%69nfer from the place %73%65%74ting%73 and other objec%74%73 you ob%73erve on the t%61%62le? Can you infer any%74%68ing from what i%73 a%62%73ent ? Can you make%20%72ea%73onable inference%73 %61%62out the weather and t%69%6De of day when thi%73 ph%6F%74ograph wa%73 taken?

Inference%73 are imp%6F%72tant in %73cience becau%73%65 they help refine gen%65%72al que%73tion%73 into %73pe%63%69fic que%73tion%73 that ca%6E%20be explored further. %46%6Fr example, a %73cienti%73%74%20might a%73k: %22What %73ub%73%74%61nce produced by thi%73 %70%61rticular mold kill%73 b%61%63teria?%22 However, keep%20%69n mind that %73cienti%73t%73%20are %73keptical of infe%72%65nce%73 that %22%73tretch%22 f%61%72 beyond the data. An %65%78ample would be inferr%69%6Eg, %73olely from%20%46leming'%73 ob%73ervation,%20%74hat %73ome mold%73 could %62%65 u%73ed to produce anti%62%69otic%73 capable of curi%6E%67 bacterial di%73ea%73e%73 i%6E%20human%73. It took much %6D%6Fre re%73earch before th%69%73 conclu%73ion wa%73 accep%74%65d among %73cienti%73t%73. A%6C%73o, it i%73 important no%74%20to confu%73e inference%73%20%77ith the ob%73ervation%73 %6F%6E which they are ba%73ed%2E%20Hearing the doorbell %72%69ng i%73 an ob%73ervation.%20%49nferring that %73omeone%20%69%73 at the door, though%20%72ea%73onable, ha%73 le%73%73 c%65%72tainty. Maybe an elec%74%72ical %73hort circuit i%73%20%63au%73ing the bell to ri%6E%67.

It i%73 al%73o %73%6F%6Detime%73 po%73%73ible to ge%6E%65ralize from quantitat%69%76e data. Thi%73 u%73ually %72%65quire%73 pooling (combi%6E%69ng) mea%73urement%73 from%20%61 very large %73ample. T%6F%20look for general patt%65%72n%73 in mea%73urement%73, i%74%20often help%73 to put th%65%20data in a graph. For %65%78ample, the graph in F%69%67ure 2-8 compare%73 the %63%68ange%73 in height%73 of t%65%65nage boy%73 and girl%73 o%76%65r time. Each point on%20%74he graph i%73 an averag%65%20mea%73urement for many %74%68ou%73and boy%73 or girl%73.%20%54he graph make%73 it ea%73%69%65r to %73pot the general%20%70attern that girl%73, , %73top g%72%6Fwing at a younger age%20%74han boy%73. Of cour%73e, %74%68ere are many individu%61%6C exception%73. %53ome female%73 do continue%20%74o grow well pa%73t the %61%76erage age when male%73 %73%74op growing. But the g%65%6Eeralization %73till hol%64%73 acro%73%73 the very larg%65%20%73ample of teen%73.

Concept Ch%65%63k 2.1 %0A How doe%73 %73cient%69%66ic inquiry differ fro%6D%20%73imply a%73king que%73tio%6E%73? %0A 2. Are%20%74he data recorded in t%68%65 table in Figure 2-2 %71%75antitative or qualita%74%69ve? Explain. %0A How i%73 Jane Goo%64%61ll'%73 work an example %6F%66 di%73covery %73cience? 4. De%73cribe%20%61n ob%73ervation you mad%65%20today and an inferenc%65%20you can make from tha%74%20ob%73ervation. %0A How are the ter%6D%73 generalization ob%73ervation %0A

SEP logo

  • Table of Contents
  • New in this Archive
  • Chronological
  • Editorial Information
  • About the SEP
  • Editorial Board
  • How to Cite the SEP
  • Special Characters
  • Support the SEP
  • PDFs for SEP Friends
  • Make a Donation
  • SEPIA for Libraries
  • Entry Contents


Academic tools.

  • Friends PDF Preview
  • Author and Citation Info
  • Back to Top

Scientific Discovery

Scientific discovery is the process or product of successful scientific inquiry. Objects of discovery can be things, events, processes, causes, and properties as well as theories and hypotheses and their features (their explanatory power, for example). Most philosophical discussions of scientific discoveries focus on the generation of new hypotheses that fit or explain given data sets or allow for the derivation of testable consequences. Philosophical discussions of scientific discovery have been intricate and complex because the term “discovery” has been used in many different ways, both to refer to the outcome and to the procedure of inquiry. In the narrowest sense, the term “discovery” refers to the purported “eureka moment” of having a new insight. In the broadest sense, “discovery” is a synonym for “successful scientific endeavor” tout court. Some philosophical disputes about the nature of scientific discovery reflect these terminological variations.

Philosophical issues related to scientific discovery arise about the nature of human creativity, specifically about whether the “eureka moment” can be analyzed and about whether there are rules (algorithms, guidelines, or heuristics) according to which such a novel insight can be brought about. Philosophical issues also arise about rational heuristics, about the characteristics of hypotheses worthy of articulation and testing, and, on the meta-level, about the nature and scope of philosophical reflection itself. This essay describes the emergence and development of the philosophical problem of scientific discovery, surveys different philosophical approaches to understanding scientific discovery, and presents the meta-philosophical problems surrounding the debates.

1. Introduction

2. scientific inquiry as discovery, 3. elements of discovery, 4. logics of discovery, 5. the distinction between the context of discovery and the context of justification, 6.1 discovery as abduction, 6.2 heuristic programming, 7. anomalies and the structure of discovery, 8.1 discoverability, 8.2 preliminary appraisal, 9.1 psychological and social conditions of creativity, 9.2 analogy, 9.3 mental models, other internet resources, related entries.

Philosophical reflection on scientific discovery occurred in different phases. Prior to the 1930s, philosophers were mostly concerned with discoveries in the broadest sense of the term, that is, with the analysis of successful scientific inquiry as a whole. Philosophical discussions focused on the question of whether there were any discernible patterns in the production of new knowledge. Because the concept of discovery did not have a specified meaning and was used in a very broad sense, almost all seventeenth- and eighteenth-century treatises on scientific method could potentially be considered as early contributions to reflections on scientific discovery. In the course of the 19 th century, as philosophy of science and science became two distinct endeavors, the term “discovery” became a technical term in philosophical discussions. Different elements of scientific inquiry were specified. Most importantly, the generation of new knowledge was clearly and explicitly distinguished from its validation, and thus the conditions for the narrower notion of discovery as the act of conceiving new ideas emerged.

The next phase in the discussion about scientific discovery began with the introduction of the so-called “context distinction,” the distinction between the “context of discovery” and the “context of justification”. It was further assumed that the act of conceiving a new idea is a non-rational process, a leap of insight that cannot be regulated. Justification, by contrast, is a systematic process of applying evaluative criteria to knowledge claims. Advocates of the context distinction argued that philosophy of science is exclusively concerned with the context of justification. The assumption underlying this argument is that philosophy is a normative project; it determines norms for scientific practice. Given these assumptions, only the justification of ideas, not their generation, can be the subject of philosophical (normative) analysis. Discovery, by contrast, can only be a topic for empirical study. By definition, the study of discovery is outside the scope of philosophy of science proper.

The introduction of the context distinction and the disciplinary distinction that was tied to it spawned meta-philosophical disputes. For a long time, philosophical debates about discovery were shaped by the notion that philosophical and empirical analyses are mutually exclusive. A number of philosophers insisted, like their predecessors prior to the 1930s, that the philosopher's tasks include the analysis of actual scientific practices and that scientific resources be used to address philosophical problems. They also maintained that it is a legitimate task for philosophy of science to develop a theory of heuristics or problem solving. But this position was the minority view during much of 20 th -century philosophy of science. Philosophers of discovery were thus compelled to demonstrate that scientific discovery was in fact a legitimate part of philosophy of science. Philosophical reflections about the nature of scientific discovery had to be bolstered by meta-philosophical arguments about the nature and scope of philosophy of science.

Today, however, there is wide agreement that philosophy and empirical research are not mutually exclusive. Not only do empirical studies of actual scientific discoveries inform philosophical thought about the structure and cognitive mechanisms of discovery, but researches in psychology, cognitive science, artificial intelligence and related fields have become an integral part of philosophical analyses of the processes and conditions of the generation of new knowledge.

Prior to the 19 th century, the term “discovery” commonly referred to the product of successful inquiry. “Discovery” was used broadly to refer to a new finding, such as a new cure, an improvement of an instrument, or a new method of measuring longitude. Several natural and experimental philosophers, notably Bacon, Descartes, and Newton, expounded accounts of scientific methods for arriving at new knowledge. These accounts were not explicitly labeled “methods of discovery ”, but the general accounts of scientific methods are nevertheless relevant for current philosophical debates about scientific discovery. They are relevant because philosophers of science have frequently presented 17 th -century theories of scientific method as a contrast class to current philosophies of discovery. The distinctive feature of the 17 th - and 18 th -century accounts of scientific method is that the methods have probative force (Nickles 1985). This means that those accounts of scientific method function as guides for acquiring new knowledge and at the same time as validations of the knowledge thus obtained (Laudan 1980; Schaffner 1993: chapter 2).

Bacon's account of his “new method” as it is presented in the Novum Organum is a prominent example. Bacon's work showed how best to arrive at knowledge about “form natures” (the most general properties of matter) via a systematic investigation of phenomenal natures. Bacon described how first to collect and organize natural phenomena and experimental facts in tables, how to evaluate these lists, and how to refine the initial results with the help of further experiments. Through these steps, the investigator would arrive at conclusions about the “form nature” that produces particular phenomenal natures. The point is that for Bacon, the procedures of constructing and evaluating tables and conducting experiments according to the Novum Organum leads to secure knowledge. The procedures thus have “probative force”.

Similarly, Newton's aim in the Philosophiae Naturalis Principia Mathematica was to present a method for the deduction of propositions from phenomena in such a way that those propositions become “more secure” than propositions that are secured by deducing testable consequences from them (Smith 2002). Newton did not assume that this procedure would lead to absolute certainty. One could only obtain moral certainty for the propositions thus secured. The point for current philosophers of science is that these approaches are generative theories of scientific method. Generative theories of scientific method assume that propositions can only be established and secured by showing that they follow from observed and experimentally produced phenomena. In contrast, non-generative theories of scientific method—such as the one proposed by Huygens—assumed that propositions must be established by comparing their consequences with observed and experimentally produced phenomena. In 20 th -century philosophy of science, this approach is often characterized as “consequentialist” (Laudan 1980; Nickles 1985).

Recent philosophers of science have used historical sketches like these to construct the prehistory of current philosophical debates about scientific discovery. The argument is that scientific discovery became a problem for philosophy of science in the 19 th century, when consequentialist theories of scientific method became more widespread. When consequentialist theories were on the rise, the two processes of conception and validation of an idea or hypothesis became distinct and the view that the merit of a new idea does not depend on the way in which it was arrived at became widely accepted.

In the course of the 19 th century, the act of having an insight—the purported “eureka moment”—was separated from processes of articulating, developing, and testing the novel insight. Philosophical discussion focused on the question of whether and to what extent rules could be devised to guide each of these processes. William Whewell's work, especially the two volumes of Philosophy of the Inductive Sciences of 1840, is an important contribution to the philosophical debates about scientific discovery precisely because he clearly separated the creative moment or “happy thought” as he called it from other elements of scientific inquiry. For Whewell, discovery comprised all three elements: the happy thought, the articulation and development of that thought, and the testing or verification of it. In most of the subsequent treatments of discovery, however, the scope of the term “discovery” is limited to either the first of these elements, the “happy thought”, or to the first two of these elements, the happy thought and its articulation. In fact, much of the controversies in the 20 th century about the possibility of a philosophy of discovery can be understood against the background of the disagreement about whether the process of discovery does or does not include the articulation and development of a novel thought.

The previous section shows that scholars like Bacon and Newton aimed to develop methodologies of scientific inquiry. They proposed “new methods” or “rules of reasoning” that guide the generation of certain propositions from observed and experimental phenomena. Whewell, by contrast, was explicitly concerned with developing a philosophy of discovery. His account was in part a description of the psychological makeup of the discoverer. For instance, he held that only geniuses could have those happy thoughts that are essential to discovery. In part, his account was an account of the methods by which happy thoughts are integrated into the system of knowledge. According to Whewell, the initial step in every discovery is what he called “some happy thought, of which we cannot trace the origin, some fortunate cast of intellect, rising above all rules. No maxims can be given which inevitably lead to discovery” (Whewell 1996 [1840]: 186). An “art of discovery” in the sense of a teachable and learnable skill does not exist according to Whewell. The happy thought builds on the known facts, but according to Whewell it is impossible to prescribe a method for having happy thoughts.

In this sense, happy thoughts are accidental. But in an important sense, scientific discoveries are not accidental. The happy thought is not a “wild guess.” Only the person whose mind is prepared to see things will actually notice them. The “previous condition of the intellect, and not the single fact, is really the main and peculiar cause of the success. The fact is merely the occasion by which the engine of discovery is brought into play sooner or later. It is, as I have elsewhere said, only the spark which discharges a gun already loaded and pointed; and there is little propriety in speaking of such an accident as the cause why the bullet hits its mark.” (Whewell 1996 [1840]: 189).

Having a happy thought is not yet a discovery, however. The second element of a scientific discovery consists in binding together—“colligating”, as Whewell called it—a set of facts by bringing them under a general conception. Not only does the colligation produce something new, but it also shows the previously known facts in a new light. More precisely, colligation works from both ends, from the facts as well as from the ideas that bind the facts together. Colligation is an extended process. It involves, on the one hand, the specification of facts through systematic observation, measurements and experiment, and on the other hand, the clarification of ideas through the exposition of the definitions and axioms that are tacitly implied in those ideas. This process is iterative. The scientists go back and forth between binding together the facts, clarifying the idea, rendering the facts more exact, and so on and so forth.

The final part of the discovery is the verification of the colligation involving the happy thought. This means, first and foremost, that the outcome of the colligation must be sufficient to explain the data at hand. Verification also involves judging the predictive power, simplicity, and “consilience” of the outcome of the colligation. “Consilience” refers to a higher range of generality (broader applicability) of the theory (the articulated and clarified happy thought) that the actual colligation produced. Whewell's account of discovery is not a deductivist system. It is essential that the outcome of the colligation be inferable from the data prior to any testing (Snyder 1997).

Whewell's theory of discovery is significant for the philosophical debate about scientific discovery because it clearly separates three elements: the non-analyzable happy thought or “eureka moment”; the process of colligation which includes the clarification and explication of facts and ideas; and the verification of the outcome of the colligation. His position that the philosophy of discovery cannot prescribe how to think happy thoughts has been a key element of 20 th -century philosophical reflection on discovery, and many philosophers have adopted the notion “happy thought” as a label for the “eureka moment” involved in discovery. Notably, however, Whewell's conception of discovery not only comprises the happy thoughts but also the processes by which the happy thoughts are to be integrated into the given system of knowledge. The procedures of articulation and test are both analyzable according to Whewell, and his conception of colligation and verification serve as guidelines for how the discoverer should proceed. A colligation, if properly done, has as such justificatory force. Similarly, the process of verification is an integral part of discovery and it too has justificatory force. Whewell's conception of verification thus comprises elements of generative and consequential methods of inquiry. To verify a hypothesis, the investigator needs to show that it accounts for the known facts, that it foretells new, previously unobserved phenomena, and that it can explain and predict phenomena which are explained and predicted by a hypothesis that was obtained through an independent happy thought-cum-colligation (Ducasse 1951).

Whewell's conceptualization of scientific discovery offers a useful framework for mapping the philosophical debates about discovery and for identifying major issues of concern in recent philosophical debates. First and foremost, nearly all recent philosophers operate with a notion of discovery that is narrower than Whewell's. In the narrower conception, what Whewell called “verification” is not part of discovery proper. Secondly, until the late 20 th century, there was wide agreement that the “eureka moment,” narrowly construed, is an unanalyzable, even mysterious leap of insight. The main disagreements concerned the question of whether the process of developing a hypothesis (the “colligation” in Whewell's terms) is or is not a part of discovery proper, and if it is, whether and how this process is guided by rules. Philosophers also disagreed on the issue of whether it is a philosophical task to explicate these rules. In recent decades, philosophical attention has shifted to the “eureka moment”. Drawing on resources from cognitive science, neuroscience, computational research, and environmental and social psychology, they have “demystified” the cognitive processes involved in the generation of new ideas.

In the early 20 th century, the view that discovery is or at least crucially involves a non-analyzable creative act of a gifted genius was widespread but not unanimously accepted. Alternative conceptions of discovery emphasize that discovery is an extended process, i.e., that the discovery process includes the reasoning processes through which a new insight is articulated and further developed. Moreover, it was assumed that there is a systematic, formal aspect to that reasoning. While the reasoning involved does not proceed according to the principles of demonstrative logic, it is systematic enough to deserve the label “logical”. Proponents of this view argued that traditional (here: Aristotelian) logic is an inadequate model of scientific discovery because it misrepresents the process of knowledge generation as grossly as the notion of “happy thought”. In this approach, the term “logic” is used in the broad sense. It is the task of the logic of discovery to draw out and give a schematic representation of the reasoning strategies that were applied in episodes of successful scientific inquiry. Early 20 th -century logics of discovery can best be described as theories of the mental operations involved in knowledge generation. Among these mental operations are classification, determination of what is relevant to an inquiry, and the conditions of communication of meaning. It is argued that these features of scientific discovery are either not or insufficiently represented by traditional logic (Schiller 1917: 236–7).

Philosophers advocating this approach agree that the logic of discovery should be characterized as a set of heuristic principles rather than as a process of applying inductive or deductive logic to a set of propositions. These heuristic principles are not understood to show the path to secure knowledge. Heuristic principles are suggestive rather than demonstrative (Carmichael 1922, 1930). One recurrent feature in these accounts of the reasoning strategies leading to new ideas is analogical reasoning (Schiller 1917; Benjamin 1934). In the 20 th century, it is widely acknowledged that analogical reasoning is a productive form of reasoning that cannot be reduced to inductive or deductive reasoning. However, these approaches to the logic of discovery remained scattered and tentative at that time, and attempts to develop more systematically the heuristics guiding discovery processes were eclipsed by the advance of the distinction between contexts of discovery and justification.

The distinction between “context of discovery” and “context of justification” dominated and shaped the discussions about discovery in 20 th -century philosophy of science. The context distinction marks the distinction between the generation of a new idea or hypothesis and the defense (test, verification) of it. As the previous sections have shown, the distinction among different features of scientific inquiry has a longer history, but in philosophy of science it became potent in the first half of the 20 th century. In the course of the ensuing discussions about scientific discovery, the distinction between the different features of scientific inquiry turned into a powerful demarcation criterion. The boundary between context of discovery (the de facto thinking processes) and context of justification (the de jure defense of the correctness of these thoughts) was now understood to determine the scope of philosophy of science. The underlying assumption is that philosophy of science is a normative endeavor. Advocates of the context distinction argue that the generation of a new idea is an intuitive, irrational process; it cannot be subject to normative analysis. Therefore, the study of scientists' actual thinking can only be the subject of psychology, sociology, and other empirical sciences. Philosophy of science, by contrast, is exclusively concerned with the context of justification.

The terms “context of discovery” and “context of justification” are often associated with Hans Reichenbach's work. Reichenbach's original conception of the context distinction is quite complex, however (Howard 2006; Richardson 2006). It does not map easily on to the disciplinary distinction mentioned above, because for Reichenbach, philosophy of science proper is partly descriptive. Reichenbach maintains that philosophy of science includes a description of knowledge as it really is. Descriptive philosophy of science reconstructs scientists' thinking processes in such a way that logical analysis can be performed on them, and it thus prepares the ground for the evaluation of these thoughts (Reichenbach 1938: § 1). Discovery, by contrast, is the object of empirical—psychological, sociological—study. According to Reichenbach, the empirical study of discoveries shows that processes of discovery often correspond to the principle of induction, but this is simply a psychological fact (Reichenbach 1938: 403).

While the terms “context of discovery” and “context of justification” are widely used, there has been ample discussion about how the distinction should be drawn and what their philosophical significance is (c.f. Kordig 1978; Gutting 1980; Zahar 1983; Leplin 1987; Hoyningen-Huene 1987; Weber 2005: chapter 3; Schickore and Steinle 2006). Most commonly, the distinction is interpreted as a distinction between the process of conceiving a theory and the validation of that theory, that is, the determination of the theory's epistemic support. This version of the distinction is not necessarily interpreted as a temporal distinction. In other words, it is not usually assumed that a theory is first fully developed and then validated. Rather, conception and validation are two different epistemic approaches to theory: the endeavor to articulate, flesh out, and develop its potential and the endeavor to assess its epistemic worth. Within the framework of the context distinction, there are two main ways of conceptualizing the process of conceiving a theory. The first option is to characterize the generation of new knowledge as an irrational act, a mysterious creative intuition, a “eureka moment”. The second option is to conceptualize the generation of new knowledge as an extended process that includes a creative act as well as some process of articulating and developing the creative idea.

Both of these accounts of knowledge generation served as starting points for arguments against the possibility of a philosophy of discovery. In line with the first option, philosophers have argued that neither is it possible to prescribe a logical method that produces new ideas nor is it possible to reconstruct logically the process of discovery. Only the process of testing is amenable to logical investigation. This objection to philosophies of discovery has been called the “discovery machine objection” (Curd 1980: 207). It is usually associates with Karl Popper's Logic of Scientific Discovery .

The initial state, the act of conceiving or inventing a theory, seems to me neither to call for logical analysis not to be susceptible of it. The question how it happens that a new idea occurs to a man—whether it is a musical theme, a dramatic conflict, or a scientific theory—may be of great interest to empirical psychology; but it is irrelevant to the logical analysis of scientific knowledge. This latter is concerned not with questions of fact (Kant's quid facti ?) , but only with questions of justification or validity (Kant's quid juris ?) . Its questions are of the following kind. Can a statement be justified? And if so, how? Is it testable? Is it logically dependent on certain other statements? Or does it perhaps contradict them? […]Accordingly I shall distinguish sharply between the process of conceiving a new idea, and the methods and results of examining it logically. As to the task of the logic of knowledge—in contradistinction to the psychology of knowledge—I shall proceed on the assumption that it consists solely in investigating the methods employed in those systematic tests to which every new idea must be subjected if it is to be seriously entertained. (Popper 2002 [1934/1959]: 7-8)

With respect to the second way of conceptualizing knowledge generation, many philosophers argue in a similar fashion that because the process of discovery involves an irrational, intuitive process, which cannot be examined logically, a logic of discovery cannot be construed. Other philosophers turn against the philosophy of discovery even though they explicitly acknowledge that discovery is an extended, reasoned process. They present a meta-philosophical objection argument, arguing that a theory of articulating and developing ideas is not a philosophical but a psychological theory.

The impact of the context distinction on studies of scientific discovery and on philosophy of science more generally can hardly be overestimated. The view that the process of discovery (however construed) is outside the scope of philosophy of science proper was widely shared amongst philosophers of science for most of the 20 th century and is still held by many. The last section shows that there were a few attempts to develop logics of discovery in the 1920s and 1930s. But for several decades, the context distinction dictated what philosophy of science should be about and how it should proceed. The dominant view was that theories of mental operations or heuristics had no place in philosophy of science. Therefore, discovery was not a legitimate topic for philosophy of science. The wide notion of discovery is mostly deployed in sociological accounts of scientific practice. In this perspective, “discovery” is understood as a retrospective label, which is attributed as a sign of accomplishment to some scientific endeavors. Sociological theories acknowledge that discovery is a collective achievement and the outcome of a process of negotiation through which “discovery stories” are constructed and certain knowledge claims are granted discovery status (Brannigan 1981; Schaffer 1986, 1994). Until the last third of the 20 th century, there were few attempts to challenge the disciplinary distinction tied to the context distinction. Only in the 1970s did the interest in philosophical approaches to discovery begin to increase. But the context distinction remained a challenge for philosophies of discovery.

There are three main lines of response to the disciplinary distinction tied to the context distinction. Each of these lines of response opens up a philosophical perspective on discovery. Each proceeds on the assumption that philosophy of science may legitimately include some form of analysis of actual reasoning patterns as well as information from empirical sciences such as cognitive science, psychology, and sociology. All of these responses reject the idea that discovery is nothing but a mystical event. Discovery is conceived as an analyzable reasoning process, not just as a creative leap by which novel ideas spring into being fully formed. All of these responses agree that the procedures and methods for arriving at new hypotheses and ideas are no guarantee that the hypothesis or idea that is thus formed is necessarily the best or the correct one. Nonetheless, it is the task of philosophy of science to provide rules for making this process better. All of these responses can be described as theories of problem solving, whose ultimate goal is to make the generation of new ideas and theories more efficient.

But the different approaches to scientific discovery employ different terminologies. In particular, the term “logic” of discovery is sometimes used in a narrow sense and sometimes broadly understood. In the narrow sense, “logic” of discovery is understood to refer to a set of formal, generally applicable rules by which novel ideas can be mechanically derived from existing data. In the broad sense, “logic” of discovery refers to the schematic representation of reasoning procedures. “Logical” is just another term for “rational”. Moreover, while each of these responses combines philosophical analyses of scientific discovery with empirical research on actual human cognition, different sets of resources are mobilized, ranging from AI research and cognitive science to historical studies of problem-solving procedures. Also, the responses parse the process of scientific inquiry differently. Often, scientific inquiry is regarded as having two aspects, viz. generation and validation of new ideas. At times, however, scientific inquiry is regarded as having three aspects, namely generation, pursuit or articulation, and validation of knowledge. In the latter framework, the label “discovery” is sometimes used to refer just to generation and sometimes to refer to both generation and pursuit.

The first response to the challenge of the context distinction draws on a broad understanding of the term “logic” to argue that a logic of scientific discovery can be developed ( section 6 ). The second response, drawing on a narrow understanding of the term “logic”, is to concede that there is no logic of discovery, i.e., no algorithm for generating new knowledge. Philosophers who take this approach argue that the process of discovery follows an identifiable, analyzable pattern ( section 7 ). Others argue that discovery is governed by a methodology . The methodology of discovery is a legitimate topic for philosophical analysis ( section 8 ). All of these responses assume that there is more to discovery than a “eureka moment.” Discovery comprises processes of articulating and developing the creative thought. These are the processes that can be examined with the tools of philosophical analysis. The third response to the challenge of the context distinction also assumes that discovery is or at least involves a creative act. But in contrast to the first two responses, it is concerned with the creative act itself. Philosophers who take this approach argue that scientific creativity is amenable to philosophical analysis ( section 9 ).

6. Logics of discovery after the context distinction

The first response to the challenge of the context distinction is to argue that discovery is a topic for philosophy of science because it is a logical process after all. Advocates of this approach to the logic of discovery usually accept the overall distinction between the two processes of conceiving and testing a hypothesis. They also agree that it is impossible to put together a manual that provides a formal, mechanical procedure through which innovative concepts or hypotheses can be derived: There is no discovery machine. But they reject the view that the process of conceiving a theory is a creative act, a mysterious guess, a hunch, a more or less instantaneous and random process. Instead, they insist that both conceiving and testing hypotheses are processes of reasoning and systematic inference, that both of these processes can be represented schematically, and that it is possible to distinguish better and worse paths to new knowledge.

This line of argument has much in common with the logics of discovery described in section 4 above but it is now explicitly pitched against the disciplinary distinction tied to the context distinction. There are two main ways of developing this argument. The first is to conceive of discovery in terms of abductive reasoning ( section 6.1 ). The second is to conceive of discovery in terms of problem-solving algorithms, whereby heuristic rules aid the processing of available data and enhance the success in finding solutions to problems ( section 6.2 ). Both lines of argument rely on a broad conception of logic, whereby the “logic” of discovery amounts to a schematic account of the reasoning processes involved in knowledge generation.

One argument, elaborated prominently by Norwood R. Hanson, is that the act of discovery—here, the act of suggesting a new hypothesis—follows a distinctive logical pattern, which is different from both inductive logic and the logic of hypothetico-deductive reasoning. The special logic of discovery is the logic of abductive or “retroductive” inferences (Hanson 1958). The argument that it is through an act of abductive inferences that plausible, promising scientific hypotheses are devised goes back to C.S. Peirce. This version of the logic of discovery characterizes reasoning processes that take place before a new hypothesis is ultimately justified. The abductive mode of reasoning that leads to plausible hypotheses is conceptualized as an inference beginning with data or, more specifically, with surprising or anomalous phenomena.

In this view, discovery is primarily a process of explaining anomalies or surprising, astonishing phenomena. The scientists' reasoning proceeds abductively from an anomaly to an explanatory hypothesis in light of which the phenomena would no longer be surprising or anomalous. The outcome of this reasoning process is not one single specific hypothesis but the delineation of a type of hypotheses that is worthy of further attention (Hanson 1965: 64). According to Hanson, the abductive argument has the following schematic form (Hanson 1960: 104):

  • Some surprising, astonishing phenomena p 1 , p 2 , p 3 … are encountered.
  • But p 1 , p 2 , p 3 … would not be surprising were an hypothesis of H's type to obtain. They would follow as a matter of course from something like H and would be explained by it.
  • Therefore there is good reason for elaborating an hypothesis of type H—for proposing it as a possible hypothesis from whose assumption p 1 , p 2 , p 3 … might be explained.

Drawing on the historical record, Hanson argues that several important discoveries were made relying on abductive reasoning, such as Kepler's discovery of the elliptic orbit of Mars (Hanson 1958). It is now widely agreed, however, that Hanson's reconstruction of the episode is not a historically adequate account of Kepler's discovery (Lugg 1985). More importantly, while there is general agreement that abductive inferences are frequent in both everyday and scientific reasoning, these inferences are no longer considered as logical inferences. Even if one accepts Hanson's schematic representation of the process of identifying plausible hypotheses, this process is a “logical” process only in the widest sense whereby the term “logical” is understood as synonymous with “rational”. Notably, some philosophers have even questioned the rationality of abductive inferences (Koehler 1991; Brem and Rips 2000).

Another argument against the above schema is that it is too permissive. There will be several hypotheses that are explanations for phenomena p 1 , p 2 , p 3 …, so the fact that a particular hypothesis explains the phenomena is not a decisive criterion for developing that hypothesis (Harman 1965; see also Blackwell 1969). Additional criteria are required to evaluate the hypothesis yielded by abductive inferences.

Finally, it is worth noting that the schema of abductive reasoning does not explain the very act of conceiving a hypothesis or hypothesis-type. The processes by which a new idea is first articulated remain unanalyzed in the above schema. The schema focuses on the reasoning processes by which an exploratory hypothesis is assessed in terms of its merits and promise (Laudan 1980; Schaffner 1993).

In more recent work on abduction and discovery, two notions of abduction are sometimes distinguished: the common notion of abduction as inference to the best explanation (selective abduction) and creative abduction (Magnani 2000, 2009). Selective abduction—the inference to the best explanation—involves selecting a hypothesis from a set of known hypotheses. Medical diagnosis exemplifies this kind of abduction. Creative abduction, by contrast, involves generating a new, plausible hypothesis. This happens, for instance, in medical research, when the notion of a new disease is articulated. However, it is still an open question whether this distinction can be drawn, or whether there is a more gradual transition from selecting an explanatory hypothesis from a familiar domain (selective abduction) to selecting a hypothesis that is slightly modified from the familiar set and to identifying a more drastically modified or altered assumption.

Another recent suggestion is to broaden Peirce's original account of abduction and to include not only verbal information but also non-verbal mental representations, such as visual, auditory, or motor representations. In Thagard's approach, representations are characterized as patterns of activity in mental populations (see also section 9.3 below). The advantage of the neural account of human reasoning is that it covers features such as the surprise that accompanies the generation of new insights or the visual and auditory representations that contribute to it. If all mental representations can be characterized as patterns of firing in neural populations, abduction can be analyzed as the combination or “convolution” (Thagard) of patterns of neural activity from disjoint or overlapping patterns of activity (Thagard 2010).

The concern with the logic of discovery has also motivated research on artificial intelligence at the intersection of philosophy of science and cognitive science. In this approach, scientific discovery is treated as a form of problem-solving activity (Simon 1973; see also Newell and Simon 1971), whereby the systematic aspects of problem solving are studied within an information-processing framework. The aim is to clarify with the help of computational tools the nature of the methods used to discover scientific hypotheses. These hypotheses are regarded as solutions to problems. Philosophers working in this tradition build computer programs employing methods of heuristic selective search (e.g., Langley et al. 1987). In computational heuristics, search programs can be described as searches for solutions in a so-called “problem space” in a certain domain. The problem space comprises all possible configurations in that domain (e.g., for chess problems, all possible arrangements of pieces on a board of chess). Each configuration is a “state” of the problem space. There are two special states, namely the goal state, i.e., the state to be reached, and the initial state, i.e., the configuration at the starting point from which the search begins. There are operators, which determine the moves that generate new states from the current state. There are path constraints, which limit the permitted moves. Problem solving is the process of searching for a solution of the problem of how to generate the goal state from an initial state. In principle, all states can be generated by applying the operators to the initial state, then to the resulting state, until the goal state is reached (Langley et al. 1987: chapter 9). A problem solution is a sequence of operations leading from the initial to the goal state.

The basic idea behind computational heuristics is that rules can be identified that serve as guidelines for finding a solution to a given problem quickly and efficiently by avoiding undesired states of the problem space. These rules are best described as rules of thumb. The aim of constructing a logic of discovery thus becomes the aim of constructing a heuristics for the efficient search for solutions to problems. The term “heuristic search” indicates that in contrast to algorithms, problem-solving procedures lead to results that are merely provisional and plausible. A solution is not guaranteed, but heuristic searches are advantageous because they are more efficient than exhaustive random trial and error searches. Insofar as it is possible to evaluate whether one set of heuristics is better—more efficacious—than another, the logic of discovery turns into a normative theory of discovery.

Arguably, because it is possible to reconstruct important scientific discovery processes with sets of computational heuristics, the scientific discovery process can be considered as a special case of the general mechanism of information processing. In this context, the term “logic” is not used in the narrow sense of a set of formal, generally applicable rules to draw inferences but again in a broad sense as a label for a set of procedural rules.

The computer programs that embody the principles of heuristic searches in scientific inquiry simulate the paths that scientists followed when they searched for new theoretical hypotheses. Computer programs such as BACON (Simon et al. 1981) and KEKADA (Kulkarni and Simon 1988) utilize sets of problem-solving heuristics to detect regularities in given data sets. The program would note, for instance, that the values of a dependent term are constant or that a set of values for a term x and a set of values for a term y are linearly related. It would thus “infer” that the dependent term always has that value or that a linear relation exists between x and y . These programs can “make discoveries” in the sense that they can simulate successful discoveries such as Kepler's third law (BACON) or the Krebs cycle (KEKADA).

AI-based theories of scientific discoveries have helped identify and clarify a number of problem-solving strategies. An example of such a strategy is heuristic means-ends analysis, which involves identifying specific differences between the present and the goal situation and searches for operators (processes that will change the situation) that are associated with the differences that were detected. Another important heuristic is to divide the problem into sub-problems and to begin solving the one with the smallest number of unknowns to be determined (Simon 1977). AI-based approaches have also highlighted the extent to which the generation of new knowledge draws on existing knowledge that constrains the development of new hypotheses.

As accounts of scientific discoveries, computational heuristics have some limitations. Most importantly, because computer programs require the data from actual experiments the simulations cover only certain aspects of scientific discoveries. They do not design new experiments, instruments, or methods. Moreover, compared to the problem spaces given in computational heuristics, the complex problem spaces for scientific problems are often ill defined, and the relevant search space and goal state must be delineated before heuristic assumptions could be formulated (Bechtel and Richardson 1993: chapter 1).

Earlier critics of AI-based theories of scientific discoveries argued that a computer cannot devise new concepts but is confined to the concepts included in the given computer language (Hempel 1985: 119–120). Subsequent work has shown that computational methods can be used to generate new results leading to refereed scientific publications in astronomy, cancer research, ecology, and other fields (Langley 2000). The most recent computational research on scientific discovery is no longer driven by philosophical interests in scientific discovery, however. Instead, the main motivation is to contribute computational tools to aid scientists in their research.

Many philosophers maintain that discovery is a legitimate topic for philosophy of science while abandoning the notion that there is a logic of discovery. One very influential approach is Thomas Kuhn's analysis of the emergence of novel facts and theories (Kuhn 1970 [1962]: chapter 6). Kuhn identifies a general pattern of discovery as part of his account of scientific change. A discovery is not a simple act, but an extended, complex process, which culminates in paradigm changes. Paradigms are the symbolic generalizations, metaphysical commitments, values, and exemplars that are shared by a community of scientists and that guide the research of that community. Paradigm-based, normal science does not aim at novelty but instead at the development, extension, and articulation of accepted paradigms. A discovery begins with an anomaly, that is, with the recognition that the expectations induced by an established paradigm are being violated. The process of discovery involves several aspects: observations of an anomalous phenomenon, attempts to conceptualize it, and changes in the paradigm so that the anomaly can be accommodated.

It is the mark of success of normal science that it does not make transformative discoveries, and yet such discoveries come about as a consequence of normal, paradigm-guided science. The more detailed and the better developed a paradigm, the more precise are its predictions. The more precisely the researchers know what to expect, the better they are able to recognize anomalous results and violations of expectations:

novelty ordinarily emerges only for the man who, knowing with precision what he should expect, is able to recognize that something has gone wrong. Anomaly appears only against the background provided by the paradigm. (Kuhn 1970 [1962]: 65)

Drawing on several historical examples, Kuhn argues that it is usually impossible to identify the very moment when something was discovered or even the individual who made the discovery. Kuhn illustrates these points with the discovery of oxygen (see Kuhn 1970 [1962]: 53–56). Oxygen had not been discovered before 1774 and had been discovered by 1777. Even before 1774, Lavoisier had noticed that something was wrong with phlogiston theory, but he was unable to move forward. Two other investigators, C. W. Scheele and Joseph Priestley, independently identified a gas obtained from heating solid substances. But Scheele's work remained unpublished until after 1777, and Priestley did not identify his substance as a new sort of gas. In 1777, Lavoisier presented the oxygen theory of combustion, which gave rise to fundamental reconceptualization of chemistry. But according to this theory as Lavoisier first presented it, oxygen was not a chemical element. It was an atomic “principle of acidity” and oxygen gas was a combination of that principle with caloric. According to Kuhn, all of these developments are part of the discovery of oxygen, but none of them can be singled out as “the” act of discovery.

In pre-paradigmatic periods or in times of paradigm crisis, theory-induced discoveries may happen. In these periods, scientists speculate and develop tentative theories, which may lead to novel expectations and experiments and observations to test whether these expectations can be confirmed. Even though no precise predictions can be made, phenomena that are thus uncovered are often not quite what had been expected. In these situations, the simultaneous exploration of the new phenomena and articulation of the tentative hypotheses together bring about discovery.

In cases like the discovery of oxygen, by contrast, which took place while a paradigm was already in place, the unexpected becomes apparent only slowly, with difficulty, and against some resistance. Only gradually do the anomalies become visible as such. It takes time for the investigators to recognize “both that something is and what it is” (Kuhn 1970 [1962]: 55). Eventually, a new paradigm becomes established and the anomalous phenomena become the expected phenomena.

Recent studies in cognitive neuroscience of brain activity during periods of conceptual change support Kuhn's view that conceptual change is hard to achieve. These studies examine the neural processes that are involved in the recognition of anomalies and compare them with the brain activity involved in the processing of information that is consistent with preferred theories. The studies suggest that the two types of data are processed differently (Dunbar et al. 2007).

8. Methodologies of discovery

Advocates of the view that there are methodologies of discovery use the term “logic” in the narrow sense of an algorithmic procedure to generate new ideas. But like the AI-based theories of scientific discovery described in section 6 , methodologies of scientific discovery interpret the concept “discovery” as a label for an extended process of generating and articulating new ideas and often describe the process in terms of problem solving. In these approaches, the distinction between the contexts of discovery and the context of justification is challenged because the methodology of discovery is understood to play a justificatory role. Advocates of a methodology of discovery usually rely on a distinction between different justification procedures, justification involved in the process of generating new knowledge and justification involved in testing it. Consequential or “strong” justifications are methods of testing. The justification involved in discovery, by contrast, is conceived as generative (as opposed to consequential) justification ( section 8.1 ) or as weak (as opposed to strong) justification ( section 8.2 ). Again, some terminological ambiguity exists because according to some philosophers, there are three contexts, not two: Only the initial conception of a new idea (the “eureka moment”) is the context of discovery proper, and between it and justification there exists a separate context of pursuit (Laudan 1980). But many advocates of methodologies of discovery regard the context of pursuit as an integral part of the process of justification. They retain the notion of two contexts and re-draw the boundaries between the contexts of discovery and justification as they were drawn in the early 20 th century.

The methodology of discovery has sometimes been characterized as a form of justification that is complementary to the methodology of testing (Nickles 1984, 1985, 1989). According to the methodology of testing, empirical support for a theory results from successfully testing the predictive consequences derived from that theory (and appropriate auxiliary assumptions). In light of this methodology, justification for a theory is “consequential justification,” the notion that a hypothesis is established if successful novel predictions are derived from the theory or claim. Generative justification complements consequential justification. Advocates of generative justification hold that there exists an important form of justification in science that involves reasoning to a claim from data or previously established results more generally.

One classic example for a generative methodology is the set of Newton's rules for the study of natural philosophy. According to these rules, general propositions are established by deducing them from the phenomena. The notion of generative justification seeks to preserve the intuition behind classic conceptions of justification by deduction. Generative justification amounts to the rational reconstruction of the discovery path in order to establish its discoverability had the researchers known what is known now, regardless of how it was first thought of (Nickles 1985, 1989). The reconstruction demonstrates in hindsight that the claim could have been discovered in this manner had the necessary information and techniques been available. In other words, generative justification—justification as “discoverability” or “potential discovery”—justifies a knowledge claim by deriving it from results that are already established. While generative justification does not retrace exactly those steps of the actual discovery path that were actually taken, it is a better representation of scientists' actual practices than consequential justification because scientists tend to construe new claims from available knowledge. Generative justification is a weaker version of the traditional ideal of justification by deduction from the phenomena. Justification by deduction from the phenomena is complete if a theory or claim is completely determined from what we already know. The demonstration of discoverability results from the successful derivation of a claim or theory from the most basic and most solidly established empirical information.

Discoverability as described in the previous paragraphs is a mode of justification. Like the testing of novel predictions derived from a hypothesis, generative justification begins when the phase of finding and articulating a hypothesis worthy of assessing is drawing to a close. Other approaches to the methodology of discovery are directly concerned with the procedures involved in devising new hypotheses. The argument in favor of this kind of methodology is that the procedures of devising new hypotheses already include elements of appraisal. These preliminary assessments have been termed “weak” evaluation procedures (Schaffner 1993). Weak evaluations are relevant during the process of devising a new hypothesis. They provide reasons for accepting a hypothesis as promising and worthy of further attention. Strong evaluations, by contrast, provide reasons for accepting a hypothesis as (approximately) true or confirmed. Both “generative” and “consequential” testing as discussed in the previous section are strong evaluation procedures. Strong evaluation procedures are rigorous and systematically organized according to the principles of hypothesis derivation or H-D testing. A methodology of preliminary appraisal, by contrast, articulates criteria for the evaluation of a hypothesis prior to rigorous derivation or testing. It aids the decision about whether to take that hypothesis seriously enough to develop it further and test it. For advocates of this version of the methodology of discovery, it is the task of philosophy of science to characterize sets of constraints and methodological rules guiding the complex process of prior-to-test evaluation of hypotheses.

In contrast to the computational approaches discussed above, strategies of preliminary appraisal are not regarded as subject-neutral but as specific to particular fields of study. Because the analysis of criteria for the appraisal of hypotheses has mostly been made with regard to the study of biological mechanism, the criteria and constraints that have been proposed are those that play a role in the discovery of biological mechanisms. Biological mechanisms are entities and activities that are organized in such a way that they produce regular changes from initial to terminal conditions (Machamer et al. 2000).

Philosophers of biology have developed a fine-grained framework to account for the generation and preliminary evaluation of these mechanisms (Darden 2002; Craver 2002; Bechtel and Richardson 1993; Craver and Darden 2013). Some philosophers have even suggested that the phase of preliminary appraisal be further divided into two phases, the phase of appraising and the phase of revising. According to Lindley Darden, the phases of generation, appraisal and revision of descriptions of mechanisms can be characterized as reasoning processes governed by reasoning strategies. Different reasoning strategies govern the different phases (Darden 1991, 2002; Craver 2002; Darden 2009). The generation of hypotheses about mechanisms, for instance, is governed by the strategy of “schema instantiation” (see Darden 2002). The discovery of the mechanism of protein synthesis involved the instantiation of an abstract schema for chemical reactions: reactant 1 + reactant 2 = product. The actual mechanism of protein synthesis was found through specification and modification of this schema.

It is important to appreciate the status of these reasoning strategies. They are not necessarily strategies that were actually used. Neither of these strategies is deemed necessary for discovery, and they are not prescriptions for biological research. Rather, these strategies are deemed sufficient for the discovery of mechanisms; they “could have been used” to arrive at the description of that mechanism (Darden 2002). The methodology of the discovery of mechanisms is an extrapolation from past episodes of research on mechanisms and the result of a synthesis of rational reconstructions of several of these historical episodes. The methodology of discovery is only weakly normative in the sense that the strategies for the discovery of mechanisms that have been identified so far may prove useful in future biological research. Moreover, the sets of reasoning strategies that have been proposed are highly specific. It is still an open question whether the analysis of strategies for the discovery of biological mechanisms can illuminate the efficiency of scientific problem solving more generally (Weber 2005: chapter 3).

9. Creativity, analogy, and mental models

The approaches to scientific discovery presented in the previous sections focus on the adoption, articulation, and preliminary evaluation of ideas or hypotheses prior to rigorous testing. They do not illuminate how a novel hypothesis or idea is first thought up. Even among philosophers of discovery, the predominant view has long been that there is an initial step of discovery that is best described as a “eureka moment”, a mysterious intuitive leap of the human mind that cannot be analyzed further.

The concept of discovery as hypothesis-formation as it is encapsulated in the traditional distinction between context of discovery and context of justification does not explicate how new ideas are formed. According to accounts of discovery informed by evolutionary biology, the generation of new ideas is akin to random, blind variations of thought processes, which have to be inspected by the critical mind and assessed as neutral, productive, or useless (Campbell 1960; see also Hull 1988). While the evolutionary approach to discovery offers a more substantial account of scientific discovery, the key processes by which random ideas are generated are still left unanalyzed.

Today, many philosophers hold the view that creativity is not mysterious and can be submitted to analysis. Margaret Boden has offered helpful analyses of the concept of creativity. According to Boden, a new development is creative if it is novel, surprising, and important. She distinguishes between psychological creativity (P-creativity) and historical creativity (H-creativity). P-creativity is a development that is new, surprising and important to the particular person who comes up with it. H-creativity, by contrast, is radically novel, surprising, and important—it is generated for the first time (Boden 2004).

The majority of recent philosophical studies of scientific discovery today focus on the act of generation of new knowledge. The distinctive feature of these studies is that they integrate approaches from cognitive science, psychology, and computational neuroscience (Thagard 2012). Recent work on creativity offers substantive analyses of the social and psychological preconditions and the cognitive mechanisms involved in generating new ideas. Some of this research aims to characterize those features that are common to all creative processes. Other research aims to identify the features that are distinctive of scientific creativity (as opposed to other forms of creativity, such as artistic creativity or creative technological invention). Studies have focused on analyses of the personality traits that are conducive to creative thinking, and the social and environmental factors that are favorable for discovery ( section 9.1 ). Two key elements of the cognitive processes involved in creative thinking are analogies ( section 9.2 ) and mental models ( section 9.3 ).

Psychological studies of creative individuals' behavioral dispositions suggest that creative scientists share certain personality traits, including confidence, openness, dominance, independence, introversion, as well as arrogance and hostility. (For overviews of recent studies on personality traits of creative scientists, see Feist 1999, 2006: chapter 5). Social situatedness has also been explored as an important resource for creativity. In this perspective, the sociocultural structures and practices in which individuals are embedded are considered crucial for the generation of creative ideas. Both approaches suggest that creative individuals usually have outsider status—they are socially deviant and diverge from the mainstream.

Outsider status is also a key feature of standpoint. According to standpoint theorists, people with standpoint are politically aware and politically engaged people outside the mainstream. Some standpoint theorists suggest exploiting this similarity for creativity research. Because people with standpoint have different experiences and access to different domains of expertise than most members of a culture, they can draw on rich conceptual resources for creative thinking. Standpoint theory may thus be an important resource for the development of social and environmental approaches to the study of creativity (Solomon 2007).

Many philosophers of science highlight the role of analogy in the development of new knowledge, whereby analogy is understood as a process of bringing ideas that are well understood in one domain to bear on a new domain (Thagard 1984; Holyoak and Thagard 1996). An important source for philosophical thought about analogy is Mary Hesse's conception of models and analogies in theory construction and development. In this approach, analogies are similarities between different domains. Hesse introduces the distinction between positive, negative, and neutral analogies (Hesse 1966: 8). If we consider the relation between gas molecules and a model for gas, namely a collection of billiard balls in random motion, we will find properties that are common to both domains (positive analogy) as well as properties that can only be ascribed to the model but not to the target domain (negative analogy). There is a positive analogy between gas molecules and a collection of billiard balls because both the balls and the molecules move randomly. There is a negative analogy between the domains because billiard balls are colored, hard, and shiny but gas molecules do not have these properties. The most interesting properties are those properties of the model about which we do not know whether they are positive or negative analogies. This set of properties is the neutral analogy. These properties are the significant properties because they might lead to new insights about the less familiar domain. From our knowledge about the familiar billiard balls, we may be able to derive new predictions about the behavior of gas molecules, which we could then test.

Hesse offers a more detailed analysis of the structure of analogical reasoning through the distinction between horizontal and vertical analogies between domains. Horizontal analogies between two domains concern the sameness or similarity between properties of both domains. If we consider sound and light waves, there are similarities between them: sound echoes, light reflects; sound is loud, light is bright, both sound and light are detectable by our senses. There are also relations among the properties within one domain, such as the causal relation between sound and the loud tone we hear and, analogously, between physical light and the bright light we see. These analogies are vertical analogies. For Hesse, vertical analogies hold the key for the construction of new theories.

Analogies play several roles in science. Not only do they contribute to discovery but they also play a role in the development and evaluation of scientific theories. Current discussions about analogy and discovery have expanded and refined Hesse's approach in various ways. Some philosophers have developed criteria for evaluating analogy arguments (Bartha 2010). Other work has identified highly significant analogies that were particularly fruitful for the advancement of science (Holyoak and Thagard 1996: 186–188; Thagard 1999: chapter 9). The majority of analysts explore the features of the cognitive mechanisms through which aspects of a familiar domain or source are applied to an unknown target domain in order to understand what is unknown. According to the influential multi-constraint theory of analogical reasoning developed by Holyoak and Thagard, the transfer processes involved in analogical reasoning (scientific and otherwise) are guided or constrained in three main ways: 1) by the direct similarity between the elements involved; 2) by the structural parallels between source and target domain; as well as 3) by the purposes of the investigators, i.e., the reasons why the analogy is considered. Discovery, the formulation of a new hypothesis, is one such purpose.

“In vivo” investigations of scientists reasoning in their laboratories have not only shown that analogical reasoning is a key component of scientific practice, but also that the distance between source and target depends on the purpose for which analogies are sought. Scientists trying to fix experimental problems draw analogies between targets and sources from highly similar domains. In contrast, scientists attempting to formulate new models or concepts draw analogies between less similar domains. Analogies between radically different domains, however, are rare (Dunbar 1997, 2001).

In current cognitive science, human cognition is often explored in terms of model-based reasoning. The starting point of this approach is the notion that much of human reasoning, including probabilistic and causal reasoning as well as problem solving takes place through mental modeling rather than through the application of logic or methodological criteria to a set of propositions (Johnson-Laird 1983; Magnani et al. 1999; Magnani and Nersessian 2002). In model-based reasoning, the mind constructs a structural representation of a real-world or imaginary situation and manipulates this structure. In this perspective, conceptual structures are viewed as models and conceptual innovation as constructing new models through various modeling operations. Analogical reasoning—analogical modeling—is regarded as one of three main forms of model-based reasoning that appear to be relevant for conceptual innovation in science. Besides analogical modeling, visual modeling and simulative modeling or thought experiments also play key roles (Nersessian 1992, 1999, 2009). These modeling practices are constructive in that they aid the development of novel mental models. The key elements of model-based reasoning are the call on knowledge of generative principles and constraints for physical models in a source domain and the use of various forms of abstraction. Conceptual innovation results from the creation of new concepts through processes that abstract and integrate source and target domains into new models (Nersessian 2009).

Some critics have argued that despite the large amount of work on the topic, the notion of mental model is not sufficiently clear. Thagard seeks to clarify the concept by characterizing mental models in terms of neural processes (Thagard 2010). In his approach, mental models are produced through complex patterns of neural firing, whereby the neurons and the interconnections between them are dynamic and changing. A pattern of firing neurons is a representation when there is a stable causal correlation between the pattern or activation and the thing that is represented. In this research, questions about the nature of model-based reasoning are transformed into questions about the brain mechanisms that produce mental representations.

The above sections show that the study of scientific discovery has become an integral part of the wider endeavor of exploring creative thinking and creativity more generally. Naturalistic philosophical approaches combine conceptual analysis of processes of knowledge generation with empirical work on creativity, drawing heavily and explicitly on current research in psychology and cognitive science, and on in vivo laboratory observations, and, most recently, on brain imaging techniques.

  • Bartha, P., 2010, By Parallel Reasoning: The Construction and Evaluation of Analogical Arguments , New York: Oxford University Press.
  • Bechtel, W. and R. Richardson, 1993, Discovering Complexity , Princeton: Princeton University Press.
  • Benjamin, A.C., 1934, “The Mystery of Scientific Discovery ” Philosophy of Science , 1: 224–36.
  • Blackwell, R.J., 1969, Discovery in the Physical Sciences , Notre Dame: University of Notre Dame Press.
  • Boden, M.A., 2004, The Creative Mind: Myths and Mechanisms , London: Routledge.
  • Brannigan, A., 1981, The Social Basis of Scientific Discoveries , Cambridge: Cambridge University Press.
  • Brem, S. and L.J. Rips, 2000, “Explanation and Evidence in Informal Argument”, Cognitive Science , 24: 573–604.
  • Campbell, D., 1960, “Blind Variation and Selective Retention in Creative Thought as in Other Knowledge Processes”, Psychological Review , 67: 380–400.
  • Carmichael, R.D., 1922, “The Logic of Discovery”, The Monist , 32: 569–608.
  • –––, 1930, The Logic of Discovery , Chicago: Open Court.
  • Craver, C.F., 2002, “Interlevel Experiments, Multilevel Mechanisms in the Neuroscience of Memory”, Philosophy of Science Supplement , 69: 83–97.
  • Craver, C.F. and L. Darden, 2013, In Search of Mechanisms: Discoveries across the Life Sciences , Chicago: University of Chicago Press.
  • Curd, M., 1980, “The Logic of Discovery: An Analysis of Three Approaches”, in T. Nickles (ed.) Scientific Discovery, Logic, and Rationality , Dordrecht: Reidel. 201–19.
  • Darden, L., 1991, Theory Change in Science: Strategies from Mendelian Genetics , New York: Oxford University Press.
  • –––, 2002, “Strategies for Discovering Mechanisms: Schema Instantiation, Modular Subassembly, Forward/Backward Chaining”, Philosophy of Science , 69: S354-S65.
  • –––, 2009, “Discovering Mechanisms in Molecular Biology. Finding and Fixing Incompleteness and Incorrectness”, in J. Meheus and T. Nickles (eds), Models of Discovery and Creativity , Dordrecht: Springer. 43–55.
  • Ducasse, C.J., 1951, “Whewell's Philosophy of Scientific Discovery II”, The Philosophical Review , 60(2): 213–34.
  • Dunbar, K., 1997, “How scientists think: On-line creativity and conceptual change in science”, in T.B. Ward, S.M. Smith, and J. Vaid (eds), Conceptual Structures and Processes: Emergence, Discovery , and Change , Washington: American Psychological Association Press.
  • –––, 2001, “The Analogical Paradox: Why Analogy is so Easy in Naturalistic Settings Yet so Difficult in Psychological Laboratories”, in D. Gentner, K.J. Holyoak, and B.N. Kokinov (eds), The Analogical Mind: Perspectives from Cognitive Science , Cambridge: MIT Press.
  • Dunbar, K, J. Fugelsang, and C Stein, 2007, “Do Naïve Theories Ever Go Away? Using Brain and Behavior to Understand Changes in Concepts”, in M. Lovett and P. Shah (eds), Thinking with Data: 33rd Carnegie Symposium on Cognition , Mahwah: Erlbaum, 193–205.
  • Feist, G.J., 1999, “The Influence of Personality on Artistic and Scientific Creativity”, in R.J. Sternberg (ed.), Handbook of Creativity , New York: Cambridge University Press, 273–96.
  • –––, 2006, The psychology of science and the origins of the scientific mind , New Haven: Yale University Press.
  • Gutting, G., 1980, “Science as Discovery”, Revue internationale de philosophie , 131: 26–48.
  • Hanson, N.R., 1958, Patterns of Discovery , Cambridge: Cambridge University Press.
  • –––, 1960, “Is there a Logic of Scientific Discovery?”, Australasian Journal of Philosophy , 38: 91–106.
  • –––, 1965, “Notes Toward a Logic of Discovery”, in R.J. Bernstein (ed.), Perspectives on Peirce. Critical Essays on Charles Sanders Peirce , New Haven and London: Yale University Press, 42–65.
  • Harman, G.H., 1965, “The Inference to the Best Explanation”, Philosophical Review , 74.
  • Hempel, C.G., 1985, “Thoughts in the Limitations of Discovery by Computer”, in K. Schaffner (ed.), Logic of Discovery and Diagnosis in Medicine , Berkeley: University of California Press, 115–22.
  • Hesse, M., 1966, Models and Analogies in Science , Notre Dame: University of Notre Dame Press.
  • Holyoak, K.J. and P. Thagard, 1996, Mental Leaps: Analogy in Creative Thought , Cambridge: MIT Press.
  • Howard, D., 2006, “Lost Wanderers in the Forest of Knowledge: Some Thoughts on the Discovery-Justification Distinction”, in J. Schickore and F. Steinle (eds), Revisiting Discovery and Justification. Historical and Philosophical Perspectives on the Context Distinction , Dordrecht: Springer, 3–22.
  • Hoyningen-Huene, P., 1987, “Context of Discovery and Context of Justification”, Studies in History and Philosophy of Science ,18: 501–15.
  • Hull, D.L., 1988, Science as Practice: An Evolutionary Account of the Social and Conceptual Development of Science , Chicago: University of Chicago Press.
  • Johnson-Laird, P., 1983, Mental Models , Cambridge: Cambridge University Press.
  • Koehler, D.J., 1991, “Explanation, Imagination, and Confidence in Judgment”, Psychological Bulletin , 110: 499–519.
  • Kordig, C., 1978, “Discovery and Justification”, Philosophy of Science , 45: 110–17.
  • Kuhn, T.S., 1970 [1962], The Structure of Scientific Revolutions , 2 nd ed., Chicago: The University of Chicago Press. First edition published in 1962.
  • Kulkarni, D. and H.A. Simon, 1988, “The processes of scientific discovery: The strategy of experimentation”, Cognitive Science , 12: 139–76.
  • Langley, P., 2000, “The Computational Support of Scientific Discovery”, International Journal of Human-Computer Studies , 53: 393–410.
  • Langley, P., H.A. Simon, G.L. Bradshaw, and J.M. Zytkow, 1987, Scientific Discovery: Computational Explorations of the Creative Processes , Cambridge, MA: MIT Press.
  • Laudan, L., 1980, “Why Was the Logic of Discovery Abandoned?” in T. Nickles (ed.), Scientific Discovery Vol. I, Dordrecht: Reidel. 173–83.
  • Leplin, J., 1987, “The Bearing of Discovery on Justification”, Canadian Journal of Philosophy , 17: 805–14.
  • Lugg, A., 1985, “The Process of Discovery”, Philosophy of Science , 52: 207–20.
  • Machamer, P., L. Darden, and C.F. Craver, 2000, “Thinking About Mechanisms”, Philosophy of Science , 67: 1–25.
  • Magnani, L., 2000, Abduction, Reason, and Science: Processes of Discovery and Explanation , Dordrecht: Kluwer.
  • –––, 2009, “Creative Abduction and Hypothesis Withdrawal”, in J. Meheus and T. Nickles (eds), Models of Discovery and Creativity , Dordrecht: Springer.
  • Magnani, L. and N.J. Nersessian, 2002, Model-Based Reasoning: Science, Technology, and Values , Dordrecht: Kluwer.
  • Magnani, L., N.J. Nersessian, and P. Thagard, 1999, Model-Based Reasoning in Scientific Discovery , Dordrecht: Kluwer.
  • Nersessian, N.J., 1992, “How do scientists think? Capturing the dynamics of conceptual change in science”, in R. Giere (ed.), Cognitive Models of Science , Minneapolis: University of Minnesota Press, 3–45.
  • –––, 1999, “Model-based reasoning in conceptual change”, in L. Magnani, N.J. Nersessian and P. Thagard (eds), Model-Based Reasoning in Scientific Discovery , New York: Kluwer. 5–22.
  • –––, 2009, “Conceptual Change: Creativity, Cognition, and Culture ” in J. Meheus and T. Nickles (eds), Models of Discovery and Creativity , Dordrecht: Springer, 127–66.
  • Newell, A. and H. A Simon, 1971, “Human Problem Solving: The State of the Theory in 1970”, American Psychologist , 26: 145–59.
  • Nickles, T., 1984, “Positive Science and Discoverability”, PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association , 1984: 13–27.
  • –––, 1985, “Beyond Divorce: Current Status of the Discovery Debate”, Philosophy of Science , 52: 177–206.
  • –––, 1989, “Truth or Consequences? Generative versus Consequential Justification in Science”, PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association , 1988, 393–405.
  • Popper, K., 2002 [1934/1959], The Logic of Scientific Discovery , London and New York: Routledge. First published in German in 1934; first English translation in 1959.
  • Reichenbach, H., 1938, Experience and Prediction. An Analysis of the Foundations and the Structure of Knowledge , Chicago: The University of Chicago Press.
  • Richardson, A., 2006, “Freedom in a Scientific Society: Reading the Context of Reichenbach's Contexts”, in J. Schickore and F. Steinle (eds), Revisiting Discovery and Justification. Historical and Philosophical Perspectives on the Context Distinction , Dordrecht: Springer, 41–54.
  • Schaffer, S., 1986, “Scientific Discoveries and the End of Natural Philosophy”, Social Studies of Science , 16: 387–420.
  • –––, 1994, “Making Up Discovery”, in M.A. Boden (ed.), Dimensions of Creativity , Cambridge: MIT Press, 13–51.
  • Schaffner, K., 1993, Discovery and Explanation in Biology and Medicine , Chicago: University of Chicago Press.
  • Schickore, J. and F. Steinle, 2006, Revisiting Discovery and Justification. Historical and Philosophical Perspectives on the Context Distinction , Dordrecht: Springer.
  • Schiller, F.C.S., 1917, “Scientific Discovery and Logical Proof”, in C.J. Singer (ed.), Studies in the History and Method of Science Vol. 1, Oxford: Clarendon. 235–89.
  • Simon, H.A., 1973, “Does Scientific Discovery Have a Logic?”, Philosophy of Science , 40: 471–80.
  • –––, 1977, Models of Discovery and Other Topics in the Methods of Science , Dordrecht: Reidel.
  • Simon, H.A., P.W. Langley, and G.L. Bradshaw, 1981, “Scientific Discovery as Problem Solving”, Synthese , 47: 1–28.
  • Smith, G.E., 2002, “The Methodology of the Principia ”, in G.E. Smith and I.B. Cohen (eds), The Cambridge Companion to Newton , Cambridge: Cambridge University Press, 138–73.
  • Snyder, L.J., 1997, “Discoverers' Induction”, Philosophy of Science , 64: 580–604.
  • Solomon, M., 2007, “Standpoint and Creativity”, Hypathia : 226–37.
  • Thagard, P., 1984, “Conceptual Combination and Scientific Discovery”, PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association , 1984: 3–12.
  • –––, 1999, How Scientists Explain Disease , Princeton: Princeton University Press.
  • –––, 2010, “How Brains Make Mental Models”, in L. Magnani, N.J. Nersessian and P. Thagard (eds), Model-Based Reasoning in Science & Technology , Berlin and Heidelberg: Springer, 447–61.
  • –––, 2012, The Cognitive Science of Science , Cambridge: MIT Press.
  • Weber, M., 2005, Philosophy of Experimental Biology , Cambridge: Cambridge University Press.
  • Whewell, W., 1996 [1840], The Philosophy of the Inductive Sciences , Vol. II. London: Routledge/Thoemmes.
  • Zahar, E., 1983, “Logic of Discovery or Psychology of Invention?”, British Journal for the Philosophy of Science , 34: 243–61.
How to cite this entry . Preview the PDF version of this entry at the Friends of the SEP Society . Look up this entry topic at the Indiana Philosophy Ontology Project (InPhO). Enhanced bibliography for this entry at PhilPapers , with links to its database.

[Please contact the author with suggestions.]

abduction | analogy and analogical reasoning | cognitive science | knowledge: analysis of | Kuhn, Thomas | models in science | molecular biology | Newton, Isaac: Philosophiae Naturalis Principia Mathematica | Popper, Karl | Whewell, William

Copyright © 2014 by Jutta Schickore < jschicko @ indiana . edu >

Support SEP

Mirror sites.

View this site from another server:

  • Info about mirror sites

Stanford Center for the Study of Language and Information

The Stanford Encyclopedia of Philosophy is copyright © 2016 by The Metaphysics Research Lab , Center for the Study of Language and Information (CSLI), Stanford University

Library of Congress Catalog Data: ISSN 1095-5054

The REU (Research Experiences for Undergraduates) Site is in the area of Consumer Networking Technologies and will investigate some important issues related to software-defined networking/virtualization , visual computing at the network edge , social virtual reality , mobile sensing & environment recognition . The REU Site is supported by NSF , MU Office of Research , MU Office of Undergraduate Research , the College of Engineering , and the Department of Electrical Engineering & Computer Science .

As broadband services and mobile devices proliferate, consumers are migrating towards advanced social, information-intensive, and personalized services. The fast pace of growth, improvement to quality of life, and significant economic contribution of networked consumer systems have been receiving increasing attention. Technical challenges in this field include quality of service, network and device heterogeneity, network performance, mobility, device intelligence, security and privacy, and user experience and knowledge, etc. In this REU Site, students will participate and develop new skills in ongoing funded research projects of the faculty mentors by investigating, implementing, and testing viable solutions to technical challenges in consumer networking technologies. This research activity will allow the students to obtain a better understanding of the technical issues, performance, and trade-offs in consumer networking. Exposing the students to collaborative research environments, fostering their enthusiasm for science and engineering, and developing skills needed for pursuing advanced degrees in research is a goal of this program.

The REU site will support typically 10 undergraduate students every summer. The participants will engage in a 10-week summer school at the University of Missouri (MU). The participants will take a 1-week short course, one to two lectures per week in the remaining weeks, work with other undergraduate and graduate students and a faculty mentor to carry out a research project, present their progress weekly throughout the summer, and write project reports and give final presentations. In addition, the undergraduate participants will present their work at their home institutions during the following semester.

The Program runs from May through July, see Important Dates . The undergraduate participants will get a stipend of $7,000, room (if needed) and meal allowance, plus support for travel to/from the Site, and potentially travel support to present their work at a conference.

If you're seeing this message, it means we're having trouble loading external resources on our website.

If you're behind a web filter, please make sure that the domains *.kastatic.org and *.kasandbox.org are unblocked.

To log in and use all the features of Khan Academy, please enable JavaScript in your browser.

Biology archive

Course: biology archive   >   unit 1, the scientific method.

  • Controlled experiments
  • The scientific method and experimental design

hypothesis science vs discovery


  • Make an observation.
  • Ask a question.
  • Form a hypothesis , or testable explanation.
  • Make a prediction based on the hypothesis.
  • Test the prediction.
  • Iterate: use the results to make new hypotheses or predictions.

Scientific method example: Failure to toast

1. make an observation., 2. ask a question., 3. propose a hypothesis., 4. make predictions., 5. test the predictions..

  • If the toaster does toast, then the hypothesis is supported—likely correct.
  • If the toaster doesn't toast, then the hypothesis is not supported—likely wrong.

Logical possibility

Practical possibility, building a body of evidence, 6. iterate..

  • If the hypothesis was supported, we might do additional tests to confirm it, or revise it to be more specific. For instance, we might investigate why the outlet is broken.
  • If the hypothesis was not supported, we would come up with a new hypothesis. For instance, the next hypothesis might be that there's a broken wire in the toaster.

Want to join the conversation?

  • Upvote Button navigates to signup page
  • Downvote Button navigates to signup page
  • Flag Button navigates to signup page

Incredible Answer

  • Skip to primary navigation
  • Skip to main content
  • Skip to primary sidebar
  • Skip to footer

hypothesis science vs discovery

Understanding Science

How science REALLY works...

  • Understanding Science 101
  • Scientific investigations may be inspired by many different factors.
  • The early stages of an investigation often involve making observations, asking questions, sharing ideas and data, and learning what’s already been discovered about the topic.
  • Science relies on the accumulated knowledge of the scientific community to move forward.

Exploration and discovery

The early stages of a scientific investigation often rely on making observations , asking questions, and initial experimentation — essentially poking around. But the routes to and from these stages are diverse. Intriguing observations sometimes arise in surprising ways, as in the discovery of radioactivity, which was inspired by the observation that photographic plates (an early version of camera film) stored next to uranium salts were unexpectedly exposed. Sometimes interesting observations (and the investigations that follow) are suddenly made possible by the development of a new technology . For example, the launch of the Hubble Space Telescope in 1990 allowed astronomers to make deeper and more focused observations of our universe than were ever before possible. These observations ultimately led to breakthroughs in areas as diverse as star and planet formation, the nature of black holes, and the expansion of the universe.

Sometimes, observations are clarified and questions arise through discussions with colleagues and reading the work of other scientists — as demonstrated by the discovery of the role of chlorofluorocarbons (CFCs) in ozone depletion…


In 1973, chemists had observed that CFCs were being released into the environment from aerosol cans, air conditioners, and other sources. But it was discussions with his colleague and advisor, Sherwood Rowland, that led Mario Molina to ask what their ultimate fate was. Since CFCs were rapidly accumulating in the atmosphere, the question was intriguing. But before he could tackle the issue (which would ultimately lead to a Nobel Prize and an explanation for the hole in the ozone layer), Molina needed more information. He had to learn more about other scientists’ studies of atmospheric chemistry, and what he learned pointed to the disturbing fate of CFCs.

  • Take a sidetrip

Read Mario Molina’s whole story.

Furthermore, though observation and questioning are essential to the process of science, they are not enough to launch a scientific investigation on their own. Generally, scientists also need scientific background knowledge — all the information and understanding they’ve gained from their scientific training in school, supplemented by discussions with colleagues and reviews of the scientific literature . As in Mario Molina’s story, an understanding of what other scientists have already figured out about a particular topic is critical to the process. This background knowledge allows scientists to recognize revealing observations for what they are, to make connections between ideas and observations, and to figure out which questions can be fruitfully tackled with available tools. The importance of content knowledge to the process of science helps explain why science is often mischaracterized as a static set of facts contained in textbooks. Science is a process, but one that relies on accumulated knowledge to move forward.


Some scientific discoveries are chalked up to the serendipity of being in the right place at the right time to make a key observation — but rarely does serendipity alone lead to a new discovery. The people who turn lucky breaks into breakthroughs are generally those with the background knowledge and scientific ways of thinking needed to make sense of the lucky observation. For example, in 1896, Henri Becquerel made a surprising observation. He found that photographic plates stored next to uranium salts were spotted, as though they’d been exposed to light rays — even though they had been kept in a dark drawer. Someone else, with a less scientific state of mind and less background knowledge about physics, might have cursed their bad luck and thrown out the ruined plates.

But Becquerel was intrigued by the observation. He recognized it as something scientifically interesting, went on to perform follow-up experiments that traced the source of the exposure to the uranium, and in the process, discovered radioactivity. The key to this story of discovery lies partly in Becquerel’s instigating observation, but also in his way of thinking. Along with the relevant background knowledge, Becquerel had a scientific state of mind. Sure, he made some key observations — but then he dug into them further, inquiring why the plates were exposed and trying to eliminate different potential causes of the exposure to get to the physical explanation behind the happy accident.

Want to develop your own scientific state of mind? Visit  Think science  to get some handy tips.

  • Teaching resources

Serendipity is often cited as a key factor in making scientific discoveries, but is that all there is — someone being lucky enough to get hit on the head by an apple? To learn more, explore  The story of serendipity .

  • Use our  web interactive  to help students document and reflect on the process of science.
  • Learn strategies for building lessons and activities around the Science Flowchart: Grades 3-5 Grades 6-8 Grades 9-12 Grades 13-16
  • Find lesson plans for introducing the Science Flowchart to your students in: Grades 3-5 Grades 6-8 Grades 9-16
  • Get  graphics and pdfs of the Science Flowchart  to use in your classroom. Translations are available in Spanish, French, Japanese, and Swahili.

A blueprint for scientific investigations

Observation beyond our eyes

Subscribe to our newsletter

  • The science flowchart
  • Science stories
  • Grade-level teaching guides
  • Teaching resource database
  • Journaling tool
  • Misconceptions

What is a scientific hypothesis?

It's the initial building block in the scientific method.

A girl looks at plants in a test tube for a science experiment. What's her scientific hypothesis?

Hypothesis basics

What makes a hypothesis testable.

  • Types of hypotheses
  • Hypothesis versus theory

Additional resources


A scientific hypothesis is a tentative, testable explanation for a phenomenon in the natural world. It's the initial building block in the scientific method . Many describe it as an "educated guess" based on prior knowledge and observation. While this is true, a hypothesis is more informed than a guess. While an "educated guess" suggests a random prediction based on a person's expertise, developing a hypothesis requires active observation and background research. 

The basic idea of a hypothesis is that there is no predetermined outcome. For a solution to be termed a scientific hypothesis, it has to be an idea that can be supported or refuted through carefully crafted experimentation or observation. This concept, called falsifiability and testability, was advanced in the mid-20th century by Austrian-British philosopher Karl Popper in his famous book "The Logic of Scientific Discovery" (Routledge, 1959).

A key function of a hypothesis is to derive predictions about the results of future experiments and then perform those experiments to see whether they support the predictions.

A hypothesis is usually written in the form of an if-then statement, which gives a possibility (if) and explains what may happen because of the possibility (then). The statement could also include "may," according to California State University, Bakersfield .

Here are some examples of hypothesis statements:

  • If garlic repels fleas, then a dog that is given garlic every day will not get fleas.
  • If sugar causes cavities, then people who eat a lot of candy may be more prone to cavities.
  • If ultraviolet light can damage the eyes, then maybe this light can cause blindness.

A useful hypothesis should be testable and falsifiable. That means that it should be possible to prove it wrong. A theory that can't be proved wrong is nonscientific, according to Karl Popper's 1963 book " Conjectures and Refutations ."

An example of an untestable statement is, "Dogs are better than cats." That's because the definition of "better" is vague and subjective. However, an untestable statement can be reworded to make it testable. For example, the previous statement could be changed to this: "Owning a dog is associated with higher levels of physical fitness than owning a cat." With this statement, the researcher can take measures of physical fitness from dog and cat owners and compare the two.

Types of scientific hypotheses

Elementary-age students study alternative energy using homemade windmills during public school science class.

In an experiment, researchers generally state their hypotheses in two ways. The null hypothesis predicts that there will be no relationship between the variables tested, or no difference between the experimental groups. The alternative hypothesis predicts the opposite: that there will be a difference between the experimental groups. This is usually the hypothesis scientists are most interested in, according to the University of Miami .

For example, a null hypothesis might state, "There will be no difference in the rate of muscle growth between people who take a protein supplement and people who don't." The alternative hypothesis would state, "There will be a difference in the rate of muscle growth between people who take a protein supplement and people who don't."

If the results of the experiment show a relationship between the variables, then the null hypothesis has been rejected in favor of the alternative hypothesis, according to the book " Research Methods in Psychology " (​​BCcampus, 2015). 

There are other ways to describe an alternative hypothesis. The alternative hypothesis above does not specify a direction of the effect, only that there will be a difference between the two groups. That type of prediction is called a two-tailed hypothesis. If a hypothesis specifies a certain direction — for example, that people who take a protein supplement will gain more muscle than people who don't — it is called a one-tailed hypothesis, according to William M. K. Trochim , a professor of Policy Analysis and Management at Cornell University.

Sometimes, errors take place during an experiment. These errors can happen in one of two ways. A type I error is when the null hypothesis is rejected when it is true. This is also known as a false positive. A type II error occurs when the null hypothesis is not rejected when it is false. This is also known as a false negative, according to the University of California, Berkeley . 

A hypothesis can be rejected or modified, but it can never be proved correct 100% of the time. For example, a scientist can form a hypothesis stating that if a certain type of tomato has a gene for red pigment, that type of tomato will be red. During research, the scientist then finds that each tomato of this type is red. Though the findings confirm the hypothesis, there may be a tomato of that type somewhere in the world that isn't red. Thus, the hypothesis is true, but it may not be true 100% of the time.

Scientific theory vs. scientific hypothesis

The best hypotheses are simple. They deal with a relatively narrow set of phenomena. But theories are broader; they generally combine multiple hypotheses into a general explanation for a wide range of phenomena, according to the University of California, Berkeley . For example, a hypothesis might state, "If animals adapt to suit their environments, then birds that live on islands with lots of seeds to eat will have differently shaped beaks than birds that live on islands with lots of insects to eat." After testing many hypotheses like these, Charles Darwin formulated an overarching theory: the theory of evolution by natural selection.

"Theories are the ways that we make sense of what we observe in the natural world," Tanner said. "Theories are structures of ideas that explain and interpret facts." 

  • Read more about writing a hypothesis, from the American Medical Writers Association.
  • Find out why a hypothesis isn't always necessary in science, from The American Biology Teacher.
  • Learn about null and alternative hypotheses, from Prof. Essa on YouTube .

Encyclopedia Britannica. Scientific Hypothesis. Jan. 13, 2022. https://www.britannica.com/science/scientific-hypothesis

Karl Popper, "The Logic of Scientific Discovery," Routledge, 1959.

California State University, Bakersfield, "Formatting a testable hypothesis." https://www.csub.edu/~ddodenhoff/Bio100/Bio100sp04/formattingahypothesis.htm  

Karl Popper, "Conjectures and Refutations," Routledge, 1963.

Price, P., Jhangiani, R., & Chiang, I., "Research Methods of Psychology — 2nd Canadian Edition," BCcampus, 2015.‌

University of Miami, "The Scientific Method" http://www.bio.miami.edu/dana/161/evolution/161app1_scimethod.pdf  

William M.K. Trochim, "Research Methods Knowledge Base," https://conjointly.com/kb/hypotheses-explained/  

University of California, Berkeley, "Multiple Hypothesis Testing and False Discovery Rate" https://www.stat.berkeley.edu/~hhuang/STAT141/Lecture-FDR.pdf  

University of California, Berkeley, "Science at multiple levels" https://undsci.berkeley.edu/article/0_0_0/howscienceworks_19

Sign up for the Live Science daily newsletter now

Get the world’s most fascinating discoveries delivered straight to your inbox.

Last Chance Lake: The unusual 'soda lake' with conditions that may have given rise to life on Earth

See stunning photos of the Atacama Desert — the driest on Earth — blooming in winter for 1st time in a decade

Red handfish: A tiny, moody fish with hands for fins and an extravagant mohawk

Most Popular

  • 2 No, NASA hasn't warned of an impending asteroid strike in 2038. Here's what really happened.
  • 3 James Webb Space Telescope spies strange shapes above Jupiter's Great Red Spot
  • 4 What defines a species? Inside the fierce debate that's rocking biology to its core
  • 5 Newly discovered asteroid larger than the Great Pyramid of Giza will zoom between Earth and the moon on Saturday
  • 2 Newly discovered asteroid larger than the Great Pyramid of Giza will zoom between Earth and the moon on Saturday
  • 3 2,000 years ago, a bridge in Switzerland collapsed on top of Celtic sacrifice victims, new study suggests
  • 4 Self-healing 'living skin' can make robots more humanlike — and it looks just as creepy as you'd expect
  • 5 Tasselled wobbegong: The master of disguise that can eat a shark almost as big as itself

hypothesis science vs discovery

Encyclopedia Britannica

  • History & Society
  • Science & Tech
  • Biographies
  • Animals & Nature
  • Geography & Travel
  • Arts & Culture
  • Games & Quizzes
  • On This Day
  • One Good Fact
  • New Articles
  • Lifestyles & Social Issues
  • Philosophy & Religion
  • Politics, Law & Government
  • World History
  • Health & Medicine
  • Browse Biographies
  • Birds, Reptiles & Other Vertebrates
  • Bugs, Mollusks & Other Invertebrates
  • Environment
  • Fossils & Geologic Time
  • Entertainment & Pop Culture
  • Sports & Recreation
  • Visual Arts
  • Demystified
  • Image Galleries
  • Infographics
  • Top Questions
  • Britannica Kids
  • Saving Earth
  • Space Next 50
  • Student Center
  • Introduction
  • Philosophy and natural science
  • Logical positivism and logical empiricism

Logics of discovery and justification

  • Bayesian confirmation
  • Eliminativism and falsification
  • Underdetermination
  • The work of Carl Hempel
  • Difficulties
  • Other approaches to explanation
  • Scientific laws
  • The axiomatic conception
  • The semantic conception
  • The historicist conception
  • Unification and reduction
  • The work of Thomas Kuhn
  • Early arguments for realism
  • The antirealism of van Fraassen, Laudan, and Fine
  • “Piecemeal” realism
  • Scientific truth
  • Science as a social activity
  • Feminist themes
  • Progress and values


  • Where did Auguste Comte go to school?
  • What was Auguste Comte best known for?
  • What did Werner Heisenberg do during World War II?
  • What is Werner Heisenberg best known for?
  • How did Werner Heisenberg contribute to atomic theory?

Aristotle (384-322 BC), Ancient Greek philosopher and scientist. One of the most influential philosophers in the history of Western thought, Aristotle established the foundations for the modern scientific method of enquiry. Statue

Discovery, justification, and falsification

Our editors will review what you’ve submitted and determine whether to revise the article.

  • Social Sci LibreTexts - Philosophy of Science
  • Routledge Encyclopedia of Philosophy - Philosophy of Science
  • The Basics of Philosophy - Philosophy of Science
  • The Stanford Encyclopedia of Philosophy - Kant’s philosophy of science
  • Open Library Publishing Platform - The Philosophy of Science
  • Table Of Contents

An ideal theory of scientific method would consist of instructions that could lead an investigator from ignorance to knowledge. Descartes and Bacon sometimes wrote as if they could offer so ideal a theory, but after the mid-20th century the orthodox view was that this is too much to ask for. Following Hans Reichenbach (1891–1953), philosophers often distinguished between the “context of discovery” and the “context of justification.” Once a hypothesis has been proposed, there are canons of logic that determine whether or not it should be accepted—that is, there are rules of method that hold in the context of justification. There are, however, no such rules that will guide someone to formulate the right hypothesis , or even hypotheses that are plausible or fruitful. The logical empiricists were led to this conclusion by reflecting on cases in which scientific discoveries were made either by imaginative leaps or by lucky accidents; a favourite example was the hypothesis by August Kekulé (1829–96) that benzene molecules have a hexagonal structure, allegedly formed as he was dozing in front of a fire in which the live coals seemed to resemble a snake devouring its own tail.

Although the idea that there cannot be a logic of scientific discovery often assumed the status of orthodoxy, it was not unquestioned. As will become clear below ( see Scientific change ), one of the implications of the influential work of Thomas Kuhn (1922–96) in the philosophy of science was that considerations of the likelihood of future discoveries of particular kinds are sometimes entangled with judgments of evidence , so discovery can be dismissed as an irrational process only if one is prepared to concede that the irrationality also infects the context of justification itself.

Sometimes in response to Kuhn and sometimes for independent reasons, philosophers tried to analyze particular instances of complex scientific discoveries, showing how the scientists involved appear to have followed identifiable methods and strategies. The most ambitious response to the empiricist orthodoxy tried to do exactly what was abandoned as hopeless—to wit , specify formal procedures for producing hypotheses in response to an available body of evidence. So, for example, the American philosopher Clark Glymour and his associates wrote computer programs to generate hypotheses in response to statistical evidence, hypotheses that often introduced new variables that did not themselves figure in the data. These programs were applied in various traditionally difficult areas of natural and social scientific research. Perhaps, then, logical empiricism was premature in writing off the context of discovery as beyond the range of philosophical analysis.

In contrast, logical empiricists worked vigorously on the problem of understanding scientific justification. Inspired by the thought that Frege, Russell, and Hilbert had given a completely precise specification of the conditions under which premises deductively imply a conclusion, philosophers of science hoped to offer a “logic of confirmation” that would identify, with equal precision, the conditions under which a body of evidence supported a scientific hypothesis . They recognized, of course, that a series of experimental reports on the expansion of metals under heat would not deductively imply the general conclusion that all metals expand when heated—for even if all the reports were correct, it would still be possible that the very next metal to be examined failed to expand under heat. Nonetheless, it seemed that a sufficiently large and sufficiently varied collection of reports would provide some support, even strong support, for the generalization. The philosophical task was to make precise this intuitive judgment about support.

hypothesis science vs discovery

During the 1940s, two prominent logical empiricists, Rudolf Carnap (1891–1970) and Carl Hempel (1905–97), made influential attempts to solve this problem. Carnap offered a valuable distinction between various versions of the question. The “qualitative” problem of confirmation seeks to specify the conditions under which a body of evidence E supports, to some degree, a hypothesis H. The “comparative” problem seeks to determine when one body of evidence E supports a hypothesis H more than a body of evidence E* supports a hypothesis H* (here E and E* might be the same, or H and H* might be the same). Finally, the “quantitative” problem seeks a function that assigns a numerical measure of the degree to which E supports H. The comparative problem attracted little attention, but Hempel attacked the qualitative problem while Carnap concentrated on the quantitative problem.

It would be natural to assume that the qualitative problem is the easier of the two, and even that it is quite straightforward. Many scientists (and philosophers) were attracted to the idea of hypothetico-deductivism, or the hypothetico-deductive method : scientific hypotheses are confirmed by deducing from them predictions about empirically determinable phenomena, and, when the predictions hold good, support accrues to the hypotheses from which those predictions derive. Hempel’s explorations revealed why so simple a view could not be maintained. An apparently innocuous point about support seems to be that, if E confirms H, then E confirms any statement that can be deduced from H. Suppose, then, that H deductively implies E, and E has been ascertained by observation or experiment. If H is now conjoined with any arbitrary statement, the resulting conjunction will also deductively imply E. Hypothetico-deductivism says that this conjunction is confirmed by the evidence. By the innocuous point, E confirms any deductive consequence of the conjunction. One such deductive consequence is the arbitrary statement. So one reaches the conclusion that E, which might be anything whatsoever, confirms any arbitrary statement.

To see how bad this is, consider one of the great predictive theories—for example, Newton’s account of the motions of the heavenly bodies. Hypothetico-deductivism looks promising in cases like this, precisely because Newton’s theory seems to yield many predictions that can be checked and found to be correct. But if one tacks on to Newtonian theory any doctrine one pleases—perhaps the claim that global warming is the result of the activities of elves at the North Pole—then the expanded theory will equally yield the old predictions. On the account of confirmation just offered, the predictions confirm the expanded theory and any statement that follows deductively from it, including the elfin warming theory.

Hempel’s work showed that this was only the start of the complexities of the problem of qualitative confirmation, and, although he and later philosophers made headway in addressing the difficulties, it seemed to many confirmation theorists that the quantitative problem was more tractable. Carnap’s own attempts to tackle that problem, carried out in the 1940s and ’50s, aimed to emulate the achievements of deductive logic. Carnap considered artificial systems whose expressive power falls dramatically short of the languages actually used in the practice of the sciences, and he hoped to define for any pair of statements in his restricted languages a function that would measure the degree to which the second supports the first. His painstaking research made it apparent that there were infinitely many functions (indeed, continuum many—a “larger” infinity corresponding to the size of the set of real number s) satisfying the criteria he considered admissible. Despite the failure of the official project, however, he argued in detail for a connection between confirmation and probability, showing that, given certain apparently reasonable assumptions, the degree-of-confirmation function must satisfy the axioms of the probability calculus .

U.S. flag

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

  • Publications
  • Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

  • Advanced Search
  • Journal List
  • Mol Endocrinol

Editorial: Would You Like A Hypothesis With Those Data? Omics and the Age of Discovery Science

The advent of microarray-based genomic technologies in the late 1980s and early 1990s, leading to the groundbreaking paper by Brown and coworkers in 1995 describing the first microarray expression analysis ( 1 ), ushered in the age of genomics. Since then, a number of approaches aimed at the “collective characterization and quantification of pools of biological molecules that translate into the structure, function, and dynamics of an organism” ( 2 ), together referred to as “omics,” have been developed and applied to a wide variety of biological systems. Omics approaches, including genomics, transcriptomics, proteomics, and metabolomics, have transformed their respective disciplines. No longer is the study of a single gene, RNA, protein, or metabolite the norm. These approaches have also transformed modern biology as well, all the while stirring up a vigorous debate between those who favor hypothesis-driven science and those who favor data-driven science ( 3 ). This dichotomy, however, is too simplistic. In this editorial, I discuss how arguments about hypothesis- vs data-driven science have impacted the integration of omics approaches into our biological tool kit, as well as the grant review panels that evaluate proposed omics research. In addition, I discuss the role of journals, such as Molecular Endocrinology , in providing a forum for the presentation of high-quality omics resources.

“The goal is to discover things we neither knew nor expected …” Patrick O. Brown

Discovery, Mechanism, and Description: Two Out of Three Ain't Bad

Although many scientists, like me, get excited by a well-executed and informative omics experiment, especially one using a newly developed method that can delve into previously inaccessible biological endpoints, others are put off by such “discovery science.” Pat Brown, an advocate of discovery science, asserted that “the goal is to discover things we neither knew nor expected, and to see relationships and connections among the elements, whether previously suspected or not. It follows that this process is not driven by hypothesis and should be as model-independent as possible” ( 4 ). In contrast, John Allen, a critic of the approach, predicted that “induction and data-mining, uninformed by ideas, can themselves produce neither knowledge nor understanding” ( 5 ). Although entertaining, thought provoking, and, at times, useful, this debate does not reflect the current realities and applications of omics 20+ years into the enterprise.

In many respects, labeling omics as purely discovery science is a misrepresentation. Although discovery is an important use for omics, testing molecular mechanisms on a global scale is equally important. Scientists have gotten better at designing omic experiments and combining them with other approaches (eg, perturbations, such as RNAi-mediated knockdown or chemical inhibition of effector proteins), demonstrating the power of genomics to test the generality of molecular mechanisms on a global scale. If there is a problem with the use and application of omics approaches, it lies not with the tools (ie, omics approaches), but those who wield them (ie, scientists), at times not so deftly.

In my view, there are 3 types of omics experiments, which I refer to as discovery-focused, mechanism-focused, and descriptive (see Box 1). Discovery-focused omics experiments are designed to “discover things we neither knew nor expected,” as Pat Brown imagined. They apply new or existing methodologies to biological systems under conditions or at time points that are most likely to reveal key aspects of the biology (hmmm … that almost sounds like a hypothesis, but I will address that in more detail below). Such experiments may be combined with high-throughput screens (eg, RNAi or chemical libraries), which allows discovery of targets on a global scale. Mechanism-focused omics experiments (also referred to as Functional omics experiments) are typically based on clear hypotheses and are designed to test, or perhaps reveal, underlying molecular mechanisms on a global scale. These are classical perturbation-effect experiments writ large, revealing the generality of a molecular mechanism, or variations in a mechanism, across the “ome” in question. Finally, descriptive omics experiments are those that are neither discovery- nor mechanism-focused, and perhaps might best be described as an omics experiment gone wrong. They survey the biological system, without leading to significant discoveries. They may also appear to address mechanisms without actually doing so. Descriptive omics experiments produce catalogs (lists of genes, transcripts, proteins, or metabolites) whose levels “change” from condition A to condition B, without revealing how or why. I think discovery- and mechanism-focused omics experiments have great value in modern biology. Descriptive omics experiments, however, are of limited value, because they are typically poorly designed and, as such, fail to capitalize on the power of omics as a tool of discovery and hypothesis testing.

Grant Me This: Study Sections and Their Love/Hate Relationship With Omics Experiments

The debate about hypothesis-driven vs data-driven science has made its way into the grant review process. For grant applicants, this can be a difficult minefield to navigate. This is disappointing, because a well-crafted and well-integrated omics experiments can contribute significantly to a research proposal. In my experience, both as a member of an NIH study section and an applicant, grant review panels whose traditional purview is mechanisms or physiology/pathophysiology are prone to a dim view of specific aims that are omics-centric. Common criticisms include: 1) lack of a hypothesis, 2) proposed experiments that are too ambitious, and 3) lack of a clear plan for sorting through and testing targets and potential mechanisms. Ironically, proposals that focus on a single gene and lack a global view are often criticized (rightly so, in my opinion) for not including the obvious omics experiment that would broaden the perspective and reveal new molecular details. Thus, it is understandable if applicants sense that grant review panels have a love/hate relationship with omics experiments.

Although each of the criticisms noted above may be valid in specific cases, I think they are too often applied in a knee-jerk manner. For example, even discovery-focused omics experiments have an implicit hypothesis: “the signature of the transcriptome (or proteome, or metabolome) is indicative of the state of the biological system and will reveal the key effectors (RNAs, proteins, metabolites) driving the biological state.” Furthermore, the proliferation of omics technologies has led to the establishment of core facilities and companies that can process samples, generate data, and assist with data analysis, making omics experiments practical, cost effective, and accessible to most researchers (although access to the computational resources needed for detailed analyses are limiting for many scientists, but that is a topic for another editorial). Nonetheless, in a practical and strategic surrender, I have taken to coaching new faculty and postdocs applying for grants or fellowships to downplay their omics experiments in their aims, presenting them as preliminary data (if initial experiments can be completed before submission) or making them more of an afterthought or “confirmation on a global scale.” Of course, none of this is ideal. Rather than furthering the scientific enterprise, it stands in the way. What can be done?

An external file that holds a picture, illustration, etc.
Object name is zmg9991556120001.jpg

I think there is shared culpability for the issues that I outlined above, and both the grant reviewers and the applicants need to be part of the solution. On the one hand, grant reviewers must get past the aversion to discovery science. Although the philosophical debate may persist, the practical debate is over. Experience from the past 20+ years tells us that omics approaches are an essential part of modern biology. They have allowed us to make discoveries and understand biology in a way that we could not have done without them. In addition, grant reviewers should be willing to accept that even a discovery-focused omics experiment can have a hypothesis. Finally, grant reviewers must make every effort to distinguish between the 3 types of omics experiments described above, weeding out those that are purely descriptive.

On the other hand, applicants must provide a clear justification for the need, as well as an explanation of the benefit, of their proposed omics experiments. Importantly, they must present clear hypotheses for the discovery- and mechanism-focused omics experiments that they propose, as noted above, and avoid descriptive omics experiments. Furthermore, applicants must include 1) a clear plan for analyzing and sorting through the data, 2) a proposal for establishing the priority of targets for follow-up analyses, 3) a description of the expected outcomes, and 4) how the outcomes will provide a test of the hypothesis. The applicant is responsible for leading the reviewer through the experiments. I think there is a tendency for applicants to include a particular omics experiment in their proposal, with the assumption that the reviewer will somehow be able to discern the path from A to B to C. For example, “performing RNA-seq” becomes shorthand for everything from the sample collection, data generation, data analysis, and the outcomes and interpretation, without actually elucidating any of the details for the reviewer. Finally, the applicant must present a credible plan for data analysis, including a description of the methods, resources, computational infrastructure, and computational personnel available for the analysis.

I think these common-sense approaches for applicants and reviewers will improve the use and integration of omics experiments into research plans, as well as improve the process of reviewing grant applications that propose to use omics approaches.

Data Resources for the Community: A Role for Journals?

The large datasets generated in omics experiments can be mined again and again, making them great resources for future experiments that were not yet conceived at the time of the initial data generation. In my own lab, we regularly integrate data from our own genomic experiments with others that are publically available. For example, in one recent study ( 6 ), we mined more than 25 different publically available genomic datasets and integrated them with a few of our own. The publically available data gave us a much broader and deeper view of the system than we could have discerned from our data alone.

The availability of data generated by other laboratories or consortia, especially the raw data, which can be more flexibly mined, is an important component of the overall omics enterprise. Although many omics datasets are associated with and released as part of publications in science journals, neither the journals, nor the authors, are equipped for, or well suited to, maintaining repositories of the data for others who wish to mine them. Rather, a number of omics data repositories, many of which are publically supported (eg, by the NIH), have been established to maintain the data as a resource to the scientific community. These include the National Center for Biotechnology Information (NCBI)'s Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/ ) for genomic data and the Metabolomics Consortium Data Repository and Coordinating Center (DRCC; http://www.metabolomicsworkbench.org/data/DRCCDataDeposit.php ) for metabolomic data. More should be done to support, maintain, and expand these repositories. Wisely, most journals have made submission of new omics datasets used in the papers they publish a requirement for acceptance and publication.

Box 1. Three Types of Omics Experiments

1) discovery-focused.

Designed to discover things neither known nor expected. They apply new or existing methodologies to biological systems under conditions or at time points that are most likely to reveal key aspects of the biology.

2) Mechanism-focused (functional)

They are based on clear hypotheses and are designed to test underlying molecular mechanisms on a global scale using perturbation-effect approaches. They reveal the generality of a molecular mechanism, or variations in a mechanism, across the ome in question.

3) Descriptive

They survey the biological system, without leading to significant discovery, and they may appear to address a mechanism without actually doing so. They produce catalogs (lists of genes, transcripts, proteins, or metabolites) whose levels change from condition A to condition B, without revealing how or why.

In addition to broad repositories, such as those noted above, some fields of research have developed their own topical resources that contain (or link to) omics data, as well as other information, relevant to the field. One such resource relevant to many readers of Molecular Endocrinology is the Nuclear Receptor Signaling Atlas (NURSA; https://www.nursa.org/nursa/index.jsf ), which organizes and contains links to published datasets in the nuclear receptor field, as well as a tool called “Transcriptomine,” which allows users to mine tissue-specific nuclear receptor signaling pathways based on public transcriptomic datasets. Field-specific resources are even being organized, supported, and maintained by some institutes at the NIH. This includes the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK)'s Information Network (dkNET; http://www.dknet.org ), which provides access to large pools of data relevant to the mission of the NIDDK. These resources are a useful launching point for exploring available omics data in a particular field.

Box 2. Criteria for Resource Papers in Molecular Endocrinology

  • Describes the development of a new method or an improved version of an existing method that is of general use to others who might conduct similar studies in the future.
  • Describes a dataset for a relevant biological system that is superior to any previously published datasets (must meet standards of quality, novelty, and uniqueness).
  • Presents the data in interesting ways that illustrate the full spectrum of the data and the biological interest.
  • Includes new or improved methods that lead to new observations or the generation of new hypotheses that were not reported previously.
  • Includes functional follow-up that tests the underlying hypotheses generated from the data.

Journals have also begun to venture in the realm of omics data resources by publishing “resource” papers that highlight new, interesting, and broadly useful datasets, especially those generated using newly developed methodologies. Omics datasets, however, are not useful to the community if they are limited, highly specific, or of poor quality. Obviously, not every RNA-seq or ChIP-seq dataset is a useful resource. Thus, a rigorous set of criteria must be applied when evaluating resource papers. The editors at Molecular Endocrinology have developed a set of criteria that we use and ask our reviewers to use when evaluating resource papers for publication (see Box 2). Understandably, papers with descriptive omics experiments are not reviewed favorably. We hope that these criteria will help authors as they consider submitting their omics papers for publication as resources.

I think my enthusiasm for the use of omics in modern biology should be evident from this editorial. I am an advocate for the use and integration of omics experiments wherever applicable. However, I also think that omics is a tool that must be wieldy deftly for maximum benefit. Continuing dialogs like this one should help us to refine the way omics tools are applied and allow us to perfect this craft.


The author thanks Gary Hon for helpful comments and feedback on this piece.

The Omics research in the author's lab is funded by grants from the National Institutes of Health (NIH)/National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), the Cancer Prevention Research Institute of Texas (CPRIT), and the Cecil H. and Ida Green Center for Reproductive Biology Sciences Endowment. W.L.K. holds the Cecil H. and Ida Green Distinguished Chair in Reproductive Biology Sciences.

Disclosure Summary: The author has nothing to disclose.

There’s a powerful story behind every headline at Ohio State Health & Discovery. As one of the largest academic health centers and health sciences campuses in the nation, we are uniquely positioned with renowned experts covering all aspects of health, wellness, science, research and education. Ohio State Health & Discovery brings this expertise together to deliver today’s most important health news and the deeper story behind the most powerful topics that affect the health of people, animals, society and the world.  Like the science and discovery news you find here? You can support more innovations fueling advances across medicine, science, health and wellness by giving today.


  • The Ohio State University
  • College of Dentistry
  • College of Medicine
  • College of Nursing
  • College of Optometry
  • College of Pharmacy
  • College of Public Health
  • College of Veterinary Medicine
  • Ohio State Wexner Medical Center
  • Ohio State's Comprehensive Cancer Center – James Cancer Hospital and Solove Research Institute

Subscribe. The latest from Ohio State Health & Discovery delivered right to your inbox.

Explaining turmeric’s health benefits — and limitations

Registered Dietitian Ohio State Wexner Medical Center

  • Share on Facebook
  • Share on Linkedin
  • Share via Email
  • Share this page

Hands holding turmeric golden milk latte with spices

Turmeric , a golden-hued spice used frequently in Indian, Middle Eastern and Southeast Asian dishes, has a long list of health benefits to its name.

Internet searches will turn up claims that turmeric can do everything from reducing inflammation and migraines to preventing and treating cancer and Alzheimer’s disease .

But which claims are based on legitimate evidence? And what are the safest, most efficient ways to get health benefits from the spice?

The good news

Turmeric has been used for medicinal purposes in Ayurveda  and Chinese medicine for thousands of years to treat a variety of inflammatory conditions. There is substantial evidence from preclinical studies (in test tubes and animals) showing that the active component of turmeric, curcumin , is a powerful anti-inflammatory and antioxidant agent.

More importantly, the number of research studies conducted on the benefits of curcumin in human subjects have been increasing, so we now have a larger body of evidence on the benefits of curcumin. Studies suggest that regular use of curcumin may lead to:

  • Improvements in blood sugar regulation, including improved A1C and improved insulin sensitivity
  • Improvements in lipids  (cholesterol, triglycerides)
  • Improvements in BMI
  • Improvements in osteoarthritis  pain and function that are comparable to NSAID medications
  • Improvements in active symptoms and markers of inflammation  for those suffering from ulcerative colitis
  • Reducing skin irritation in breast cancer patients who have received radiation treatments.
  • Reducing the number of heart attacks in those who have had bypass surgery.

Unless you have an allergy or severe food intolerance , it’s safe to use turmeric in a paste applied to your skin or to consume a few teaspoons of turmeric daily in your food. It’s also a fun way to give foods more flavor and color.

The bad news

The participants in these human studies usually consumed a supplement of curcumin alone, instead of whole turmeric itself. These supplements often hold a concentration of curcumin you’re unlikely to consume in everyday life, even if you add turmeric to every food you eat.

Curcumin from turmeric powder can never be absorbed as well as it is from a curcumin supplement. Turmeric powder has about 2% curcumin, while supplements are about 95% curcumin. So, it’s hard to say if you’ll get the same benefit shown in studies if you’re only consuming turmeric in its whole-food form.

However, the robust historical use of turmeric as an anti-inflammatory addition to the diet leads me to think it is still beneficial to include some whole-food turmeric for the above benefits, even if you also choose to also take a curcumin supplement.

How can I get the benefits of turmeric?

Your body best absorbs the curcumin in turmeric when paired with fat and black pepper components. This is likely why recipes for turmeric “golden milk” drinks are so popular among wellness-seekers — the recipes typically include milk (containing fat) and pinches of black pepper.

In cooking, turmeric pairs well with root vegetables and dark, leafy greens, such as kale, spinach and collard greens. When added to eggs, rice or tofu, it kicks up otherwise bland flavors. It can also be added to any grain you are cooking, such as rice, quinoa, barley or farro. Another suggestion I give my patients is to add it to salad dressings and sauces. When all else fails, you can try making a golden milk drink for yourself:

  • 1 cup whole milk or dairy-free alternative
  • 1 tsp turmeric powder (can start with 1/2 tsp to get used to the taste)
  • Sprinkle of black pepper
  • Other spices to taste (like cinnamon, cardamom)
  • Sweetener to taste (maple syrup, honey)
  • ½ tsp to 1 tsp ghee/coconut oil for additional fat if desired

Combine milk and spices and bring to a boil, then reduce heat and simmer for a few minutes. Pour into a mug and let cool to a warm drinkable temperature, then add desired sweetener and additional fat if desired.

Should I take a curcumin supplement?

While most people are safely able to ingest up to 4 grams of concentrated curcumin supplements per day, curcumin can interact with some medications and make them toxic.

Among the drugs made more potent with curcumin are anticoagulant or antiplatelet medications, such as aspirin, clopidogrel (Plavix), enoxaparin (Lovenox) and heparin.

There’s also some evidence that in those with hormone-sensitive cancers, such as breast, ovarian or prostate cancer, a curcumin supplement could worsen the cancer or interfere with cancer medications. And in women who are pregnant or breastfeeding, the supplement isn’t recommended  because researchers just haven’t fully studied the risks.

For these reasons, it’s important to talk with your doctor before beginning a curcumin supplement. But there’s no risk to consuming a few teaspoons of turmeric per day, so feel free to spice away with the whole-food form!

The bottom line

It seems that what many people hope to find in turmeric are overall health benefits and reduced inflammation. There are no health risks in using turmeric as a spice added to food and, if your doctor approves, a curcumin supplement could be beneficial to your health.

To get the most benefits from added turmeric, incorporate it into a diet rich with anti-inflammatory foods, plenty of fruits and vegetables, limited sugar and lots of plant-based protein. A long-term lifestyle change involving regular exercise and a well-rounded diet will help more than a supplement alone.

There’s still much left for us to understand about how different foods and spices digest and interact in our bodies, so I always advocate for incorporating whole foods and spices into what we eat, instead of trying to reduce it down to a single nutrient that can be taken in pill form.

While nutrition research is important for understanding how to optimize eating habits, it’s equally important not to get swept away by headlines promising that all of our health ailments will be cured by adding one certain food, spice or herb to our kitchen rotation.

Healthy eating is within your reach!

Make an appointment with our dietitians or nutritionists.

Mary Mosquera Cochran

  • Anti-Inflammatory Diet ,
  • Nutrition ,
  • Ohio State Wexner Medical Center ,
  • Exercise and Nutrition

Related websites

Articles on wellness.

The dangers of orthorexia

The dangers of orthorexia

By Stacy Cleveland, RDN, LD

Healthy eating can become an obsession. An Ohio State dietitian explains how.

Does cooling starchy food after cooking really lower its calories?

Does cooling starchy food after cooking really lower its calories?

By Kristine Dilley, RDN, LD

Here’s a trick to eating pasta that cuts the calories, explained by an Ohio State dietitian.

Is protein powder good for you?

By Liz Weinandy, MPH, RDN, LD

Magnesium: Are you getting enough?

By Candace Pumper, MS, RD, LD

When do you really need to add hydration tablets or other electrolytes to your exercise?

By Sarah Wick, RD, LD, CSSD

Get articles and stories about health, wellness, medicine, science and education delivered right to your inbox from the experts at Ohio State.

Required fields

By clicking "Subscribe" you agree to our Terms of Use . Learn more about how we use your information by reading our Privacy Policy .

illustration of a neanderthal skull, DNA helix and humanoid skull

‘A history of contact’: Princeton geneticists are rewriting the narrative of Neanderthals and other ancient humans

Illustration by Matilda Luk, Office of Communications

Ever since the first Neanderthal bones were discovered, people have wondered about these ancient hominins. How are they different from us? How much are they like us? Did our ancestors get along with them? Fight them? Love them? The recent discovery of a group called Denisovans, a Neanderthal-like group who populated Asia and Oceania, added its own set of questions.

Now, an international team of geneticists and AI experts are adding whole new chapters to our shared hominin history. Under the leadership of Joshua Akey, a professor in Princeton’s Lewis-Sigler Institute for Integrative Genomics, the researchers have found a history of genetic intermingling and exchange that suggests a much more intimate connection between these early human groups than previously believed.

“This is the first time that geneticists have identified multiple waves of modern human-Neanderthal admixture,” said Liming Li, a professor in the Department of Medical Genetics and Developmental Biology at Southeast University in Nanjing, China, who performed this work as an associate research scholar in Akey’s lab.

“We now know that for the vast majority of human history, we've had a history of contact between modern humans and Neanderthals,” said Akey. The hominins who are our most direct ancestors split from the Neanderthal family tree about 600,000 years ago, then evolved our modern physical characteristics about 250,000 years ago.

“From then until the Neanderthals disappeared — that is, for about 200,000 years — modern humans have been interacting with Neanderthal populations,” he said.

The results of their work appear in the current issue of the journal Science.

Neanderthals, once stereotyped as slow-moving and dim-witted, are now seen as skilled hunters and tool makers who treated each other’s injuries with sophisticated techniques and were well adapted to thrive in the cold European weather.

(Note: All of these hominin groups are humans, but to avoid saying “Neanderthal humans,” “Denisovan humans,” and “ancient-versions-of-our-own-kind-of-humans,” most archaeologists and anthropologists use the shorthand Neanderthals, Denisovans, and modern humans.)

Mapping the gene flow

Using genomes from 2,000 living humans as well as three Neanderthals and one Denisovan, Akey and his team mapped the gene flow between the hominin groups over the past quarter-million years.

The researchers used a genetic tool they designed a few years ago called IBDmix, which uses machine learning techniques to decode the genome. Previous researchers depended on comparing human genomes against a “reference population” of modern humans believed to have little or no Neanderthal or Denisovan DNA.

Akey’s team has established that even those referenced groups, who live thousands of miles south of the Neanderthal caves, have trace amounts of Neanderthal DNA, probably carried south by voyagers (or their descendants).

With IBDmix, Akey’s team identified a first wave of contact about 200-250,000 years ago, another wave 100-120,000 years ago, and the largest one about 50-60,000 years ago.

That contrasts sharply with previous genetic data. “To date, most genetic data suggests that modern humans evolved in Africa 250,000 years ago, stayed put for the next 200,000 years, and then decided to disperse out of Africa 50,000 years ago and go on to people the rest of the world,” said Akey.

“Our models show that there wasn’t a long period of stasis, but that shortly after modern humans arose, we've been migrating out of Africa and coming back to Africa, too,” he said. “To me, this story is about dispersal, that modern humans have been moving around and encountering Neanderthals and Denisovans  much more than we previously recognized.”

That vision of humanity on the move coincides with the archaeological and paleoanthropological research suggesting cultural and tool exchange between the hominin groups.

A DNA insight

Li and Akey’s key insight was to look for modern-human DNA in the genomes of the Neanderthals, instead of the other way around. “The vast majority of genetic work over the last decade has really focused on how mating with Neanderthals impacted modern human phenotypes and our evolutionary history — but these questions are relevant and interesting in the reverse case, too,” said Akey.

They realized that the offspring of those first waves of Neanderthal-modern matings must have stayed with the Neanderthals, therefore leaving no record in living humans. “Because we can now incorporate the Neanderthal component into our genetic studies, we are seeing these earlier dispersals in ways that we weren't able to before,” Akey said.

The final piece of the puzzle was discovering that the Neanderthal population was even smaller than previously believed.

Genetic modeling has traditionally used variation — diversity — as a proxy for population size. The more diverse the genes, the larger the population. But using IBDmix, Akey’s team showed that a significant amount of that apparent diversity came from DNA sequences that had been lifted from modern humans, with their much larger population.

As a result, the effective population of Neanderthals was revised down from about 3,400 breeding individuals down to about 2,400.

How Neanderthals vanished

Put together, the new findings paint a picture of how the Neanderthals vanished from the record, some 30,000 years ago.

“I don’t like to say ‘extinction,’ because I think Neanderthals were largely absorbed,” said Akey. His idea is that Neanderthal populations slowly shrank until the last survivors were folded into modern human communities.

This “assimilation model” was first articulated by Fred Smith, an anthropology professor at Illinois State University, in 1989. “Our results provide strong genetic data consistent with Fred’s hypothesis, and I think that's really interesting,” said Akey.

“Neanderthals were teetering on the edge of extinction, probably for a very long time,” he said. “If you reduce their numbers by 10 or 20%, which our estimates do, that's a substantial reduction to an already at-risk population.

“Modern humans were essentially like waves crashing on a beach, slowly but steadily eroding the beach away. Eventually we just demographically overwhelmed Neanderthals and incorporated them into modern human populations.”

“ Recurrent gene flow between Neanderthals and modern humans over the past 200,000 years ,” by Liming Li, Troy J. Comi , Rob F. Bierma, and Joshua M. Akey, appears in the July 13 issue of the journal Science (DOI: 10.1126/science.adi1768 ). This research was supported by the National Institutes of Health (grant R01GM110068 to JMA).

Related Stories

Map of places where DNA have been found

Origin story: Rewriting human history through our DNA .

Joshua Akey, a professor in the Lewis-Sigler Institute for Integrative Genomics, uses "genetic archaeology" to reveal a larger picture of how we evolved as a species. The research divulges a complex history of the intermixing of early humans, indicative of several millennia of population movements across the globe.

face of a neanderthal man in an outline of the African continent

New study identifies Neanderthal ancestry in African populations and describes its origin .

After sequencing the Neanderthal genome, scientists discovered all modern humans carry some Neanderthal ancestry in their DNA — including Africans, which was previously not known. 

Illustration of Flores village, pygmy, Homo florensiensis, and cave

Modern Flores Island pygmies show no genetic link to extinct 'hobbits' .

A modern pygmy population living on an Indonesian island near a cave with Homo floresiensis fossils appears to have evolved short stature independently.

Genomic detectives crack the case of the missing heritability .

Despite years of research, the genetic factors behind many human diseases and characteristics remain unknown, and has been called the "missing heritability" problem. A new study by Princeton University researchers, however, suggests that heritability in humans may be hidden due only to the limitations of modern research tools, but could be discovered if scientists know where (and how) to look.  

Fernandina giant tortoise

'Fantastic giant tortoise,' believed extinct, confirmed alive in the Galápagos .

A tortoise from a Galápagos species long believed extinct has been found alive and now confirmed to be a member of the species. Fernanda, named after her Fernandina Island home, is the first of her species identified in more than a century.

A gene that shaped the evolution of Darwin's finches .

Researchers from Princeton University and Uppsala University in Sweden have identified a gene in the Galápagos finches studied by English naturalist Charles Darwin that influences beak shape and that played a role in the birds' evolution from a common ancestor more than 1 million years ago. The study illustrates the genetic foundation of evolution, including how genes can flow from one species to another, and how different versions of a gene within a species can contribute to the formation of entirely new species.

In one of nature's innovations, a single cell smashes and rebuilds its own genome .

A study led by Princeton University researchers found that a pond-dwelling, single-celled organism has the remarkable ability to break its own DNA into nearly a quarter-million pieces and rapidly reassemble those pieces when it's time to mate. This elaborate process could provide a template for understanding how chromosomes in more complex animals such as humans break apart and reassemble, as can happen during the onset of cancer.

Royal Society of Chemistry

Automation and machine learning augmented by large language models in a catalysis study

ORCID logo

First published on 26th June 2024

Recent advancements in artificial intelligence and automation are transforming catalyst discovery and design from traditional trial-and-error manual mode into intelligent, high-throughput digital methodologies. This transformation is driven by four key components, including high-throughput information extraction, automated robotic experimentation, real-time feedback for iterative optimization, and interpretable machine learning for generating new knowledge. These innovations have given rise to the development of self-driving labs and significantly accelerated materials research. Over the past two years, the emergence of large language models (LLMs) has added a new dimension to this field, providing unprecedented flexibility in information integration, decision-making, and interacting with human researchers. This review explores how LLMs are reshaping catalyst design, heralding a revolutionary change in the fields.

1 Introduction

Recently, large language models (LLMs) such as GPT-x, ERNIE Bot, Claude-x, and Lamma-x, 41 have begun to dramatically enhance these four technological pillars. By processing natural language, automating code generation and data analysis, optimizing design of experiment (DoE) algorithms, and facilitating human–computer interaction, 16,42–47 LLMs are setting new standards for efficiency and innovation in catalysis research ( Fig. 1 ). These capabilities allow for the extraction and utilization of data from diverse and unstructured sources such as scattered texts, videos, and images, previously inaccessible to more traditional ML technologies that relied on well-organized datasets.

The workflow of catalyst design and discovery with information extraction, automated chemical experimentation, active machine learning, and interpretable machine learning.

Moreover, automated and intelligent robotic systems, which have seen significant adoption over the last decade, spanning from flow systems 19,48,49 to desktops 50,51 and humanoid mobile robots, 4,5 now seamlessly integrate with advanced LLMs. This synergy is reshaping decision-making strategies within the field, transitioning from traditional methods like Bayesian optimization 4 and active learning 32 to more sophisticated, LLM-enhanced approaches, 45,47 towards more talented self-driving labs for closed-loop discovery. This is only the beginning of a shifting paradigm to on-demand catalyst development and in silico performance scanning for catalyst design and optimization.

Despite these technological advances, the role of the human researcher remains indispensable. The interpretability of ML methods is crucial for harnessing human intellectual engagement and deriving scientific insights that can inform new design principles for high-performance catalysts. 36–39 Artificial neural networks (ANNs) 52 used to be regarded as black-box models that are hard to explain, but recent innovations such as SHapley Additive exPlanations (SHAP) 53 for graph neural networks (GNNs) and attention mechanisms in transformer models are enhancing the transparency of artificial neural networks, which were previously considered opaque. In addition, LLMs have also showcased their capabilities in extracting data mapping and articulating them in a clear plain language format.

Given the rapid pace of these advancements, it is timely to review the revolutionary shift in AI applications for catalysis research and development. This review will delve into how the integration of LLMs is redefining the four foundational ML technologies in catalysis, providing a historical perspective and discussing recent implementations that foreshadow the future of AI-assisted catalyst design.

2 High-throughput chemical information extraction

2.1 information extraction from figures.

SMARTS (SMILES Arbitrary Target Specification): an extension of SMILES, SMARTS allows for defining substructural patterns within molecules, enhancing search and analysis capabilities.

InChI (International Chemical Identifier): provides a structured and layered representation of chemical data, facilitating interoperability across different data systems.

SELFIES (Self-referencing Embedded Strings): designed to ensure the validity of molecules represented, enhancing data integrity.

These string representations, integral to systematic chemical naming, have become increasingly valuable with the advent of language models. The seamless integration of these formats into LLMs enhances their utility, making them more than just systematic nomenclature but a dynamic part of molecular data processing. Furthermore, the development of multi-modal large models allows for directly translating structural drawings to the string representations without prior conversion, marking a significant advancement in the field. 56

In 2014, Simone Marinai et al. 70 made an improvement by introducing a Markov logic-based probabilistic logic inference engine ( Fig. 2 ). This development improved the ability to clean up noisy extractions, although challenges with fragmented elements persisted. More recently, in 2021, Yifei Wang et al. 59 advanced the field further by employing a Single Shot MultiBox Detector (SSD) neural network combined with a Non-Maximum Area Suppression (NMAS) algorithm. This combination was specifically designed to enhance object identification within a single frame, significantly improving segmentation accuracy to 89.5% on a dataset of 2100 handwritten cyclic compound samples.

Scheme of the Markov logic OCSR with low-level image information extraction and probabilistic logic inference. Reproduced with permission from ref. Copyright 2014, American Chemical Society.

(1) Insufficient understanding of embedded rules: the complexity of the embedded rules can lead to misinterpretations and errors in data extraction.

(2) Susceptibility to noise: the intricate rules are prone to interference from noisy data, which can degrade the quality of the output.

Overview of the integrated DECIMER workflow including image segmentation, classification, and translation to obtain SMILES. Reproduced with permission from ref. under CC BY license.

The accuracy and reliability of OCSR continue to improve as newer models are developed and refined. The use of multiple models for cross-validation purposes enhances robustness, offering better performance than what could be achieved by a single model. This progress is vital as it addresses the significant challenge of extracting organic reaction data on a large scale, a task that is increasingly crucial due to the exponential growth of available chemical data.

The continuous improvement of multimodal LLMs is expected to revolutionize how scientific results are communicated and utilized. As these models become more sophisticated, they will enable the scientific community to integrate vast amounts of data in unprecedented ways. This integration is anticipated to lead to the development of new tools that could dramatically enhance the efficiency and creativity of catalyst design processes. The ability to compile and analyze the extensive data generated globally by researchers represents a transformative shift towards data-driven science, promising significant advancements in how we discover and develop new materials.

2.2 Text information extraction with language models

Scheme of the automated chemical reaction extraction from scientific literature. Reproduced with permission from ref. Copyright 2019, American Chemical Society.
Scheme of the ChatGPT chemistry assistant workflow to extract synthesis information of MOFs from the literature. Reproduced with permission from ref. Copyright 2023, American Chemical Society.

2.3 Summary

Method Type Extracted content Supported modality Open source Reference
CLiDE Rule-based Molecular structures and charge Text & image Yes
OSRA Rule-based Molecular structures Text & image Yes
Imago Rule-based Depicted molecules with up and down stereo bonds and pseudostems Text & image Yes
MSE-DUDL ML-based Structures of natural products and peptide sequences Image No
DECIMER ML-based Chemical classes, species, organism parts, and spectral data Image Yes
MolMiner ML-based Molecule structures Image No
ChemDataExtractor LM-based Identifiers, spectroscopic attributes, and chemical property attributes (e.g., melting point, oxidation/reduction potentials, photoluminescence lifetime, and quantum yield) Text Yes and
SciBERT LM-based Identifiers of chemicals Text Yes
ChemRxnExtractor LM-based Reactants, catalysts, and solvents for reactions Text Yes
GPT-3.5 LM-based MOF synthesis Text No
GPT-4 LM-based   Text & image No  

The rule-based OCSR systems, once dominant, are now increasingly complemented or surpassed by neural network-based methods due to their flexibility and growing accuracy. These machine learning-based systems are not only more adaptable but also continue to improve as they learn from more data. The incorporation of rule-based techniques as a supplementary approach provides a layered methodological depth that enhances the overall robustness and generalizability of these technologies.

Language model-based systems, particularly those utilizing advanced LLMs, represent the frontier of chemical information extraction. Although their full potential is yet to be realized, the rapid evolution into multimodal models suggests that transformative developments could emerge shortly. These models are particularly promising for handling the vast and complex data typical in catalysis research.

The transition to open-source methods has also played a critical role in this field. Beginning with systems like OSRA in the 1990s, the move towards open-source has not only facilitated wider access to advanced tools but has also spurred innovation and customization, enhancing the collective capability of the research community.

This evolving landscape of chemical information extraction methods underscores the importance of continual adaptation and development to harness the ever-increasing volumes of data in catalysis and other fields of chemistry.

3 Automated and intelligent chemical robotic system

Over the past few decades, significant advancements in automation have led to reductions in costs and enhancements in the efficiency, accuracy, and reproducibility of experiments. 83–86 The origins of chemical automation date back to the 1960s and 1970s with the development of automated devices like automated peptide synthesizers, 87 DNA synthesizers, 88 and organic synthesis modules. 89 This was followed by the emergence of high-throughput automated synthesis systems in the era of combinatorial chemistry. 90–96 More recently, the introduction of humanoid chemical robots 4–6 and autonomous flow-based synthesis platforms 17–19 has marked a new era of innovation in intelligent chemical synthesis.

A notable feature of this latest advancement is the interactive “ask and tell” process, such as active learning, where models are continuously trained on current observations and actively request additional data. This interactive approach can significantly accelerate discovery efficiency compared to traditional screening strategies. 97 Therefore, experimental processes must be designed to be not only high-throughput but also sufficiently flexible to allow frequent access and modifications. This is also the stage where LLMs can contribute, integrating crucial domain knowledge to enhance exploration and decision-making processes.

In this section, we will discuss how advancements in hardware design, coupled with LLMs, enhance operational flexibility. Later, we will explore the promising potential of LLM-driven active learning in the subsequent section.

3.1 Automated and intelligent chemical experiment platform

(1) Humanoid robotic systems: this approach relies on the usage of multi-axis arms that provide a high degree of operation flexibility, mimicking the behavior of human operators.

(2) Automated flow chemical systems: these systems are designed on the foundation of fluid dynamics and transport pipelines to achieve precise chemical operations, which can be seamlessly interfaced with analytical instruments.

3.2 Humanoid robotic system

The AGV-based autonomous mobile robot system launched by Andrew I. Cooper et al. 4 is a remarkable advance in chemical automation. The team found improved photocatalysts for producing hydrogen from water after autonomous running for 8 days, completing 688 experiments in a design space of 10 variables. The robot ( Fig. 6 ) can handle sample vials among eight workstations distributed around the lab, including a solid reagent dispensing system, a mixed liquid dispensing system and capping module, an ultrasound module, a photolysis module, a gas chromatography (GC) analysis module, and three separate sample storage modules to achieve a variety of experimental tasks.

Autonomous mobile robot and experimental stations. The mobile robotic chemist (a), the roadmap of the whole laboratory (b) and several workstations (c–e) are shown. Reproduced with permission from ref. Copyright 2020, Springer Nature.

Despite the great advances, the mobile robotic chemist from Cooper's group is purely driven by Bayesian algorithms and does not capture existing chemical knowledge or include theoretical or physical models. Later, a comprehensive artificial intelligence chemistry laboratory ( Fig. 7 ) was developed by Jun Jiang's team. 5 This AI-Chemist consists of three modules, including a machine-reading module, a mobile robot module, and a computational module. The AI-Chemist system responds to scientific questions posed by researchers by tapping into vast amounts of literature. It digitizes and standardizes experimental protocols, enriching its knowledge base. The platform manages tasks, monitors the mobile robots, customizes experiment workflows, and stores the data for future use. The research team used the platform to find the best combinations of several Martian meteorite rocks to synthesize efficient water oxidation catalysts for future use in Martian exploration. 98

Design of the all-round AI-Chemist with a scientific mind. It includes three modules for chemistry knowledge, autonomous experimentation, and theoretical computation and machine learning (A). The workflow of the AI-Chemist to study various systems are shown in (B). Reproduced with permission from ref. Copyright 2022, China Science Publishing & Media Ltd.

The recent A-lab, developed by Gerbrand Ceder et al. , 6 represents a significant advancement in the field of solid material synthesis. Despite some controversy on the actual phases of the fabricated materials, the hallmark of the A-lab is its high degree of automation, which encompasses the entire synthesis and characterization process, including critical steps such as powder dosing, sample heating, and X-ray diffraction (XRD) for product characterization.

One critical issue with the robotic arm system in laboratory settings is its moderate capacity to parallelize experimental tasks. While robotic arms bring automation and precision to the table, they still mimic human researchers to conduct multiple operations one by one. This constraint is particularly evident in high-throughput settings where speed and efficiency are paramount. To address this, integrating robotic systems with other automated solutions might be necessary.

3.3 Automated flow chemical system

Physical implementation of the synthesis platform Chemputer. The scheme (A) and the actual set-up (B) of the Chemputer are shown respectively. Reproduced with permission from ref. Copyright 2019, AAAS.

One drawback of many flow systems is the lack of flexibility for different experiment tasks. One solution is to use general modules and their combination to support wider experiments. Alternatively, the modules can be reaction-specific as long as they can be designed and fabricated efficiently. Leroy Cronin et al. 49 showcased a portable, suitcase-sized chemical synthesis platform with automated on-demand 3D printing of groups of reactors for different reactions. Researchers demonstrated the broad applicability of this system by synthesizing five organic small molecules, four oligopeptides, and four oligonucleotides, achieving good yields and purity.

The implementation of batch reactors with increased throughput has accelerated the search for catalysts in more complex systems that involve multiphase reactions. Cheng Wang et al. 106 developed a fast screening platform with a coherent implementation of automated flow cell assembly and GC characterization. It was used for parallel synthesis, electrochemical characterization, and catalytic performance evaluation of electrocatalysts for the reduction of CO 2 to C 2+ products, which led to the discovery of a Mg–Cu bimetallic catalyst with competitive CO 2 to C 2+ performance and good stability compared to the top catalysts from other literature reports ( Fig. 9 ).

Fast screening platform for screening bimetallic catalysts for the CO RR. (a) Schematic illustration of the fast screening platform for the CO RR. (b) Exploded view of a 3D-printed flow cell. (c) Heat map of the relative FE of C H over Cu-based bimetallic catalysts. Elements in black font represent tested metal salt additives and elements in grey font represent the untested ones. Reproduced with permission from ref. Copyright 2022, Wiley.

Timothy F. Jamison et al. 115 developed a flexible, manually reconfigurable benchtop flow chemistry platform ( Fig. 10 ), including various reactor modules for heating/cooling, photochemical reaction, and packed bed reaction. In addition, the platform integrates liquid–liquid separation technology and is equipped with inline analysis tools such as high performance liquid chromatography (HPLC), Fourier transform infrared spectroscopy (FTIR), Raman spectroscopy, and mass spectrometry.

Plug-and-play, reconfigurable, continuous-flow chemical synthesis system. The workflow (A), the design of the flow system (B) and its actual setup (C) with interchangeable modules (D) are shown in the figure. Reproduced with permission from ref. Copyright 2018, AAAS.

One issue of the continuous flow system is its high cost in paralleling and adaptation. To partly address this issue, Kerry Gilmore et al. 116 reported a “radial synthesizer” based on a series of continuous flow modules arranged radially around a central switching station, which allows selective access to individual reactors and avoids equipment redundancies and reconfiguration among different reactions. Storing stable intermediates inside fluidic pathways enables simultaneous optimization of subsequent steps during route development. Online monitoring via infrared (IR) and 1 H/ 19 F NMR spectroscopy enables fast post-reaction analysis and feedback. The performance of this system has been demonstrated in transition metal-catalyzed C–C and C–N cross-coupling, olefination, reductive amination, nucleophilic aromatic substitution reactions, light-driven oxidation-reduction catalysis, and continuous multi-step reactions. In addition, flow selection valve technology can be used to create different process combinations, as demonstrated by Nathan Collins et al. 117 in an advanced automated continuous flow synthesizer called AutoSyn, which can access 3800 unique process combinations and up to seven consecutive reaction steps for efficiently preparing a variety of pharmaceutical small molecule compounds with a scale from milligrams to grams within hours.

To make the fluidic system even more adaptive, Klavs F. Jensen et al. 123 combined the robotic arm and the flow system ( Fig. 11 ): the robotic arm is responsible for assembling modular process units, including reactors and separators, into a continuous flow path. After the synthesis, the robotic arm can disconnect the reagent lines and move the processing module to the appropriate storage location. Pneumatic grippers are used to ensure tight connections between process chambers. In 2023, the same group introduced a prototype that further incorporates machine learning with robotics to autonomously design, synthesize, and analyze dye-like molecules with minimal human intervention. 124 This system successfully synthesized and characterized 303 new dyes, advancing the efficiency of chemical discovery.

A robotically reconfigurable flow chemistry platform. Reproduced with permission from ref. Copyright 2019, AAAS.

Flow chemistry systems, while revolutionizing chemical synthesis and processing, present several limitations in automation. The setup and maintenance of these systems are complex and resource-intensive. Establishing precise control over flow rates, temperature, and pressure requires specialized equipment and expertise. This complexity also extends to scalability issues; while flow systems excel in scaling up certain types of reactions, they may be less adaptable for reactions requiring long residence times or intricate synthesis steps. Additionally, the rigidity in altering reaction conditions can limit their flexibility, making them less suitable for laboratories that frequently switch between diverse chemical processes. Material compatibility is another concern, as the construction materials of the flow reactors must withstand a wide range of chemicals and conditions, limiting their use with highly reactive or corrosive substances. Furthermore, while adept at handling large-scale production, flow chemistry systems can be less efficient for small-scale synthesis, often leading to inefficiencies and wastage when dealing with minute quantities.

3.4 Large language models and robots

First, LLMs have facilitated the development of robotics, including log information extraction, assisted robot design, 125 and task generation and planning. 42,43,126,127 As pointed out by Francesco Stella et al. , 125 LLMs can be the creator for designing the automating system, be the mentor and copilot for domain scientists who do not have the necessary educational background to implement automation in their research, and be an assistant to debugging, troubleshooting, and method selection during the technology implementation phase to accelerate the process.

Second, LLMs, especially the multimodal ones, can help develop next-generation robots with increased flexibility. Vemprala and others from the Microsoft team 126 proposed a strategy that combines prompt engineering and a high-level feature library to enable ChatGPT to handle various robotic tasks and scenarios. An open-source tool called PromptCraft was introduced, which includes a collaboration platform and a ChatGPT-integrated sample robot simulator. However, the LLM-controlled robotic movement is not robust enough for direct use in chemistry experiments where safety and reliability are of primary concern.

Third, LLMs also offer solutions to program robots. Kourosh Darvish et al. introduced the CLAIRIFY method, 42 which combines automatic iterative prompting with program verification to ensure the syntactic accuracy of task plans and their alignment with environmental constraints. The system's objective is to produce a syntactically correct task plan suitable for robotic action as a prompt for LLMs to generate a program. However, the generated plan needs to be verified to detect any compilation error and pass the error messages as subsequent input prompts for iterative interaction with the LLMs. The capability of this method was demonstrated by translating natural language to an abstract and concise high-level chemical description language (χDL), which was originally developed and used in the control of Chemputers. 18

Compared to high-level descriptive codes, generating low-level operational codes to interface directly with the robotic system can be more complicated. Genki N. Kanda et al. 43 demonstrated that GPT-4 can generate low-level operational Python scripts for automated robots like Opentrons-2 (OT-2) from natural language instructions. They designed a pipeline based on GPT-4 to automatically translate natural language experimental descriptions into Python scripts compatible with OT-2. Leveraging OpenAI, this approach iteratively queries the model, extracts, and validates scripts using a simulator of OT-2, and provides feedback on any errors for correction. This shift towards natural language instruction simplifies the automation process, making it accessible to a broader range of researchers and promoting the automation of biological experiments.

3.5 Summary

Type Description Synthesis operations Characterization Target compounds Reference
Humanoid Mobile robotic chemist Solid dispensing, liquid dispensing, capping/uncapping, heating, and sonication Gas chromatography Catalysts for photolysis of water to produce hydrogen
An all-round AI-Chemist Solid dispensing, liquid dispensing, magnetic stirring, sonication, drying, centrifugation, and liquid extraction UV-vis, fluorescence, and Raman spectroscopy, and gas chromatography Materials for electrocatalysts, photocatalysts, and luminescence
A-lab, an autonomous laboratory Powder dosing and sample heating X-ray diffraction (XRD) Primarily oxides and phosphates identified through extensive ab initio phase-stability data
Flow: Batch reactors Modular robotic synthesis system Mixing, filtration, liquid–liquid separation, evaporation, and chromatographic separation Organic molecules
A portable suitcase-sized chemical synthesis platform Liquid transfer, temperature control, evaporation, filtration, and separation Organic molecules
Fast screening platform for the CO RR Liquid handling, electric cell preparation, and electrolysis Micro-fast gas chromatography Electrocatalysts for the CO RR
Flow: continuous flow reactors Benchtop flow chemistry platform Liquid handling, heating, cooling, photoreaction, extraction and purification High-performance liquid chromatography (HPLC), IR spectroscopy, Raman spectroscopy, and mass spectrometry Reconfigurable system for automated optimization of diverse chemical reactions
Radial synthesizer system Liquid transfer, mixing, and dilution IR spectrometry and nuclear magnetic resonance Cross-coupling, olefination, reductive amination, nucleophilic aromatic substitution reactions, light-driven redox catalysis, and continuous multi-step reactions
An automated multistep chemical synthesizer Heating, liquid–liquid separation, gas–liquid separation, and heterogeneous catalysis Liquid chromatography-mass spectrometry (LC-MS) Pharmaceutical small molecules
Humanoid robotic system with flow reactors A robotic platform for flow synthesis of organic compounds Liquid handling, separation, and temperature adjustment High-performance liquid chromatography and nuclear magnetic resonance Organic molecules

We expect a much enhanced automation level in chemistry research. However, current automation in chemistry still faces challenges, particularly in the trade-offs between the flexibility and throughput of automated systems. For instance, although capable of vast amounts of operations compared to flow systems, humanoid robotic systems are usually slower in operational speed to ensure accuracy and safety. On the other hand, flow chemistry systems can handle hundreds or thousands of experiments per day, but are more task-specific with limited flexibility. New developments in these strategies are required to enhance flexibility, throughput, and robustness at the same time.

Another challenge lies in the control part of the robotic systems. Although digital twins are very common for humanoid robotics and in industry, the development of digital twins for the whole automated chemistry system is still at its initial stage despite a few efforts. 4,18,128 Ensuring the integrity and safety of experimental procedures remains paramount in automation labs. Therefore, greater attention must be directed toward enhancing the capability to simulate experimental procedures and detect any potential physical or chemical issues during the development of various robotic systems. Furthermore, despite the rapid advancements in novel algorithms, such as reinforcement learning, the control of robots in chemistry labs often relies on hardcoded programming. This limitation restricts their ability to perform complex tasks and adds challenges to the maintenance, transferability, and future development of the systems. LLMs appear promising in introducing flexibility to control systems. However, the reliability of LLM-generated code must be verified either by human experts or through digital twins. It is foreseeable that digital twins and LLMs will soon be more cohesively integrated into the control of chemical robotic systems.

4 Design and discovery of catalysts with active machine learning

4.1 design of catalysts guided by machine learning.

BO is an optimization strategy that balances the exploration of uncertain regions and the exploitation of known regions with superior objective values. It is generally used to optimize a black-box function and consists of three key components:

(1) Surrogate model: this is a predictive model designed to approximate the underlying function. A wide range of machine learning models can be employed for this purpose, such as the Gaussian process, 131 ensembles of artificial neural networks, 132 and Bayesian neural networks. 133–136

(2) Acquisition function: an acquisition function is a scoring function used to rank sampling points within the input space based on the surrogate model's predictions. Examples of such functions include expected improvement (EI), 137,138 probability of improvement (PI), 139 and upper confidence boundary (UCB). 140 The acquisition function is instrumental in selecting the most promising candidates for further evaluation.

(3) Bayesian inference: 141 this is a foundational technique in Bayesian optimization, utilized for training the surrogate model. It uses Bayes' theorem to update the probability of a hypothesis or event based on observed evidence.

On the other hand, AL is a family of machine learning techniques that aims to minimize the number of labelled data points while obtaining a high-performance model. It can usually be achieved through an adaptive sampling strategy, which prioritizes the labelling of data points with the highest uncertainty and information gain for the model.

Both BO 4,20–28 and AL 29–34 have been applied in the design of and search for catalysts. BO can efficiently explore the vast parameter space of catalyst design and select experiments that are likely to yield the desired products. By iteratively updating the ML model and selecting new experiments based on the retrained model, BO can guide the search for optimal catalysts. AL, in the meantime, can assist in selecting the most informative data points for labelling, reducing labelling costs while improving model performance. It has been applied in many fields including materials design, 142,143 retrosynthesis, 144,145 and drug discovery. 146,147 Besides the original purpose of AL, its application in catalyst design also demonstrated its capability for global optimization, presenting a remarkable analogy to the BO framework. The applications of BO and AL in the field of catalysis will be discussed respectively below.

4.2 Bayesian optimization

Bayesian optimization of the methanol electro-oxidation process. (a) Peak current density of methanol electro-oxidation as a function of the number of BO rounds. (b) A contour plot showing the peak current density and a ternary plot depicting the chemical composition in the electrolyte solution. Reproduced with permission from ref. Copyright 2020, Royal Chemical Society.

In 2020, Bayesian experiments for autonomous researchers (BEAR) 21 combined BO with high-throughput automated experiment systems to achieve self-driven material discovery—a cycle of the design of experiments, automated experiment feedback, and retraining of machine learning models to design new experiments. As discussed before, Andrew I. Cooper et al. 4 developed an AI chemist to improve the catalytic performance for hydrogen production with BO ( Fig. 13 ). It successfully discovered a mixture of photocatalysts that exhibited six times higher activity than the original formulation. Compared to manual operations, the experimental time cost is reduced by approximately 60 times.

(a) Maximum rate of hydrogen evolution from photolyzed water reaching 21.05 μmol h after 688 experiments during an 8-day autonomous search. (b) Radar plot illustrating the sampling in the search space during experimentation. Reproduced with permission from ref. Copyright 2020, Springer Nature.

In 2021, Jan Rossmeisl et al. 22 developed a computational framework that combines density functional theory (DFT) calculations, ML-driven kinetic modelling, and BO to explore a wide range of composition space to search for multi-component high entropy alloys for the oxygen reduction reaction (ORR). To accelerate catalyst discovery, the authors integrated kinetic modelling with BO, where a Gaussian-process-based surrogate model provided suggestions for alloy compositions. The proposed compositions were evaluated using the kinetic model, and the surrogate model was updated based on the ORR activity predicted by the kinetic model. BO effectively identified optimal compositions through 150 iterations, including Ag 18 Pd 82 , Ir ≈50 Pt ≈50 , and Ir ≈10 Pd ≈60 Ru ≈30 ( Fig. 14 ). These compositions closely matched the optimal compositions found through grid search in the same chemical space. Experimental confirmation of the three optimized compositions by high-throughput thin-film synthesis and ORR testing in the Ag–Pd, Ir–Pt, and Pd–Ru binary alloy spaces, reveals the best-performing compositions of Ag 14 Pd 86 , Ir 35 Pt 65 , and Pd 65 Ru 35 . The experimental results reasonably matched the results of BO, and BO can accelerate the discovery of optimal catalysts by up to 20 times.

BO for composition optimization of an Ag–Ir–Pd–Pt–Ru system for the ORR. (a) Pseudo-ternary plots (Ir, Pt, and Ru collected at one concentration) showing pseudo-functions after sampling 15, 28, 54, and 150 compositions. Yellow indicates regions with higher absolute values of the simulated current densities, and blue indicates regions corresponding to lower values. (b) Current densities sampled during BO (black solid line) and the emergence of the three most active locally optimal compositions (blue dashed line). (c) and (d) Variation of the GP-squared exponential kernel function with respect to the constant term (c) and the length scale (d) hyper-parameters. Reproduced with permission from ref. Copyright 2021, Wiley.

4.3 Active learning

Yousung Jung et al. 30 proposed an active learning method in the discovery of catalysts for the CO 2 RR driven by uncertainty and prediction error. It utilizes cost-effective non- ab initio input features, i.e. , LMTO d-band width and electronegativity, as chemisorption descriptors to predict adsorption energies on alloy surfaces. Screening of large-scale materials is carried out by combining these descriptors with two machine learning models: an ensemble of artificial neural networks (ANNs) and kernel ridge regression (KRR). The catalytic performance of a set of 263 alloy systems was studied by predicting *CO binding energy using the models. During the active learning process, an ensemble consisting of five neural networks with the same architecture but varied initial weights was trained on an initial dataset. The ensemble was used to predict the *CO binding energy on the rest of the dataset to find candidates with the highest prediction variance, which will be included in the next training process. As an alternative machine learning model, the performance of KRR 148,149 was also elaborated. It involves the training of a KRR model on the initial dataset with *CO binding energy as the output. Then, an additional KRR model was trained on the prediction error from the previously trained model as an error predictor. 148,150 Later, the KRR error predictor was used to estimate the error rate for the rest of the dataset, which helps select candidates for the next round of training. Both models (ensemble of ANNs and the KRR model) were used to predict the adsorption energy of CO on (100) crystalline surfaces. The best model gives an RMSE of only 0.05 eV without the d-band center as a descriptor. The authors discovered Cu 3 Y@Cu* to be a highly active and cost-effective catalyst for the CO 2 RR.

Besides the original purpose of using active learning to establish an accurate and reliable model, it can also be utilized for global optimization. In 2018, Zachary W. Ulissi et al. 31 proposed a cyclic workflow with ideas from agent-based model optimization and active learning for screening electrocatalysts for the CO 2 RR and HER. This workflow, illustrated in Fig. 15 , involves machine learning screening, DFT validation, and machine learning retraining. To start, the researchers obtained a search space of intermetallic crystals and their corresponding surfaces from the Materials Project. 151 They then selected a series of materials as optimal candidates for catalysis using a machine-learning model. DFT calculations for the selected candidates were performed, providing more accurate predictions of the catalytic properties. The DFT results were then used to retrain the machine learning model, creating an iterative process for continuously improving the catalyst database. In their study, the authors considered a total of 31 elements, composed of 50% d-block elements and 33% p-block elements. The search space consists of 1499 intermetallics for potential catalysis applications. 131 possible surfaces from 54 alloys and 258 possible surfaces from 102 alloys were identified as valid candidates for the CO 2 RR and HER, respectively. The number of candidate alloy catalysts can be further reduced to 10 and 14 for the CO 2 RR and HER. This comprehensive screening approach allowed for the identification of theoretically promising catalysts for the CO 2 RR and HER.

Workflow for automating theoretical materials discovery. (a) and (b) The experimental workflow for catalyst discovery is accelerated by the ab initio DFT workflow. (c) Scientists relied on their expertise and experimental results to screen data for DFT calculations traditionally. (d) This work uses ML to select DFT data automatically and systematically. Reproduced with permission from ref. Copyright 2018, Springer Nature.
Screening of CO RR electrocatalysts using an active learning algorithm based on the DFT framework. (a) A two-dimensional activity volcano plot of the CO RR. (b) A two-dimensional selectivity volcano plot of the CO RR. (c) DFT calculations were performed on approximately 4000 adsorption sites of Cu-containing alloys identified by t-SNE. On the right, the Cu–Al clusters are labeled numerically. (d) Representative coordination sites for each cluster are labeled in the t-SNE. Reproduced with permission from ref. Copyright 2020, Springer Nature.

While BO and AL are initially different approaches, they tend to converge on the catalyst optimization task. BO usually uses a probabilistic model with the goal of optimization, while AL can adopt more diverse models with the goal of efficiently constructing a machine learning model. When AL also used a probabilistic model and assessed uncertainty in making the decision about which point to explore next, it is equivalent to exploration-oriented BO, but the ultimate goal of AL is to improve the model most efficiently, which is beyond the uncertainty strategy.

When all the obtainable information about the system comes from the previous experimental/calculation results, BO and AL are mathematically sound methods to most efficiently explore the space. However, when domain knowledge is available, it is possible to come up with a more efficient strategy by combining the testing information with domain knowledge. The addition of domain knowledge into the process can be achieved by using LLMs.

4.4 Design and synthesis of catalysts guided by large language models

In BO and AL, a machine learning model (or a surrogate model) is necessary to approximate a mapping. Traditional machine learning models can take continuous, discrete, or categorical variables as the input. In contrast, LLMs, with their inherent capabilities to process natural language descriptions and generate new content accordingly, can be potentially used as a surrogate model, which can support a versatile input format. To incorporate the training data into the models, in-context learning (ICL), a technique that includes training data as examples in the prompt for LLMs, can be used. Alternatively, fine-tuning the models using the existing dataset represents another viable approach.

Andrew D. White's group 45 demonstrated the usage of LLMs as the surrogate model in Bayesian optimization. The aim is to use a generative pre-trained transformer (GPT) as a surrogate model to predict the properties of the product according to the experimental procedure. Both fine-tuning and ICL were used for model training. To introduce prediction uncertainty when querying the LLMs, they designed two prompting strategies, a (1) multiple-choice option template and (2) top k completions template for regression. With the multiple-choice template, the LLM will treat the regression problem as a multi-option question to give a predicted value in one of the five ranges. Furthermore, the probability of selecting each option can be accessed. In the top k completion template, the question will be queried k times to the LLM to generate k answers. Both strategies generated a discrete probability distribution of the output, which can be used in Bayesian optimization. The authors used a series of models from OpenAI (text-curie-001, text-davinci-003, GPT-4, etc. ) with in-context learning or fine-tuning to predict the C 2 yield for oxidative coupling of methane based on synthesis procedures. Gaussian process regression was used as a baseline with text embedding to convert the synthesis description to a numeric input. Among the LLMs, GPT-4 is the best model in either ICL or fine-tuning. When GPT-4 and the top-k completion strategies were used, the ICL model showed comparable performance (mean absolute error, which is abbreviated as MAE, of 1.854) to the Gaussian process regression (MAE of 1.893). When the fine-tuning was implemented, the MAE of the model was further decreased to 1.325. Later, the authors implemented Bayesian optimization using the Gaussian process or LLMs with ICL as the surrogate model. The ICL model reached 99% quantile after 15 samples, after which the performance did not improve significantly and failed to find the maximum value in the sample pool. Although the GPR model also failed to find the maximum in the sample pool, it was a little closer to the maximum and showed a higher efficiency in the optimization. Due to the token size limitation and the complexity of the C 2 data, the authors only selected the five most relevant examples during ICL, which can be the reason why the ICL model did not perform as well as the GPR model in Bayesian optimization. However, as a proof-of-concept, it is enough to demonstrate that LLMs have the potential to guide researchers in decision-making.

The in-context learning ability of the LLMs is promising for building an interactive workflow where an AI agent iteratively assists and instructs human experts to increase search efficiency through experimental feedback. Recently, Omar M. Yaghi and his coworkers have built such a workflow and demonstrated its capability in the synthesis of MOFs with prompt engineering and in-context learning. 47 This innovative workflow involves three components: ChemScope, ChemNavigator, and ChemExecutor ( Fig. 17 ). With the usage of ChemScope, the human researchers offer GPT-4 the project goals and necessary information like the literature of reticular chemistry and availability of lab resources to generate a project blueprint. Here, GPT-4 reads the general concepts of reticular chemistry and constructs a scheme of the project with multiple stages, where each stage contains well-defined objectives and indicators for their completion. Then, ChemNavigator and ChemExecutor were used coherently to go through the stages and achieve the objectives defined by ChemScope. ChemNavigator was used to define tasks to complete the objectives of the current stage. It takes the project scheme from ChemScope, previous trial-and-error summaries, human feedback, and current situations to update the summaries and generate three tasks accordingly. With the updated summary and tasks from ChemNavigator, the ChemExecutor outputs step-by-step instructions to complete the task. Additionally, ChemExecutor also defines a template to record the experimental feedback from the human researchers, which will be used later in the next iteration. At this point, the human researchers will perform the experiments and fill up the template. The interaction among ChemExecutor, ChemExecutor, and human researchers was iterated several times until the completion of the project. The recording of experimental feedback and consistent updating of the summary enabled GPT-4 to learn from experiment outcomes and optimize protocols to complete the complex tasks. Using this human–computer interactive workflow, the researchers successfully discovered and characterized a series of isomorphic MOF-521s. This work highlights the advantages of the large language model in interacting with human experts in natural language without coding skills, making it easy to use for all chemists. Additionally, the in-context learning facilitated by GPT-4 can continuously optimize experimental protocols to complete complicated research tasks. When such a workflow is integrated with automated robotic systems, it paves the way for a new paradigm of self-driving labs, where the design and discovery of catalysts go beyond a purely data-driven approach.

Framework diagram for GPT-4-directed MOF synthesis. The workflow consists of three phases: Reticular ChemScope, Reticular ChemNavigator, and Reticular Executor. The ICL capability of GPT-4 is achieved by combining pre-designed prompt systems with continuous human feedback. Reproduced with permission from ref. Copyright 2021, Wiley.

Despite the potential applications of LLMs in the design of and search for catalysts, there are still some problems to be addressed. The major problem is the well-known hallucinations in the context generated by LLMs. Although researchers have tried to mitigate this issue through methods such as prompt engineering, in-context learning, and fine-tuning, further improvements are needed to enable the accuracy and reliability of these models. Secondly, LLMs with direct domain expertise are still lacking. Thus, when dealing with domain-specific scientific problems, the models need to be fine-tuned; otherwise they can show low accuracy and misunderstanding. While LLMs hold promise in chemical research, further research and improvements are necessary to overcome the existing limitations and bring the application of artificial intelligence in the research of catalysts into a new era.

4.5 Summary

Type Surrogate models Variables (input) Target (output) Research systems Reference
Bayesian optimization Random forest and Gaussian process Ratio of a metal precursor (continuous) Current density Electrocatalytic oxidation of methanol
Gaussian process Reagent concentration for catalyst synthesis (continuous) Hydrogen evolution rate Photocatalytic hydrogen generation
Gaussian process Alloy compositions (continuous) Current density Electrocatalytic O reduction
Large language models from open AI Experimental procedure as text C yield Oxidative coupling of methane
Active learning Artificial neural networks and kernel ridge regression Electronegativity and d-band width of alloys (continuous) *CO binding energy (*CO refers to adsorbed CO on a solid surface) Electrocatalytic CO reduction
Extra tree regressor, random forest, Gaussian process,etc. Fingerprints of the surface and sites of intermetallics (discrete) Adsorption energies of CO and H Electrocatalytic CO reduction and H evolution
Random forest and boosted tree Fingerprints of adsorption sites from copper-containing metals (discrete) CO adsorption energy Electrocatalytic CO reduction
GPT-4 Synthesis procedure as text input Success or failure of the synthesis MOF synthesis

Several challenges persist in implementing active machine learning, particularly related to surrogate models. These models excel in interpolation rather than extrapolation, making them prone to overfitting and necessitating training data of a specific scale. Many efforts are made to improve the surrogate models for higher generality ( e.g. , Phoenics 135 ) and extend variables from simple continuous variables to discrete or categorical variables ( e.g. , Gryffin 136 ). Additionally, a crucial challenge lies in selecting relevant catalysis features compatible with surrogate models. Incorporating irrelevant descriptors can impede the effectiveness of active learning algorithms, reducing their performance to that of uniform random search. The difficulty in feature selection confines certain closed-loop searches to mere recipe optimization, treating the process as a black box and adjusting only continuous variables such as reagent ratios or concentrations ( Table 3 ). However, catalytic reaction activity and selectivity are closely linked to explicit factors such as intermediate adsorption energy, d-band center, electronegativity, and steric hindrance, which inherently serve as valid features. These features can be assessed through ab initio theoretical calculations or in situ characterization. While the advent of automated laboratories has alleviated concerns regarding insufficient data acquisition, it remains a costly endeavor, especially considering the challenges in automating certain characterization techniques. Consequently, strategies for evaluating and selecting an appropriate subset from these explicit features require further refinement. The subsequent section will delve into the detailed elaboration of chemical descriptors employed in machine learning algorithms.

5 Interpretable machine learning for catalysis

5.1 descriptors for traditional machine learning.

The design of a new ligand library for enantioselective ketone propargylation (A). It was shown the steric effect from the oxazoline group had limited influence on the reaction (B) while the electronic effect from the substract is more dominant (C). Thus, a new ligand library with a quinoline group and varied steric effects is design for further screening, as shown in (D and E). Reproduced with permission from ref. Copyright 2016, American Chemical Society.

5.2 Descriptor selection and machine learning

Artificial Neural Networks (ANNs): these models automatically extract and continuously refine descriptors through the iterative updating of network weights.

5.3 Incorporating chemical knowledge through LLMs

One such pre-trained model, Uni-Mol, incorporates 3D information in its self-training reconstruction process and has outperformed state-of-the-art models in molecular property prediction. It demonstrates strong performance in tasks that require spatial information, such as predicting protein-ligand binding poses and generating molecular conformations. 211 Similarly, Payel Das et al. showed that a motif-based transformer applied to 3D heterogeneous molecular graphs (MOLFORMER) excels by utilizing attention mechanisms to capture spatial relationships within molecules. 157 Another innovative approach, the Chemical Space Explorer (ChemSpacE), uses pre-trained deep generative models for exploring chemical space in an interpretable and interactive manner. 212 The ChemSpacE model has exhibited impressive capabilities in molecule optimization and manipulation tasks across both single-property and multi-property scenarios. This process not only enhances the interpretability of deep generative models by navigating through their latent spaces but also facilitates human-in-the-loop exploration of chemical spaces and molecule design.

Despite these advancements, caution is necessary when considering the information used during pre-training. Unlike natural languages, which are imbued with rich contextual and cultural knowledge, pure chemical structures typically contain limited information, often constrained to basic chemical rules such as the octet rule. Pre-training models solely on 2D chemical structures or 1D SMILES strings without incorporating additional chemical knowledge may lead to models that lack substantial chemical understanding.

Pre-trained models, with their capacity for insightful interpretations and enhancements in molecular predictions, hold significant promise for transforming areas in catalyst design, molecular property prediction, and reaction optimization.

Recent initiatives have leveraged the capabilities of pre-trained language models like GPT-3, fine-tuning them with chemically curated data. In 2023, Berend Smit et al. published an influential paper titled “Is GPT-3 all you need for low-data discovery in chemistry” 15 first on preprint. The title was apparently inspired by the seminal Google paper on transformers. The paper was later published in Nature Machine Intelligence with a modified title “Leveraging large language models for predictive chemistry”. 156 They experimented with fine-tuning GPT-3 using chemistry data written in a sentence and used it as a general machine learning model for classification and regression. The chemicals are represented by either SMILES or IUPAC names in natural language, which makes no difference in the prediction performance. The fine-tuned GPT-3 model achieved superior performance over traditional models in predicting material properties and reaction yields, especially in data-scarce scenarios. Its ability to accept the IUPAC names of chemicals as inputs facilitates non-specialist use. The authors explored the model's potential in generating molecules based on specific requirements and tested its in-context learning capabilities, which also showed promising results.

It is interesting to discuss what aspect of the LLM's ability is used in the task of learning chemistry data. Most likely, the LLM's abilities to learn new patterns and apply basic chemical logic are critical in these tasks. It is not clear if the LLM's general knowledge about specific molecule or functional groups is used or not. It is important to recognize that these models may not fully “understand” the underlying chemistry and should be used with caution due to their potential for producing misleading results or hallucinations. Despite these limitations, this work introduces a novel paradigm in machine learning that utilizes language models to foster advancements in low-data learning within the field of chemistry.

5.4 Interpreting machine learning results

1. Interpretability: ideally, prompts should be phrased in natural language to ensure they are easily understood by human users.

2. Accuracy: prompts must accurately map input features to outputs, providing a clear and logical explanation of the data.

There are a variety of auto-prompting methods based on gradient descent to search for a prompt that can map the input feature to the output values. 222–224 However, as a result of gradient descent, it is not guaranteed that these searched prompts are generally interpretable. Additionally, gradient-descent-based methods are usually computationally expensive. To address these two problems, Jianfeng Gao et al. 44 introduced an interpretable auto-prompting method (iPrompt) using LLMs to directly generate and modify the prompts. There are three steps to search for ideal prompts in this method:

(1) Prompt proposal: in this stage, a prefix of data points is fed to the LLMs, requiring them to complete the prompts that map the input features to the output values. It generates a series of candidate prompts that will be evaluated further.

(2) Reranking: the performance of the candidate prompts from (1) is evaluated, and those that maximize the accuracy will be maintained.

(3) Iterate with exploration: the top candidate prompts from (2) will be truncated randomly. Then the truncated prompts will be fed to LLMs to regenerate new prompts while maintaining accuracy.

This iterative process continues until no further improvements are observed. The direct generation and modification of prompts by LLMs in steps 1 and 3 enhance interpretability, while accuracy is optimized in step 2. However, despite their impressive capabilities, LLMs may still lack depth in mathematical rigor, theoretical simulation, or specialized domain knowledge required for some catalysis applications. Incorporating AI agents equipped with a comprehensive toolkit could potentially address these limitations, enhancing both the interpretability and accuracy of machine learning models in catalysis.

5.5 Summary

Recent developments in graph-based and latent space descriptors of pre-trained models are attracting increasing attention, despite sometimes not providing direct insights. These descriptors are valued for their potential to bridge sophisticated computational models with practical chemical understanding, a connection that is strengthening due to ongoing algorithmic improvements.

Model-agnostic methods like SHAP, LIME, and PDP provide robust frameworks for interpreting machine learning models. However, the methods need a significant update to meet the new challenge due to the involvement of LLMs.

As we look to the future, the enhancement of interpretable models and the expansion of model-agnostic methods are set to increase AI's utility beyond mere speed and accuracy. By integrating tailored, interpretable descriptors across different systems, this approach not only deepens chemical insights but also empowers the use of machine learning to quantitatively analyze structure–activity relationships, thus broadening AI's impact on scientific discovery.

6 Conclusions and perspectives

Automated extraction of unstructured chemical data, facilitated by optical character recognition and large language models (LLMs), lays the groundwork for robust data-driven approaches. Automated robotic platforms streamline experimentation, enabling real-time decision-making and facilitating closed-loop optimizations. Active learning algorithms optimize experiment selection based on accumulated data to minimize trial numbers. Interpretable machine learning models disclose underlying structure–property relationships, providing critical insights for rational catalyst design.

Despite these advances, challenges persist. Information extraction needs to evolve to handle diverse unstructured data formats reliably. Current technologies like image segmentation tools 225,226 are still advancing towards fully autonomous capabilities for extracting and analyzing raw chemical data from figures. Moreover, the integration of text and figure data demands enhanced anaphora resolution and inference capabilities to support detailed analyses. Future developments in multimodal AI, capable of processing text, images, video, and voice, will be crucial in this aspect.

LLMs have demonstrated potential in comprehending complex data and have been applied successfully in projects like the one-pot synthesis conditions of MOFs. Yet, the full scope of their capabilities, especially in formatting conditions for multi-step synthesis procedures, remains underexplored. The cost and operational speed of robotic systems also limits their widespread adoption in chemical laboratories, necessitating innovations in specialized post-synthesis processing and auto-sampling for diverse catalytic systems.

The variability in control interfaces across different laboratory equipment poses another challenge, limiting hardware transferability among research communities. Standardizing control languages or systems could enhance collaborative efforts. Although natural language is commonly used to instruct experiments, its ambiguity necessitates sophisticated mapping to specific robotic operations, a task where LLMs could play a transformative role if their reliability is proven in more complex scenarios.

Furthermore, the high-dimensional nature of catalysis design and the chemical consumption in high-throughput processes suggest that automated platforms should be capable of managing varied reaction scales, from small-scale synthesis and characterization to larger-scale production.

As machine learning approaches become more integrated into catalyst design, it is anticipated that they will address increasingly complex design problems. Incorporating scientific hypotheses into the discovery process requires an iterative approach, where hypotheses are generated and modified, and data are queried for validation. AI agents, 227 e.g. , ChemCrow 228 equipped with tools for automated experimentation, information retrieval, and machine learning, show promise in bridging these capabilities to create a self-evolving, intelligent system.

Although human feedback should ideally not exist in the process, it can be used for safety checks or as alternative solutions if any of the functions ( e.g. , automated experimentation) are missing in the toolset, as demonstrated by Omar M. Yaghi et al. 47 In the iteration, the AI agents should be instructed to generate or modify hypotheses together with their validation procedures within the toolset. Later the toolset can be utilized to give feedback to the AI agents for further improvement of the hypotheses via LLMs directly or Bayesian inference.

In conclusion, the last decade's advances have shifted the paradigm from traditional methods to a more efficient, systematic approach to experimental design in catalyst research. The integration of LLMs and AI agents promises to further enhance the capability, flexibility, and efficiency of these systems, paving the way for a future where intelligent systems can autonomously explore vast chemical spaces and contribute to scientific discovery in unprecedented ways.

Data availability

Author contributions, conflicts of interest, acknowledgements, notes and references.

  • J. Gasteiger, Chemistry in times of artificial intelligence, Chemphyschem , 2020, 21 , 2233–2242  CrossRef   CAS   PubMed .
  • T. Mueller, A. G. Kusne and R. Ramprasad, Machine learning in materials science: Recent progress and emerging applications, Rev. Comput. Chem. , 2016, 29 , 186–273  CAS .
  • P. Raccuglia, K. C. Elbert, P. D. Adler, C. Falk, M. B. Wenny, A. Mollo, M. Zeller, S. A. Friedler, J. Schrier and A. J. Norquist, Machine-learning-assisted materials discovery using failed experiments, Nature , 2016, 533 , 73–76  CrossRef   CAS   PubMed .
  • B. Burger, P. M. Maffettone, V. V. Gusev, C. M. Aitchison, Y. Bai, X. Wang, X. Li, B. M. Alston, B. Li, R. Clowes, N. Rankin, B. Harris, R. S. Sprick and A. I. Cooper, A mobile robotic chemist, Nature , 2020, 583 , 237–241  CrossRef   CAS   PubMed .
  • Q. Zhu, F. Zhang, Y. Huang, H. Xiao, L. Zhao, X. Zhang, T. Song, X. Tang, X. Li, G. He, B. Chong, J. Zhou, Y. Zhang, B. Zhang, J. Cao, M. Luo, S. Wang, G. Ye, W. Zhang, X. Chen, S. Cong, D. Zhou, H. Li, J. Li, G. Zou, W. Shang, J. Jiang and Y. Luo, An all-round AI-Chemist with a scientific mind, Natl. Sci. Rev. , 2022, 9 , nwac190  CrossRef   CAS   PubMed .
  • N. J. Szymanski, B. Rendy, Y. Fei, R. E. Kumar, T. He, D. Milsted, M. J. McDermott, M. Gallant, E. D. Cubuk, A. Merchant, H. Kim, A. Jain, C. J. Bartel, K. Persson, Y. Zeng and G. Ceder, An autonomous laboratory for the accelerated synthesis of novel materials, Nature , 2023, 624 , 86–91  CrossRef   CAS   PubMed .
  • K. Rajan, H. O. Brinkhaus, A. Zielesny and C. Steinbeck, A review of optical chemical structure recognition tools, J. Cheminf. , 2020, 12 , 60  CAS .
  • K. T. Mukaddem, E. J. Beard, B. Yildirim and J. M. Cole, ImageDataExtractor: A Tool To Extract and Quantify Data from Microscopy Images, J. Chem. Inf. Model. , 2020, 60 , 2492–2509  CrossRef   CAS   PubMed .
  • E. J. Beard and J. M. Cole, ChemSchematicResolver: a toolkit to decode 2D chemical diagrams with labels and R-groups into annotated chemical named entities, J. Chem. Inf. Model. , 2020, 60 , 2059–2072  CrossRef   CAS   PubMed .
  • F. Musazade, N. Jamalova and J. Hasanov, Review of techniques and models used in optical chemical structure recognition in images and scanned documents, J. Cheminf. , 2022, 14 , 61  Search PubMed .
  • I. Beltagy, K. Lo and A. Cohan, SciBERT: A Pretrained Language Model for Scientific Text, arXiv , 2019, preprint, arXiv:1903.10676,  DOI: 10.48550/arXiv.1903.10676 .
  • A. Kumar, S. Ganesh, D. Gupta and H. Kodamana, A text mining framework for screening catalysts and critical process parameters from scientific literature - A study on Hydrogen production from alcohol, Chem. Eng. Res. Des. , 2022, 184 , 90–102  CrossRef   CAS .
  • L. Wang, Y. Gao, X. Chen, W. Cui, Y. Zhou, X. Luo, S. Xu, Y. Du and B. Wang, A corpus of CO2 electrocatalytic reduction process extracted from the scientific literature, Sci. Data , 2023, 10 , 175  CrossRef   CAS   PubMed .
  • Y. Gao, L. Wang, X. Chen, Y. Du and B. Wang, Revisiting Electrocatalyst Design by a Knowledge Graph of Cu-Based Catalysts for CO2 Reduction, ACS Catal. , 2023, 13 , 8525–8534  CrossRef   CAS .
  • K. M. Jablonka, P. Schwaller, A. Ortega-Guerrero and B. Smit, Is GPT-3 all you need for low-data discovery in chemistry?, ChemRxiv , 2023, preprint,  DOI: 10.26434/chemrxiv-2023-fw8n4 .
  • Z. Zheng, O. Zhang, C. Borgs, J. T. Chayes and O. M. Yaghi, ChatGPT Chemistry Assistant for Text Mining and the Prediction of MOF Synthesis, J. Am. Chem. Soc. , 2023, 145 , 18048–18062  CrossRef   CAS   PubMed .
  • S. Steiner, J. Wolf, S. Glatzel, A. Andreou, J. M. Granda, G. Keenan, T. Hinkley, G. Aragon-Camarasa, P. J. Kitson, D. Angelone and L. Cronin, Organic synthesis in a modular robotic system driven by a chemical programming language, Science , 2019, 363 , eaav2211  CrossRef   CAS   PubMed .
  • S. H. M. Mehr, M. Craven, A. I. Leonov, G. Keenan and L. Cronin, A universal system for digitization and automatic execution of the chemical synthesis literature, Science , 2020, 370 , 101–108  CrossRef   CAS   PubMed .
  • S. Rohrbach, M. Šiaučiulis, G. Chisholm, P.-A. Pirvan, M. Saleeb, S. H. M. Mehr, E. Trushina, A. I. Leonov, G. Keenan, A. Khan, A. Hammer and L. Cronin, Digitization and validation of a chemical synthesis literature database in the ChemPU, Science , 2022, 377 , 172–180  CrossRef   CAS   PubMed .
  • A. S. Nugraha, G. Lambard, J. Na, M. S. A. Hossain, T. Asahi, W. Chaikittisilp and Y. Yamauchi, Mesoporous trimetallic PtPdAu alloy films toward enhanced electrocatalytic activity in methanol oxidation: unexpected chemical compositions discovered by Bayesian optimization, J. Mater. Chem. A , 2020, 8 , 13532–13540  RSC .
  • A. E. Gongora, B. Xu, W. Perry, C. Okoye, P. Riley, K. G. Reyes, E. F. Morgan and K. A. Brown, A Bayesian experimental autonomous researcher for mechanical design, Sci. Adv. , 2020, 6 , eaaz1708  CrossRef   PubMed .
  • J. K. Pedersen, C. M. Clausen, O. A. Krysiak, B. Xiao, T. A. A. Batchelor, T. Löffler, V. A. Mints, L. Banko, M. Arenz, A. Savan, W. Schuhmann, A. Ludwig and J. Rossmeisl, Bayesian Optimization of High-Entropy Alloy Compositions for Electrocatalytic Oxygen Reduction, Angew. Chem., Int. Ed. , 2021, 60 , 24144–24152  CrossRef   CAS   PubMed .
  • G. O. Kayode, A. F. Hill and M. M. Montemore, Bayesian optimization of single-atom alloys and other bimetallics: efficient screening for alkane transformations, CO2 reduction, and hydrogen evolution, J. Mater. Chem. A , 2023, 11 , 19128–19137  RSC .
  • X. Wang, Y. Huang, X. Xie, Y. Liu, Z. Huo, M. Lin, H. Xin and R. Tong, Bayesian-optimization-assisted discovery of stereoselective aluminum complexes for ring-opening polymerization of racemic lactide, Nat. Commun. , 2023, 14 , 3647  CrossRef   CAS   PubMed .
  • Y. Okazaki, Y. Fujita, H. Murata, N. Masuyama, Y. Nojima, H. Ikeno, S. Yagi and I. Yamada, Composition-Designed Multielement Perovskite Oxides for Oxygen Evolution Catalysis, Chem. Mater. , 2022, 34 , 10973–10981  CrossRef   CAS .
  • Y. Zhang, T. C. Peck, G. K. Reddy, D. Banerjee, H. Jia, C. A. Roberts and C. Ling, Descriptor-Free Design of Multicomponent Catalysts, ACS Catal. , 2022, 12 , 10562–10571  CrossRef   CAS .
  • V. A. Mints, J. K. Pedersen, A. Bagger, J. Quinson, A. S. Anker, K. M. Ø. Jensen, J. Rossmeisl and M. Arenz, Exploring the Composition Space of High-Entropy Alloy Nanoparticles for the Electrocatalytic H2/CO Oxidation with Bayesian Optimization, ACS Catal. , 2022, 12 , 11263–11271  CrossRef   CAS .
  • Q. Liang, A. E. Gongora, Z. Ren, A. Tiihonen, Z. Liu, S. Sun, J. R. Deneault, D. Bash, F. Mekki-Berrada, S. A. Khan, K. Hippalgaonkar, B. Maruyama, K. A. Brown, J. Fisher Iii and T. Buonassisi, Benchmarking the performance of Bayesian optimization across multiple experimental materials science domains, npj Comput. Mater. , 2021, 7 , 188  CrossRef .
  • Y. Ureel, M. R. Dobbelaere, O. Akin, R. J. Varghese, C. G. Pernalete, J. W. Thybaut and K. M. Van Geem, Active learning-based exploration of the catalytic pyrolysis of plastic waste, Fuel , 2022, 328 , 125340  CrossRef   CAS .
  • J. Noh, S. Back, J. Kim and Y. Jung, Active learning with non-ab initio input features toward efficient CO2 reduction catalysts, Chem. Sci. , 2018, 9 , 5152–5159  RSC .
  • K. Tran and Z. W. Ulissi, Active learning across intermetallics to guide discovery of electrocatalysts for CO2 reduction and H2 evolution, Nat. Catal. , 2018, 1 , 696–703  CrossRef   CAS .
  • M. Zhong, K. Tran, Y. Min, C. Wang, Z. Wang, C.-T. Dinh, P. De Luna, Z. Yu, A. S. Rasouli, P. Brodersen, S. Sun, O. Voznyy, C.-S. Tan, M. Askerka, F. Che, M. Liu, A. Seifitokaldani, Y. Pang, S.-C. Lo, A. Ip, Z. Ulissi and E. H. Sargent, Accelerated discovery of CO2 electrocatalysts using active machine learning, Nature , 2020, 581 , 178–183  CrossRef   CAS   PubMed .
  • M. Kim, Y. Kim, M. Y. Ha, E. Shin, S. J. Kwak, M. Park, I.-D. Kim, W.-B. Jung, W. B. Lee, Y. Kim and H.-T. Jung, Exploring Optimal Water Splitting Bifunctional Alloy Catalyst by Pareto Active Learning, Adv. Mater. , 2023, 35 , 2211497  CrossRef   CAS   PubMed .
  • M. Kim, M. Y. Ha, W.-B. Jung, J. Yoon, E. Shin, I.-d. Kim, W. B. Lee, Y. Kim and H.-t. Jung, Searching for an Optimal Multi-Metallic Alloy Catalyst by Active Learning Combined with Experiments, Adv. Mater. , 2022, 34 , 2108900  CrossRef   CAS   PubMed .
  • V. Tshitoyan, J. Dagdelen, L. Weston, A. Dunn, Z. Rong, O. Kononova, K. A. Persson, G. Ceder and A. Jain, Unsupervised word embeddings capture latent knowledge from materials science literature, Nature , 2019, 571 , 95–98  CrossRef   CAS   PubMed .
  • M. Erdem Günay and R. Yıldırım, Recent advances in knowledge discovery for heterogeneous catalysis using machine learning, Catal. Rev. , 2021, 63 , 120–164  CrossRef .
  • T. Toyao, Z. Maeno, S. Takakusagi, T. Kamachi, I. Takigawa and K.-i. Shimizu, Machine Learning for Catalysis Informatics: Recent Applications and Prospects, ACS Catal. , 2019, 10 , 2260–2297  CrossRef .
  • T. Mou, H. S. Pillai, S. Wang, M. Wan, X. Han, N. M. Schweitzer, F. Che and H. Xin, Bridging the complexity gap in computational heterogeneous catalysis with machine learning, Nat. Catal. , 2023, 6 , 122–136  CrossRef .
  • A. J. Medford, M. R. Kunz, S. M. Ewing, T. Borders and R. Fushimi, Extracting Knowledge from Data through Catalysis Informatics, ACS Catal. , 2018, 8 , 7403–7429  CrossRef   CAS .
  • M. Abolhasani and E. Kumacheva, The rise of self-driving labs in chemical and materials sciences, Nat. Synth. , 2023, 2 , 483–492  CrossRef .
  • W. X. Zhao, K. Zhou, J. Li, T. Tang, X. Wang, Y. Hou, Y. Min, B. Zhang, J. Zhang, Z. Dong, Y. Du, C. Yang, Y. Chen, Z. Chen, J. Jiang, R. Ren, Y. Li, X. Tang, Z. Liu, P. Liu, J.-Y. Nie and J.-R. Wen, A Survey of Large Language Models, arXiv , 2023, preprint, arXiv:2303.18223,  DOI: 10.48550/arXiv.2303.18223 .
  • N. Yoshikawa, M. Skreta, K. Darvish, S. Arellano-Rubach, Z. Ji, L. Bjørn Kristensen, A. Z. Li, Y. Zhao, H. Xu, A. Kuramshin, A. Aspuru-Guzik, F. Shkurti and A. Garg, Large language models for chemistry robotics, Auton. Robots , 2023, 47 , 1056–1086  CrossRef .
  • T. Inagaki, A. Kato, K. Takahashi, H. Ozaki and G. N. Kanda, LLMs can generate robotic scripts from goal-oriented instructions in biological laboratory automation, arXiv , 2023, preprint, arXiv:2304.10267,  DOI: 10.48550/arXiv.2304.10267 .
  • C. Singh, J. X. Morris, J. Aneja, A. M. Rush and J. Gao, Explaining Patterns in Data with Language Models via Interpretable Autoprompting, arXiv , 2022, preprint, arXiv:2210.01848,  DOI: 10.48550/arXiv.2210.01848 .
  • M. Caldas Ramos, S. S. Michtavy, M. D. Porosoff and A. D. White, Bayesian Optimization of Catalysts With In-context Learning, arXiv , 2023, preprint, arXiv:2304.05341,  DOI: 10.48550/arXiv.2304.05341 .
  • D. A. Boiko, R. MacKnight, B. Kline and G. Gomes, Autonomous chemical research with large language models, Nature , 2023, 624 , 570–578  CrossRef   CAS   PubMed .
  • Z. Zheng, Z. Rong, N. Rampal, C. Borgs, J. T. Chayes and O. M. Yaghi, A GPT-4 Reticular Chemist for Guiding MOF Discovery, Angew. Chem., Int. Ed. , 2023, 62 , e202311983  CrossRef   CAS   PubMed .
  • D. S. Salley, G. A. Keenan, D.-L. Long, N. L. Bell and L. Cronin, A modular programmable inorganic cluster discovery robot for the discovery and synthesis of polyoxometalates, ACS Cent. Sci. , 2020, 6 , 1587–1593  CrossRef   CAS   PubMed .
  • J. S. Manzano, W. Hou, S. S. Zalesskiy, P. Frei, H. Wang, P. J. Kitson and L. Cronin, An autonomous portable platform for universal chemical synthesis, Nat. Chem. , 2022, 14 , 1311–1318  CrossRef   CAS   PubMed .
  • A. R. Frisbee, M. H. Nantz, G. W. Kramer and P. L. Fuchs, Laboratory automation. 1: Syntheses via vinyl sulfones. 14. Robotic orchestration of organic reactions: Yield optimization via an automated system with operator-specified reaction sequences, J. Am. Chem. Soc. , 1984, 106 , 7143–7145  CrossRef   CAS .
  • B. P. MacLeod, F. G. L. Parlane, T. D. Morrissey, F. Hse, L. M. Roch, K. E. Dettelbach, R. Moreira, L. P. E. Yunker, M. B. Rooney, J. R. Deeth, V. Lai, G. J. Ng, H. Situ, R. H. Zhang, M. S. Elliott, T. H. Haley, D. J. Dvorak, A. Aspuru-Guzik, J. E. Hein and C. P. Berlinguette, Self-driving laboratory for accelerated discovery of thin-film materials, Sci. Adv. , 2020, 6 , eaaz8867  CrossRef   CAS   PubMed .
  • F. Emmert-Streib, Z. Yang, H. Feng, S. Tripathi and M. Dehmer, An Introductory Review of Deep Learning for Prediction Models With Big Data, Front. Artif. Intell. , 2020, 3  DOI: 10.3389/frai.2020.00004 .
  • S. Lundberg and S.-I. Lee, A Unified Approach to Interpreting Model Predictions, arXiv , 2017, preprint, arXiv:1705.07874,  DOI: 10.48550/arXiv.1705.07874 .
  • Reaxys, https://www.reaxys.com, accessed May 12, 2021.
  • SciFinder, https://scifinder.cas.org, accessed May 12, 2021.
  • X. Zeng, H. Xiang, L. Yu, J. Wang, K. Li, R. Nussinov and F. Cheng, Accurate prediction of molecular properties and drug targets using a self-supervised image representation learning framework, Nat. Mach. Intell. , 2022, 4 , 1004–1016  CrossRef .
  • R. Rozas and H. Fernandez, Automatic processing of graphics for image databases in science, J. Chem. Inf. Comput. Sci. , 1990, 30 , 7–12  CrossRef   CAS .
  • S. S. Bukhari, Z. Iftikhar and A. Dengel, Chemical structure recognition (CSR) system: automatic analysis of 2D chemical structures in document images presented in part at the 2019 International Conference on Document Analysis and Recognition (ICDAR) , 2019  Search PubMed .
  • Y. Wang, T. Zhang and X. Yu, A component-detection-based approach for interpreting off-line handwritten chemical cyclic compound structures presented in part at the 2021 IEEE International Conference on Engineering, Technology & Education (TALE) , 2021  Search PubMed .
  • J. R. McDaniel and J. R. Balmuth, Kekule: OCR-optical chemical (structure) recognition, J. Chem. Inf. Comput. Sci. , 1992, 32 , 373–378  CrossRef   CAS .
  • N. M. Sadawi, A. P. Sexton and V. Sorge, Chemical structure recognition: a rule-based approach presented in part at the Document Recognition and Retrieval XIX , 2012  Search PubMed .
  • A. Fujiyoshi, K. Nakagawa and M. Suzuki, Robust method of segmentation and recognition of chemical structure images in cheminfty presented in part at the Pre-proceedings of the 9th IAPR international workshop on graphics recognition , GREC, 2011  Search PubMed .
  • C. Hong, X. Du and L. Zhang, Research on chemical expression images recognition presented in part at the 2015 Joint International Mechanical, Electronic and Information Technology Conference (JIMET-15) , 2015  Search PubMed .
  • J. Park, G. R. Rosania, K. A. Shedden, M. Nguyen, N. Lyu and K. Saitou, Automated extraction of chemical structure information from digital raster images, Chem. Cent. J. , 2009, 3 , 4  CrossRef   PubMed .
  • R. Casey, S. Boyer, P. Healey, A. Miller, B. Oudot and K. Zilles, Optical recognition of chemical graphics presented in part at the Proceedings of 2nd International Conference on Document Analysis and Recognition (ICDAR'93) , 1993  Search PubMed .
  • P. Ibison, M. Jacquot, F. Kam, A. Neville, R. W. Simpson, C. Tonnelier, T. Venczel and A. P. Johnson, Chemical literature data extraction: the CLiDE Project, J. Chem. Inf. Comput. Sci. , 1993, 33 , 338–344  CrossRef   CAS .
  • A. T. Valko and A. P. Johnson, CLiDE Pro: the latest generation of CLiDE, a tool for optical chemical structure recognition, J. Chem. Inf. Model. , 2009, 49 , 780–787  CrossRef   CAS   PubMed .
  • I. V. Filippov and M. C. Nicklaus, Optical structure recognition software to recover chemical information: OSRA, an open source solution, J. Chem. Inf. Model. , 2009, 49 , 740–743  CrossRef   CAS   PubMed .
  • V. Smolov, F. Zentsev and M. Rybalkin, Imago: Open-Source Toolkit for 2D Chemical Structure Image Recognition presented in part at the TREC , 2011  Search PubMed .
  • P. Frasconi, F. Gabbrielli, M. Lippi and S. Marinai, Markov logic networks for optical chemical structure recognition, J. Chem. Inf. Model. , 2014, 54 , 2380–2390  CrossRef   CAS   PubMed .
  • J. Staker, K. Marshall, R. Abel and C. M. McQuaw, Molecular Structure Extraction from Documents Using Deep Learning, J. Chem. Inf. Model. , 2019, 59 , 1017–1029  CrossRef   CAS   PubMed .
  • K. Rajan, A. Zielesny and C. Steinbeck, DECIMER: towards deep learning for chemical image recognition, J. Cheminf. , 2020, 12 , 65  CAS .
  • K. Rajan, H. O. Brinkhaus, M. Sorokina, A. Zielesny and C. Steinbeck, DECIMER-Segmentation: Automated extraction of chemical structure depictions from scientific literature, J. Cheminf. , 2021, 13 , 20  CAS .
  • K. Rajan, H. O. Brinkhaus, M. I. Agea, A. Zielesny and C. Steinbeck, DECIMER. ai-An open platform for automated optical chemical structure identification, segmentation and recognition in scientific publications, Nat. Commun. , 2023, 14 , 5045  CrossRef   CAS   PubMed .
  • Y. Xu, J. Xiao, C.-H. Chou, J. Zhang, J. Zhu, Q. Hu, H. Li, N. Han, B. Liu, S. Zhang, J. Han, Z. Zhang, S. Zhang, W. Zhang, L. Lai and J. Pei, MolMiner: You only look once for chemical structure recognition, J. Chem. Inf. Model. , 2022, 62 , 5321–5328  CrossRef   CAS   PubMed .
  • Y. Qian, J. Guo, Z. Tu, Z. Li, C. W. Coley and R. Barzilay, MolScribe: Robust Molecular Structure Recognition with Image-to-Graph Generation, J. Chem. Inf. Model. , 2023, 63 , 1925–1934  CrossRef   CAS   PubMed .
  • W. Hemati and A. Mehler, LSTMVoter: chemical named entity recognition using a conglomerate of sequence labeling tools, J. Cheminf. , 2019, 11 , 3  Search PubMed .
  • D. Mahendran, C. Tang and B. T. McInnes, Graph Convolutional Networks for Chemical Relation Extraction presented in part at the Companion Proceedings of the Web Conference 2022 , Virtual Event, Lyon, France, 2022  Search PubMed .
  • M. Suvarna, A. C. Vaucher, S. Mitchell, T. Laino and J. Pérez-Ramírez, Language models and protocol standardization guidelines for accelerating synthesis planning in heterogeneous catalysis, Nat. Commun. , 2023, 14 , 7964  CrossRef   CAS   PubMed .
  • M. C. Swain and J. M. Cole, ChemDataExtractor: A Toolkit for Automated Extraction of Chemical Information from the Scientific Literature, J. Chem. Inf. Model. , 2016, 56 , 1894–1904  CrossRef   CAS   PubMed .
  • J. Mavračić, C. J. Court, T. Isazawa, S. R. Elliott and J. M. Cole, ChemDataExtractor 2.0: Autopopulated Ontologies for Materials Science, J. Chem. Inf. Model. , 2021, 61 , 4280–4289  CrossRef   PubMed .
  • J. Guo, A. S. Ibanez-Lopez, H. Gao, V. Quach, C. W. Coley, K. F. Jensen and R. Barzilay, Automated chemical reaction extraction from scientific literature, J. Chem. Inf. Model. , 2021, 62 , 2035–2045  CrossRef   PubMed .
  • R. B. Merrifield and J. M. Stewart, Automated peptide synthesis, Nature , 1965, 207 , 522–523  CrossRef   CAS   PubMed .
  • C. Porte, W. Debreuille, F. Draskovic and A. Delacroix, Automation and optimization by simplex methods of 6-chlorohexanol synthesis, Process Control Qual. , 1996, 4 , 111–122  Search PubMed .
  • R. W. Wagner, F. Li, H. Du and J. S. Lindsey, Investigation of cocatalysis conditions using an automated microscale multireactor workstation: Synthesis of m eso-tetramesitylporphyrin, Org. Process Res. Dev. , 1999, 3 , 28–37  CrossRef   CAS .
  • J. S. Lindsey, A retrospective on the automation of laboratory synthetic chemistry, Chemom. Intell. Lab. Syst. , 1992, 17 , 15–45  CrossRef   CAS .
  • R. B. Merrifield, J. M. Stewart and N. Jernberg, Instrument for automated synthesis of peptides, Anal. Chem. , 1966, 38 , 1905–1914  CrossRef   CAS   PubMed .
  • G. Alvarado-Urbina, G. M. Sathe, W.-C. Liu, M. F. Gillen, P. D. Duck, R. Bender and K. K. Ogilvie, Automated synthesis of gene fragments, Science , 1981, 214 , 270–274  CrossRef   CAS   PubMed .
  • M. Legrand and A. Foucard, Automation on the laboratory bench, J. Chem. Educ. , 1978, 55 , 767  CrossRef   CAS .
  • J. Bajorath, Integration of virtual and high-throughput screening, Nat. Rev. Drug Discovery , 2002, 1 , 882–894  CrossRef   CAS   PubMed .
  • R. L. Martin, C. M. Simon, B. Smit and M. Haranczyk, In silico design of porous polymer networks: high-throughput screening for methane storage materials, J. Am. Chem. Soc. , 2014, 136 , 5006–5022  CrossRef   CAS   PubMed .
  • R. Macarron, M. N. Banks, D. Bojanic, D. J. Burns, D. A. Cirovic, T. Garyantes, D. V. S. Green, R. P. Hertzberg, W. P. Janzen, J. W. Paslay, U. Schopfer and G. Sitta Sittampalam, Impact of high-throughput screening in biomedical research, Nat. Rev. Drug Discovery , 2011, 10 , 188–195  CrossRef   CAS   PubMed .
  • S. M. Senkan, High-throughput screening of solid-state catalyst libraries, Nature , 1998, 394 , 350–353  CrossRef   CAS .
  • X. D. Xiang, X. Sun, G. Briceno, Y. Lou, K.-A. Wang, H. Chang, W. G. Wallace-Freedman, S.-W. Chen and P. G. Schultz, A combinatorial approach to materials discovery, Science , 1995, 268 , 1738–1740  CrossRef   CAS   PubMed .
  • W. F. Maier, K. Stoewe and S. Sieg, Combinatorial and high-throughput materials science, Angew. Chem., Int. Ed. , 2007, 46 , 6016–6067  CrossRef   CAS   PubMed .
  • L. Cheng, R. S. Assary, X. Qu, A. Jain, S. P. Ong, N. N. Rajput, K. Persson and L. A. Curtiss, Accelerating electrolyte discovery for energy storage with high-throughput screening, J. Phys. Chem. Lett. , 2015, 6 , 283–291  CrossRef   CAS   PubMed .
  • J. S. Smith, B. Nebgen, N. Lubbers, O. Isayev and A. E. Roitberg, Less is more: Sampling chemical space with active learning, J. Chem. Phys. , 2018, 148  DOI: 10.1063/1.5023802 .
  • Q. Zhu, Y. Huang, D. Zhou, L. Zhao, L. Guo, R. Yang, Z. Sun, M. Luo, F. Zhang, H. Xiao, X. Tang, X. Zhang, T. Song, X. Li, B. Chong, J. Zhou, Y. Zhang, B. Zhang, J. Cao, G. Zhang, S. Wang, G. Ye, W. Zhang, H. Zhao, S. Cong, H. Li, L.-L. Ling, Z. Zhang, W. Shang, J. Jiang and Y. Luo, Automated synthesis of oxygen-producing catalysts from Martian meteorites by a robotic AI chemist, Nat. Synth. , 2024, 3 , 319–328  CrossRef .
  • A. Adamo, R. L. Beingessner, M. Behnam, J. Chen, T. F. Jamison, K. F. Jensen, J.-C. M. Monbaliu, A. S. Myerson, E. M. Revalor, D. R. Snead, T. Stelzer, N. Weeranoppanant, S. Y. Wong and P. Zhang, On-demand continuous-flow production of pharmaceuticals in a compact, reconfigurable system, Science , 2016, 352 , 61–67  CrossRef   CAS   PubMed .
  • M. Baumann and I. R. Baxendale, The synthesis of active pharmaceutical ingredients (APIs) using continuous flow chemistry, Beilstein J. Org. Chem. , 2015, 11 , 1194–1219  CrossRef   CAS   PubMed .
  • J. Wegner, S. Ceylan and A. Kirschning, Flow chemistry–a key enabling technology for (multistep) organic synthesis, Adv. Synth. Catal. , 2012, 354 , 17–57  CrossRef   CAS .
  • S. Kobayashi, Flow “fine” synthesis: high yielding and selective organic synthesis by flow methods, Chem.–Asian J. , 2016, 11 , 425–436  CrossRef   CAS   PubMed .
  • M. E. Briggs, A. G. Slater, N. Lunt, S. Jiang, M. A. Little, R. L. Greenaway, T. Hasell, C. Battilocchio, S. V. Ley and A. I. Cooper, Dynamic flow synthesis of porous organic cages, Chem. Commun. , 2015, 51 , 17390–17393  RSC .
  • G. Jas and A. Kirschning, Continuous flow techniques in organic synthesis, Chem. - Eur. J. , 2003, 9 , 5708–5723  CrossRef   CAS   PubMed .
  • D. Angelone, A. J. S. Hammer, S. Rohrbach, S. Krambeck, J. M. Granda, J. Wolf, S. Zalesskiy, G. Chisholm and L. Cronin, Convergence of multiple synthetic paradigms in a universally programmable chemical synthesis machine, Nat. Chem. , 2021, 13 , 63–69  CrossRef   CAS   PubMed .
  • M. Xie, Y. Shen, W. Ma, D. Wei, B. Zhang, Z. Wang, Y. Wang, Q. Zhang, S. Xie, C. Wang and Y. Wang, Fast Screening for Copper-Based Bimetallic Electrocatalysts: Efficient Electrocatalytic Reduction of CO 2 to C 2+ Products on Magnesium-Modified Copper, Angew. Chem., Int. Ed. , 2022, 61 , e202213423  CrossRef   CAS   PubMed .
  • M. B. Plutschack, B. u. Pieber, K. Gilmore and P. H. Seeberger, The hitchhiker's guide to flow chemistry, Chem. Rev. , 2017, 117 , 11796–11893  CrossRef   CAS   PubMed .
  • A. R. Bogdan and Y. Wang, A high-throughput synthesis of 1, 2, 4-oxadiazole and 1, 2, 4-triazole libraries in a continuous flow reactor, RSC Adv. , 2015, 5 , 79264–79269  RSC .
  • S. E. Lohse, J. R. Eller, S. T. Sivapalan, M. R. Plews and C. J. Murphy, A simple millifluidic benchtop reactor system for the high-throughput synthesis and functionalization of gold nanoparticles with different sizes and shapes, ACS Nano , 2013, 7 , 4135–4150  CrossRef   CAS   PubMed .
  • E. J. Roberts, S. E. Habas, L. Wang, D. A. Ruddy, E. A. White, F. G. Baddour, M. B. Griffin, J. A. Schaidle, N. Malmstadt and R. L. Brutchey, High-throughput continuous flow synthesis of nickel nanoparticles for the catalytic hydrodeoxygenation of guaiacol, ACS Sustain. Chem. Eng. , 2017, 5 , 632–639  CrossRef   CAS .
  • M. Ago, S. Huan, M. Borghei, J. Raula, E. I. Kauppinen and O. J. Rojas, High-throughput synthesis of lignin particles (∼30 nm to ∼2 μm) via aerosol flow reactor: Size fractionation and utilization in pickering emulsions, ACS Appl. Mater. Interfaces , 2016, 8 , 23302–23310  CrossRef   CAS   PubMed .
  • E. M. Chan, C. Xu, A. W. Mao, G. Han, J. S. Owen, B. E. Cohen and D. J. Milliron, Reproducible, high-throughput synthesis of colloidal nanocrystals for optimization in multidimensional parameter space, Nano Lett. , 2010, 10 , 1874–1885  CrossRef   CAS   PubMed .
  • H. Long, T.-S. Chen, J. Song, S. Zhu and H.-C. Xu, Electrochemical aromatic C--H hydroxylation in continuous flow, Nat. Commun. , 2022, 13 , 3945  CrossRef   CAS   PubMed .
  • B. Winterson, T. Rennigholtz and T. Wirth, Flow electrochemistry: a safe tool for fluorine chemistry, Chem. Sci. , 2021, 12 , 9053–9059  RSC .
  • A.-C. Bédard, A. Adamo, K. C. Aroh, M. G. Russell, A. A. Bedermann, J. Torosian, B. Yue, K. F. Jensen and T. F. Jamison, Reconfigurable system for automated optimization of diverse chemical reactions, Science , 2018, 361 , 1220–1225  CrossRef   PubMed .
  • S. Chatterjee, M. Guidi, P. H. Seeberger and K. Gilmore, Automated radial synthesis of organic molecules, Nature , 2020, 579 , 379–384  CrossRef   CAS   PubMed .
  • N. Collins, D. Stout, J.-P. Lim, J. P. Malerich, J. D. White, P. B. Madrid, M. Latendresse, D. Krieger, J. Szeto and V.-A. Vu, Fully automated chemical synthesis: toward the universal synthesizer, Org. Process Res. Dev. , 2020, 24 , 2064–2077  CrossRef   CAS .
  • C. D. Scott, R. Labes, M. Depardieu, C. Battilocchio, M. G. Davidson, S. V. Ley, C. C. Wilson and K. Robertson, Integrated plug flow synthesis and crystallisation of pyrazinamide, React. Chem. Eng. , 2018, 3 , 631–634  RSC .
  • G. Niu, L. Zhang, A. Ruditskiy, L. Wang and Y. Xia, A droplet-reactor system capable of automation for the continuous and scalable production of noble-metal nanocrystals, Nano Lett. , 2018, 18 , 3879–3884  CrossRef   CAS   PubMed .
  • E. W. Yeap, D. Z. Ng, D. Lai, D. J. Ertl, S. Sharpe and S. A. Khan, Continuous flow droplet-based crystallization platform for producing spherical drug microparticles, Org. Process Res. Dev. , 2018, 23 , 93–101  CrossRef .
  • B. Rimez, J. Septavaux and B. Scheid, The coupling of in-flow reaction with continuous flow seedless tubular crystallization, React. Chem. Eng. , 2019, 4 , 516–522  RSC .
  • O. Okafor, K. Robertson, R. Goodridge and V. Sans, Continuous-flow crystallisation in 3D-printed compact devices, React. Chem. Eng. , 2019, 4 , 1682–1688  RSC .
  • C. W. Coley, D. A. Thomas III, J. A. M. Lummiss, J. N. Jaworski, C. P. Breen, V. Schultz, T. Hart, J. S. Fishman, L. Rogers, H. Gao, R. W. Hicklin, P. P. Plehiers, J. Byington, J. S. Piotti, W. H. Green, A. J. Hart, T. F. Jamison and K. F. Jensen, A robotic platform for flow synthesis of organic compounds informed by AI planning, Science , 2019, 365 , eaax1566  CrossRef   CAS   PubMed .
  • B. A. Koscher, R. B. Canty, M. A. McDonald, K. P. Greenman, C. J. McGill, C. L. Bilodeau, W. Jin, H. Wu, F. H. Vermeire, B. Jin, T. Hart, T. Kulesza, S.-C. Li, T. S. Jaakkola, R. Barzilay, R. Gómez-Bombarelli, W. H. Green and K. F. Jensen, Autonomous, multiproperty-driven molecular discovery: From predictions to measurements and back, Science , 2023, 382 , eadi1407  CrossRef   CAS   PubMed .
  • F. Stella, C. Della Santina and J. Hughes, How can LLMs transform the robotic design process?, Nat. Mach. Intell. , 2023, 1–4  Search PubMed .
  • S. Vemprala, R. Bonatti, A. Bucker and A. Kapoor, Chatgpt for robotics: Design principles and model abilities, Microsoft Auton. Syst. Robot. Res , 2023, 2 , 20  Search PubMed .
  • L. Wang, Y. Ling, Z. Yuan, M. Shridhar, C. Bao, Y. Qin, B. Wang, H. Xu and X. Wang, GenSim: Generating Robotic Simulation Tasks via Large Language Models, arXiv , 2023, preprint, arXiv:2310.01361,  DOI: 10.48550/arXiv.2310.01361 .
  • F. Olsson, A literature survey of active machine learning in the context of natural language processing, Report 11003154 (ISSN), Swedish Institute of Computer Science , Kista, Sweden, 2009  Search PubMed .
  • Y. Ureel, M. R. Dobbelaere, Y. Ouyang, K. De Ras, M. K. Sabbe, G. B. Marin and K. M. Van Geem, Active Machine Learning for Chemical Engineers: A Bright Future Lies Ahead, Engineering , 2023, 27 , 23–30  CrossRef   CAS .
  • X. Wang, Y. Jin, S. Schmitt and M. Olhofer, Recent Advances in Bayesian Optimization, ACM Comput. Surv. , 2023, 55 , 287  Search PubMed .
  • R. González Perea, E. Camacho Poyato, P. Montesinos and J. A. Rodríguez Díaz, Optimisation of water demand forecasting by artificial intelligence with short data sets, Biosyst. Eng. , 2019, 177 , 59–66  CrossRef .
  • J. M. Hernández-Lobato and R. P. Adams, Probabilistic Backpropagation for Scalable Learning of Bayesian Neural Networks presented in part at the International Conference on Machine Learning , 2015  Search PubMed .
  • A. Graves, Practical variational inference for neural networks presented in part at the Proceedings of the 24th International Conference on Neural Information Processing Systems , Granada, Spain, 2011  Search PubMed .
  • F. Häse, L. M. Roch, C. Kreisbeck and A. Aspuru-Guzik, Phoenics: A Bayesian Optimizer for Chemistry, ACS Cent. Sci. , 2018, 4 , 1134–1145  CrossRef   PubMed .
  • F. Häse, M. Aldeghi, R. J. Hickman, L. M. Roch and A. Aspuru-Guzik, Gryffin: An algorithm for Bayesian optimization of categorical variables informed by expert knowledge, Applied Physics Reviews , 2021, 8  DOI: 10.1063/5.0048164 .
  • J. Močkus, in Optimization Techniques IFIP Technical Conference: Novosibirsk, July 1–7, 1974 , ed. G. I. Marchuk, Springer Berlin Heidelberg, Berlin, Heidelberg, 1975, pp. 400–404,  DOI: 10.1007/978-3-662-38527-2_55 .
  • A. Zilinskas, Optimization of one-dimensional multimodal functions, J. R. Stat. Soc., C: Appl. Stat. , 1978, 27 , 367–375  Search PubMed .
  • H. J. Kushner, A New Method of Locating the Maximum Point of an Arbitrary Multipeak Curve in the Presence of Noise, J. Basic Eng. , 1964, 86 , 97–106  CrossRef .
  • P. Auer, Using confidence bounds for exploitation-exploration trade-offs, J. Mach. Learn. Res. , 2003, 3 , 397–422  Search PubMed .
  • J.-B. Masson, Counting biomolecules with Bayesian inference, Nat. Comput. Sci. , 2022, 2 , 74–75  CrossRef   PubMed .
  • L. Bassman Oftelie, P. Rajak, R. K. Kalia, A. Nakano, F. Sha, J. Sun, D. J. Singh, M. Aykol, P. Huck, K. Persson and P. Vashishta, Active learning for accelerated design of layered materials, npj Comput. Mater. , 2018, 4 , 74  CrossRef .
  • K. M. Jablonka, G. M. Jothiappan, S. Wang, B. Smit and B. Yoo, Bias free multiobjective active learning for materials design and discovery, Nat. Commun. , 2021, 12 , 2312  CrossRef   CAS   PubMed .
  • N. S. Eyke, W. H. Green and K. F. Jensen, Iterative experimental design based on active machine learning reduces the experimental burden associated with reaction screening, React. Chem. Eng. , 2020, 5 , 1963–1972  RSC .
  • S. Viet Johansson, H. Gummesson Svensson, E. Bjerrum, A. Schliep, M. Haghir Chehreghani, C. Tyrchan and O. Engkvist, Using Active Learning to Develop Machine Learning Models for Reaction Yield Prediction, Mol. Inf. , 2022, 41 , e2200043  CrossRef   PubMed .
  • D. Reker, P. Schneider, G. Schneider and J. B. Brown, Active learning for computational chemogenomics, Future Med. Chem. , 2017, 9 , 381–402  CrossRef   CAS   PubMed .
  • D. Reker and G. Schneider, Active-learning strategies in computer-assisted drug discovery, Drug Discovery Today , 2015, 20 , 458–465  CrossRef   PubMed .
  • F. Douak, F. Melgani and N. Benoudjit, Kernel ridge regression with active learning for wind speed prediction, Appl. Energy , 2013, 103 , 328–340  CrossRef .
  • M. Sharma and M. Bilgic, Evidence-based uncertainty sampling for active learning, Data Min. Knowl. Discov. , 2017, 31 , 164–202  CrossRef .
  • F. Douak, N. Benoudjit and F. Melgani, A two-stage regression approach for spectroscopic quantitative analysis, Chemom. Intell. Lab. Syst. , 2011, 109 , 34–41  CrossRef   CAS .
  • A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek, S. Cholia, D. Gunter, D. Skinner, G. Ceder and K. A. Persson, Commentary: The Materials Project: A materials genome approach to accelerating materials innovation, APL Mater. , 2013, 1 , 011002  CrossRef .
  • T. Guo, K. Guo, B. Nan, Z. Liang, Z. Guo, N. Chawla, O. Wiest and X. Zhang, What can large language models do in chemistry? a comprehensive benchmark on eight tasks, arXiv , 2023, preprint, arXiv:2305.18365,  DOI: 10.48550/arXiv.2305.18365 .
  • T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry and A. Askell, Language models are few-shot learners, Adv. Neural Inf. Process Syst. , 2020, 33 , 1877–1901  Search PubMed .
  • A. E. Blanchard, J. Gounley, D. Bhowmik, M. Chandra Shekar, I. Lyngaas, S. Gao, J. Yin, A. Tsaris, F. Wang and J. Glaser, Language models for the prediction of SARS-CoV-2 inhibitors, Int. J. High Perform. Comput. , 2022, 36 , 587–602  CrossRef   PubMed .
  • C. Xu, Y. Wang and A. Barati Farimani, TransPolymer: a Transformer-based language model for polymer property predictions, npj Comput. Mater. , 2023, 9 , 64  CrossRef .
  • K. M. Jablonka, P. Schwaller, A. Ortega-Guerrero and B. Smit, Leveraging large language models for predictive chemistry, Nat. Mach. Intell. , 2024, 6 , 161–169  CrossRef .
  • J. Ross, B. Belgodere, V. Chenthamarakshan, I. Padhi, Y. Mroueh and P. Das, Large-scale chemical language representations capture molecular structure and properties, Nat. Mach. Intell. , 2022, 4 , 1256–1264  CrossRef .
  • Z. Yang, Y. Wang and L. Zhang, AI becomes a masterbrain scientist, bioRxiv , 2023, preprint,  DOI: 10.1101/2023.04.19.537579 .
  • J. Kim, D. Kang, S. Kim and H. W. Jang, Catalyze Materials Science with Machine Learning, ACS Mater. Lett. , 2021, 3 , 1151–1171  CrossRef   CAS .
  • S. Singh and R. B. Sunoj, Molecular Machine Learning for Chemical Catalysis: Prospects and Challenges, Acc. Chem. Res. , 2023, 56 , 402–412  CrossRef   CAS   PubMed .
  • A. G. Maldonado and G. Rothenberg, Predictive modeling in homogeneous catalysis: a tutorial, Chem. Soc. Rev. , 2010, 39 , 1891–1902  RSC .
  • B. Kim, S. Lee and J. Kim, Inverse design of porous materials using artificial neural networks, Sci. Adv. , 2020, 6 , eaax9324  CrossRef   CAS   PubMed .
  • D. A. Carr, M. Lach-hab, S. Yang, I. I. Vaisman and E. Blaisten-Barojas, Machine learning approach for structure-based zeolite classification, Microporous Mesoporous Mater. , 2009, 117 , 339–349  CrossRef   CAS .
  • A. Jain and T. Bligaard, Atomic-position independent descriptor for machine learning of material properties, Phys. Rev. B , 2018, 98 , 214112  CrossRef .
  • K. Li and D. Xue, Estimation of Electronegativity Values of Elements in Different Valence States, J. Phys. Chem. A , 2006, 110 , 11332–11337  CrossRef   CAS   PubMed .
  • Z. Wu, B. Ramsundar, E. N. Feinberg, J. Gomes, C. Geniesse, A. S. Pappu, K. Leswing and V. Pande, MoleculeNet: a benchmark for molecular machine learning, Chem. Sci. , 2018, 9 , 513–530  RSC .
  • P. Schlexer Lamoureux, K. T. Winther, J. A. Garrido Torres, V. Streibel, M. Zhao, M. Bajdich, F. Abild-Pedersen and T. Bligaard, Machine Learning for Computational Heterogeneous Catalysis, ChemCatChem , 2019, 11 , 3581–3601  CrossRef   CAS .
  • J. L. Durant, B. A. Leland, D. R. Henry and J. G. Nourse, Reoptimization of MDL Keys for Use in Drug Discovery, J. Chem. Inf. Comput. Sci. , 2002, 42 , 1273–1280  CrossRef   CAS   PubMed .
  • D. Butina, Unsupervised Data Base Clustering Based on Daylight's Fingerprint and Tanimoto Similarity: A Fast and Automated Way To Cluster Small and Large Data Sets, J. Chem. Inf. Comput. Sci. , 1999, 39 , 747–750  CrossRef   CAS .
  • X. He, Y. Su, J. Zhu, N. Fang, Y. Chen, H. Liu, D. Zhou and C. Wang, Uncovering the influence of the modifier redox potential on CO2 reduction through combined data-driven machine learning and hypothesis-driven experimentation, J. Mater. Chem. A , 2023, 11 , 18106–18114  RSC .
  • D. Rogers and M. Hahn, Extended-Connectivity Fingerprints, J. Chem. Inf. Model. , 2010, 50 , 742–754  CrossRef   CAS   PubMed .
  • Y. Guo, X. He, Y. Su, Y. Dai, M. Xie, S. Yang, J. Chen, K. Wang, D. Zhou and C. Wang, Machine-Learning-Guided Discovery and Optimization of Additives in Preparing Cu Catalysts for CO2 Reduction, J. Am. Chem. Soc. , 2021, 143 , 5755–5762  CrossRef   CAS   PubMed .
  • J. P. Janet and H. J. Kulik, Resolving Transition Metal Chemical Space: Feature Selection for Machine Learning and Structure–Property Relationships, J. Phys. Chem. A , 2017, 121 , 8939–8954  CrossRef   CAS   PubMed .
  • C. B. Santiago, J.-Y. Guo and M. S. Sigman, Predictive and mechanistic multivariate linear regression models for reaction development, Chem. Sci. , 2018, 9 , 2398–2412  RSC .
  • X. Ma, Z. Li, L. E. K. Achenie and H. Xin, Machine-Learning-Augmented Chemisorption Model for CO2 Electroreduction Catalyst Screening, J. Phys. Chem. Lett. , 2015, 6 , 3528–3533  CrossRef   CAS   PubMed .
  • H. Mai, T. C. Le, D. Chen, D. A. Winkler and R. A. Caruso, Machine Learning for Electrocatalyst and Photocatalyst Design and Discovery, Chem. Rev. , 2022, 122 , 13478–13515  CrossRef   CAS   PubMed .
  • M. S. Sigman, K. C. Harper, E. N. Bess and A. Milo, The Development of Multidimensional Analysis Tools for Asymmetric Catalysis and Beyond, Acc. Chem. Res. , 2016, 49 , 1292–1301  CrossRef   CAS   PubMed .
  • J. P. Reid and M. S. Sigman, Comparing quantitative prediction methods for the discovery of small-molecule chiral catalysts, Nat. Rev. Chem , 2018, 2 , 290–305  CrossRef   CAS .
  • K. C. Harper and M. S. Sigman, Three-Dimensional Correlation of Steric and Electronic Free Energy Relationships Guides Asymmetric Propargylation, Science , 2011, 333 , 1875–1878  CrossRef   CAS   PubMed .
  • C. W. Yap, PaDEL-descriptor: an open source software to calculate molecular descriptors and fingerprints, J. Comput. Chem. , 2011, 32 , 1466–1474  CrossRef   CAS   PubMed .
  • Q. Yang, Y. Liu, J. Cheng, Y. Li, S. Liu, Y. Duan, L. Zhang and S. Luo, An Ensemble Structure and Physicochemical (SPOC) Descriptor for Machine-Learning Prediction of Chemical Reaction and Molecular Properties, Chemphyschem , 2022, 23 , e202200255  CrossRef   CAS   PubMed .
  • E. L. Willighagen, H. M. G. W. Denissen, R. Wehrens and L. M. C. Buydens, On the Use of 1H and 13C 1D NMR Spectra as QSPR Descriptors, J. Chem. Inf. Model. , 2006, 46 , 487–494  CrossRef   CAS   PubMed .
  • T. Jin, Q. Zhao, A. B. Schofield and B. M. Savoie, Machine learning models capable of chemical deduction for identifying reaction products, ChemRxiv , 2023, preprint,  DOI: 10.26434/chemrxiv-2023-l6lzp .
  • X. Wang, S. Jiang, W. Hu, S. Ye, T. Wang, F. Wu, L. Yang, X. Li, G. Zhang, X. Chen, J. Jiang and Y. Luo, Quantitatively Determining Surface–Adsorbate Properties from Vibrational Spectroscopy with Interpretable Machine Learning, J. Am. Chem. Soc. , 2022, 144 , 16069–16076  CrossRef   CAS   PubMed .
  • D. Loffreda, F. Delbecq, F. Vigné and P. Sautet, Fast Prediction of Selectivity in Heterogeneous Catalysis from Extended Brønsted–Evans–Polanyi Relations: A Theoretical Insight, Angew. Chem., Int. Ed. , 2009, 48 , 8978–8980  CrossRef   CAS   PubMed .
  • F. Wei and L. Zhuang, Unsupervised machine learning reveals eigen reactivity of metal surfaces, Sci. Bull. , 2024, 69 , 756–762  CrossRef   CAS   PubMed .
  • S. Ringe, The importance of a charge transfer descriptor for screening potential CO(2) reduction electrocatalysts, Nat. Commun. , 2023, 14 , 2598  CrossRef   CAS   PubMed .
  • W. T. Hong, R. E. Welsch and Y. Shao-Horn, Descriptors of Oxygen-Evolution Activity for Oxides: A Statistical Evaluation, J. Phys. Chem. C , 2015, 120 , 78–86  CrossRef .
  • C. Ren, S. Lu, Y. Wu, Y. Ouyang, Y. Zhang, Q. Li, C. Ling and J. Wang, A Universal Descriptor for Complicated Interfacial Effects on Electrochemical Reduction Reactions, J. Am. Chem. Soc. , 2022, 144 , 12874–12883  CrossRef   CAS   PubMed .
  • M. Andersen, A. J. Medford, J. K. Nørskov and K. Reuter, Scaling-Relation-Based Analysis of Bifunctional Catalysis: The Case for Homogeneous Bimetallic Alloys, ACS Catal. , 2017, 7 , 3960–3967  CrossRef   CAS .
  • J. Liu, W. Luo, L. Wang, J. Zhang, X.-Z. Fu and J.-L. Luo, Toward Excellence of Electrocatalyst Design by Emerging Descriptor-Oriented Machine Learning, Adv. Funct. Mater. , 2022, 32 , 2110748  CrossRef   CAS .
  • X. Lin, Y. Wang, X. Chang, S. Zhen, Z. J. Zhao and J. Gong, High-Throughput Screening of Electrocatalysts for Nitrogen Reduction Reactions Accelerated by Interpretable Intrinsic Descriptor, Angew Chem. Int. Ed. Engl. , 2023, 62 , e202300122  CrossRef   CAS   PubMed .
  • X. Wang, S. Ye, W. Hu, E. Sharman, R. Liu, Y. Liu, Y. Luo and J. Jiang, Electric Dipole Descriptor for Machine Learning Prediction of Catalyst Surface–Molecular Adsorbate Interactions, J. Am. Chem. Soc. , 2020, 142 , 7737–7743  CrossRef   CAS   PubMed .
  • L. H. Mou, T. Han, P. E. S. Smith, E. Sharman and J. Jiang, Machine Learning Descriptors for Data-Driven Catalysis Study, Advanced Science , 2023, 10 , e2301020  CrossRef   PubMed .
  • O. Isayev, C. Oses, C. Toher, E. Gossett, S. Curtarolo and A. Tropsha, Universal fragment descriptors for predicting properties of inorganic crystals, Nat. Commun. , 2017, 8 , 15679  CrossRef   CAS   PubMed .
  • C. W. Coley, W. Jin, L. Rogers, T. F. Jamison, T. S. Jaakkola, W. H. Green, R. Barzilay and K. F. Jensen, A graph-convolutional neural network model for the prediction of chemical reactivity, Chem. Sci. , 2019, 10 , 370–377  RSC .
  • L. P. Hammett, Linear free energy relationships in rate and equilibrium phenomena, Trans. Faraday Soc. , 1938, 34 , 156–165  RSC .
  • R. W. Taft Jr, Linear Free Energy Relationships from Rates of Esterification and Hydrolysis of Aliphatic and Ortho-substituted Benzoate Esters, J. Am. Chem. Soc. , 1952, 74 , 2729–2732  CrossRef .
  • J. M. Crawford, C. Kingston, F. D. Toste and M. S. Sigman, Data Science Meets Physical Organic Chemistry, Acc. Chem. Res. , 2021, 54 , 3136–3148  CrossRef   CAS   PubMed .
  • W. L. Williams, L. Zeng, T. Gensch, M. S. Sigman, A. G. Doyle and E. V. Anslyn, The Evolution of Data-Driven Modeling in Organic Chemistry, ACS Cent. Sci. , 2021, 7 , 1622–1637  CrossRef   CAS   PubMed .
  • J. P. Reid and M. S. Sigman, Holistic prediction of enantioselectivity in asymmetric catalysis, Nature , 2019, 571 , 343–348  CrossRef   CAS   PubMed .
  • S. M. Lundberg, G. Erion, H. Chen, A. DeGrave, J. M. Prutkin, B. Nair, R. Katz, J. Himmelfarb, N. Bansal and S.-I. Lee, From local explanations to global understanding with explainable AI for trees, Nat. Mach. Intell. , 2020, 2 , 56–67  CrossRef   PubMed .
  • R. Ouyang, S. Curtarolo, E. Ahmetcik, M. Scheffler and L. M. Ghiringhelli, SISSO: A compressed-sensing method for identifying the best low-dimensional descriptor in an immensity of offered candidates, Phys. Rev. Mater. , 2018, 2 , 083802  CrossRef   CAS .
  • B. Weng, Z. Song, R. Zhu, Q. Yan, Q. Sun, C. G. Grice, Y. Yan and W. J. Yin, Simple descriptor derived from symbolic regression accelerating the discovery of new perovskite catalysts, Nat. Commun. , 2020, 11 , 3513  CrossRef   CAS   PubMed .
  • C. J. Bartel, S. L. Millican, A. M. Deml, J. R. Rumptz, W. Tumas, A. W. Weimer, S. Lany, V. Stevanović, C. B. Musgrave and A. M. Holder, Physical descriptor for the Gibbs energy of inorganic crystalline solids and temperature-dependent materials chemistry, Nat. Commun. , 2018, 9 , 4168  CrossRef   PubMed .
  • W. Xu, M. Andersen and K. Reuter, Data-Driven Descriptor Engineering and Refined Scaling Relations for Predicting Transition Metal Oxide Reactivity, ACS Catal. , 2020, 11 , 734–742  CrossRef .
  • Z.-K. Han, D. Sarker, R. Ouyang, A. Mazheika, Y. Gao and S. V. Levchenko, Single-atom alloy catalysts designed by first-principles calculations and artificial intelligence, Nat. Commun. , 2021, 12 , 1833  CrossRef   CAS   PubMed .
  • M. E. Tipping and C. M. Bishop, Probabilistic principal component analysis, J. R. Stat. Soc., B: Stat. Methodol. , 1999, 61 , 611–622  CrossRef .
  • C. Singh, A. Askari, R. Caruana and J. Gao, Augmenting interpretable models with large language models during training, Nat. Commun. , 2023, 14 , 7913  CrossRef   CAS   PubMed .
  • J. Xia, Y. Zhu, Y. Du, Y. Liu and S. Z. Li, A Systematic Survey of Chemical Pre-trained Models , IJCAI, 2023  Search PubMed .
  • G. Zhou, Z. Gao, Q. Ding, H. Zheng, H. Xu, Z. Wei, L. Zhang and G. Ke, Uni-Mol: A Universal 3D Molecular Representation Learning Framework, The Eleventh International Conference on Learning Representations , 2023  Search PubMed .
  • Y. Du, X. Liu, N. M. Shah, S. Liu, J. Zhang and B. Zhou, ChemSpacE: Interpretable and Interactive Chemical Space Exploration, ChemRxiv , 2022, preprint,  DOI: 10.26434/chemrxiv-2022-x49mh-v3 .
  • A. Yüksel, E. Ulusoy, A. Ünlü and T. Doğan, SELFormer: molecular representation learning via SELFIES language models, Mach. Learn.: Sci. Technol. , 2023, 4 , 025035  Search PubMed .
  • J. Born and M. Manica, Regression Transformer enables concurrent sequence regression and generation for molecular language modelling, Nat. Mach. Intell. , 2023, 5 , 432–444  CrossRef .
  • D. Flam-Shepherd, K. Zhu and A. Aspuru-Guzik, Language models can learn complex molecular distributions, Nat. Commun. , 2022, 13 , 3293  CrossRef   CAS   PubMed .
  • F. Grisoni, Chemical language models for de novo drug design: Challenges and opportunities, Curr. Opin. Struct. Biol. , 2023, 79 , 102527  CrossRef   CAS   PubMed .
  • A. G. Yohannes, C. Lee, P. Talebi, D. H. Mok, M. Karamad, S. Back and S. Siahrostami, Combined High-Throughput DFT and ML Screening of Transition Metal Nitrides for Electrochemical CO2 Reduction, ACS Catal. , 2023, 13 , 9007–9017  CrossRef   CAS .
  • A. Chakkingal, P. Janssens, J. Poissonnier, A. J. Barrios, M. Virginie, A. Y. Khodakov and J. W. Thybaut, Machine learning based interpretation of microkinetic data: a Fischer–Tropsch synthesis case study, React. Chem. Eng. , 2022, 7 , 101–110  RSC .
  • M. T. Ribeiro, S. Singh and C. Guestrin, “Why Should I Trust You?”: Explaining the Predictions of Any Classifier presented in part at the Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , San Francisco, California, USA, 2016  Search PubMed .
  • H. S. Pillai, Y. Li, S. H. Wang, N. Omidvar, Q. Mu, L. E. K. Achenie, F. Abild-Pedersen, J. Yang, G. Wu and H. Xin, Interpretable design of Ir-free trimetallic electrocatalysts for ammonia oxidation with graph neural networks, Nat. Commun. , 2023, 14 , 792  CrossRef   CAS   PubMed .
  • K. Vellayappan, Y. Yue, K. H. Lim, K. Cao, J. Y. Tan, S. Cheng, T. Wang, T. Z. H. Gani, I. A. Karimi and S. Kawi, Impacts of catalyst and process parameters on Ni-catalyzed methane dry reforming via interpretable machine learning, Appl. Catal., B , 2023, 330  DOI: 10.1016/j.apcatb.2023.122593 .
  • T. Shin, Y. Razeghi, R. L. Logan IV, E. Wallace and S. Singh, AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Prompts, arXiv , 2020, preprint, arXiv:2010.15980,  DOI: 10.48550/arXiv.2010.15980 .
  • X. L. Li and P. Liang, Prefix-Tuning: Optimizing Continuous Prompts for Generation, arXiv , 2021, preprint, arXiv:2101.00190,  DOI: 10.48550/arXiv.2101.00190 .
  • K. Hambardzumyan, H. Khachatrian and J. May, WARP: Word-level Adversarial ReProgramming , Online, August, 2021  Search PubMed .
  • A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y. Lo, P. Dollár and R. Girshick, Segment Anything, arXiv , 2023, preprint, arXiv:2304.02643,  DOI: 10.48550/arXiv.2304.02643 .
  • O. Aydin and M. Y. Yassikaya, Validity and Reliability Analysis of the PlotDigitizer Software Program for Data Extraction from Single-Case Graphs, Perspect. Behav. Sci. , 2022, 45 , 239–257  CrossRef   PubMed .
  • H. Yang, S. Yue and Y. He, Auto-GPT for Online Decision Making: Benchmarks and Additional Opinions, arXiv , 2023, preprint, arXiv:2306.02224,  DOI: 10.48550/arXiv.2306.02224 .
  • A. M. Bran, S. Cox, O. Schilter, C. Baldassari, A. D. White and P. Schwaller, Augmenting large language models with chemistry tools, Nat. Mach. Intell. , 2024, 6 (5), 525–535  CrossRef   PubMed .


  1. PPT

    hypothesis science vs discovery

  2. 8 Discovery vs Hypothesis Based Science Notes

    hypothesis science vs discovery

  3. Hypothesis-testing and discovery-oriented research as complementary

    hypothesis science vs discovery

  4. What is the Difference Between Discovery Science and Hypothesis-driven

    hypothesis science vs discovery

  5. PPT

    hypothesis science vs discovery

  6. Discovery vs hypothesis-driven research proposal

    hypothesis science vs discovery


  1. Misunderstanding The Null Hypothesis

  2. Review of Hypothesis Testing 4 z test

  3. The Scientific Method

  4. Hypothesis vs Prediction 1

  5. Hypothesis vs Action#motivation

  6. Top 10 Scientific Discoveries Reshaping Our Understanding of the Universe


  1. 1.2 The Process of Science

    Both types of logical thinking are related to the two main pathways of scientific study: descriptive science and hypothesis-based science. Descriptive (or discovery) science aims to observe, explore, and discover, while hypothesis-based science begins with a specific question or problem and a potential answer or solution that can be tested. The ...

  2. 1.2: The Science of Biology

    Descriptive (or discovery) science, which is usually inductive, aims to observe, explore, and discover, while hypothesis-based science, which is usually deductive, begins with a specific question or problem and a potential answer or solution that can be tested. ... Descriptive science and hypothesis-based science are in continuous dialogue ...

  3. 1.2: The Process of Science

    Both types of logical thinking are related to the two main pathways of scientific study: descriptive science and hypothesis-based science. Descriptive (or discovery) science aims to observe, explore, and discover, while hypothesis-based science begins with a specific question or problem and a potential answer or solution that can be tested.

  4. The Process of Science in Biology

    Both types of logical thinking are related to the two main pathways of scientific study: descriptive science and hypothesis-based science. Descriptive (or discovery) science, which is usually inductive, aims to observe, explore, and discover, while hypothesis-based science, which is usually deductive, begins with a specific question or problem ...

  5. Scientific Discovery

    The view that the process of discovery (however construed) is outside the scope of philosophy of science proper was widely shared amongst philosophers of science for most of the 20 th century. The last section shows that there were some attempts to develop logics of discovery in the 1920s and 1930s, especially in the pragmatist tradition.

  6. Discovery science

    Discovery science (also known as discovery-based science) is a scientific methodology which aims to find new patterns, correlations, and form hypotheses through the analysis of large-scale experimental data.The term "discovery science" encompasses various fields of study, including basic, translational, and computational science and research. ...

  7. Discovery vs. Hypothesis-Based Science

    Discovery vs. Hypothesis-Based Science. Science Inquiry, a search for information and explanation, often focusing on specific questions. ... Discovery science can lead to important conclusions based on a type of logic called induction, where we derive generalizations from a large number of specific observations.

  8. Chapter 2: Concept 2.1

    Biology blends two main forms of scientific exploration: discovery science and hypothesis-based science. Discovery science, as you'll read later in this section, is mostly about describing nature. Hypothesis-based science, as you'll read in Concept 2.2, is mostly about explaining nature. Most scientists practice a combination of these two ...

  9. Scientific Discovery

    The concept of discovery as hypothesis-formation as it is encapsulated in the traditional distinction between context of discovery and context of justification does not explicate how new ideas are formed. ... L., 2000, Abduction, Reason, and Science: Processes of Discovery and Explanation, Dordrecht: Kluwer. ---, 2009, "Creative ...

  10. PDF Hypothesis-Driven Research

    Hypothesis-Driven Research Research types • Descriptive science: observe, describe and categorize the facts • Discovery science: measure variables to decide general patterns based on inductive reasoning • Hypothesis-driven science: make a hypothesis and then test the hypothesis using deductive reasoning

  11. The scientific method (article)

    The results of a test may either support or contradict—oppose—a hypothesis. Results that support a hypothesis can't conclusively prove that it's correct, but they do mean it's likely to be correct. On the other hand, if results contradict a hypothesis, that hypothesis is probably not correct.

  12. Theory vs. Hypothesis: Basics of the Scientific Method

    Science & Tech Theory vs. Hypothesis: Basics of the Scientific Method. Written by MasterClass. Last updated: Jun 7, 2021 • 2 min read. Though you may hear the terms "theory" and "hypothesis" used interchangeably, these two scientific terms have drastically different meanings in the world of science. Articles. Videos. Instructors.

  13. Scientific hypothesis

    scientific hypothesis, an idea that proposes a tentative explanation about a phenomenon or a narrow set of phenomena observed in the natural world.The two primary features of a scientific hypothesis are falsifiability and testability, which are reflected in an "If…then" statement summarizing the idea and in the ability to be supported or refuted through observation and experimentation.

  14. Exploration and discovery

    The early stages of a scientific investigation often rely on making observations, asking questions, and initial experimentation — essentially poking around. But the routes to and from these stages are diverse. Intriguing observations sometimes arise in surprising ways, as in the discovery of radioactivity, which was inspired by the observation that photographic plates (an early version of ...

  15. 1.2: The Process of Science

    Both types of logical thinking are related to the two main pathways of scientific study: descriptive science and hypothesis-based science. Descriptive (or discovery) science aims to observe, explore, and discover, while hypothesis-based science begins with a specific question or problem and a potential answer or solution that can be tested. The ...

  16. What is a scientific hypothesis?

    Thus, the hypothesis is true, but it may not be true 100% of the time. Scientific theory vs. scientific hypothesis. The best hypotheses are simple. They deal with a relatively narrow set of phenomena.

  17. Philosophy of science

    Following Hans Reichenbach (1891-1953), philosophers often distinguished between the "context of discovery" and the "context of justification." Once a hypothesis has been proposed, there are canons of logic that determine whether or not it should be accepted—that is, there are rules of method that hold in the context of justification.

  18. Editorial: Would You Like A Hypothesis With Those Data? Omics and the

    Pat Brown, an advocate of discovery science, asserted that "the goal is to discover things we neither knew nor expected, and to see relationships and connections among the elements, whether previously suspected or not. ... The debate about hypothesis-driven vs data-driven science has made its way into the grant review process. For grant ...

  19. What's the Difference Between a Fact, a Hypothesis, a ...

    It's so thoroughly proven, you might even call it a "scientific fact." Unfortunately, all of these common impressions aren't quite right. The words "fact," "hypothesis," "theory," and "law" have very specific meanings in the world of science, and they don't exactly match the ones we use in everyday language.

  20. Process of Science Flashcards

    What is the difference between discovery-based science and hypothesis testing? In discovery science, you make observations or analyze data. In hypothesis-based science, you form an educated guess, or hypothesis. Know the general difference between inductive and deductive reasoning.

  21. Chap #1 Flashcards

    Study with Quizlet and memorize flashcards containing terms like What is the difference between discovery science and hypothesis-driven science? A) Discovery science involves predictions about outcomes, whereas hypothesis-driven science involves tentative answers to specific questions. B) Discovery science "discovers" new knowledge, whereas hypothesis-driven science does not. C) Discovery ...

  22. Discovery Science vs. Hypothesis Based Science

    Blog. April 18, 2024. Use Prezi Video for Zoom for more engaging meetings; April 16, 2024. Understanding 30-60-90 sales plans and incorporating them into a presentation

  23. Explaining turmeric's health benefits

    Your source for health, wellness, innovation, research and science news from the experts at Ohio State. There's a powerful story behind every headline at Ohio State Health & Discovery. As one of the largest academic health centers and health sciences campuses in the nation, we are uniquely positioned with renowned experts covering all aspects ...

  24. 'A history of contact': Princeton geneticists are rewriting the

    A gene that shaped the evolution of Darwin's finches.. Researchers from Princeton University and Uppsala University in Sweden have identified a gene in the Galápagos finches studied by English naturalist Charles Darwin that influences beak shape and that played a role in the birds' evolution from a common ancestor more than 1 million years ago.

  25. Automation and machine learning augmented by large ...

    1 Introduction The field of catalyst design and discovery is undergoing a profound transformation, facilitated by the convergence of artificial intelligence (AI) 1-3 and automation systems, 4-6 as well as utilization of large data. This shift is propelled by advancements in four crucial areas: high-throughput information extraction, 7-16 automated robotic systems for chemical ...