The scientific method refers to the principles and approaches used to develop knowledge by formulating questions about natural and social phenomena, collecting evidence through observation and experimentation, and critically examining hypotheses and theories. Science is not simply the accumulation of facts or the application of existing theories to explain observations. It also involves discovering new problems, proposing explanations for observed phenomena, and investigating how well those explanations correspond to reality.
The scientific method is often presented as a sequence: Observation → Hypothesis → Experiment → Verification → Conclusion. This description is useful for introducing scientific inquiry, but actual research does not always proceed in this order. Researchers may begin with a hypothesis and design observations to test it. Alternatively, unexpected experimental results may lead them to reconsider the original problem. In some scientific disciplines, observation, comparative studies, historical evidence, or mathematical models play a greater role than direct experimentation.
Nevertheless, an important principle is shared across many forms of scientific inquiry: We should examine not only evidence that supports our ideas but also evidence that might show them to be mistaken. Karl Popper (1902–1994) articulated this principle through his influential concept of falsifiability.
Is There a Fixed Procedure for the Scientific Method?
When considering the scientific method, it is useful to distinguish between the process of generating hypotheses or theories and the process of evaluating them.
New ideas may arise from observations, contradictions within existing theories, analogies with other fields, unexpected discoveries, or intuition. Some investigations begin by collecting and analyzing large amounts of data without a clearly defined initial hypothesis.
For example, when an unexpected property of a material is discovered, researchers may develop new hypotheses to explain it.
Conversely, researchers may derive predictions from an established theory and design experiments to determine whether the predicted phenomena occur.
In the first case, observation precedes hypothesis formation. In the second, the hypothesis guides the design of observations.
Neither approach is inherently more scientific than the other.
In philosophy of science, a distinction has traditionally been made between the context of discovery, which concerns how hypotheses arise, and the context of justification, which concerns how those hypotheses are evaluated.
Although this distinction is not always clear-cut, it is useful for separating the originality of an idea from the validity of the claims made about it.
Even if a promising hypothesis originates from intuition, its origin does not establish its correctness.
Conversely, a hypothesis discovered by accident or through trial and error may contribute to scientific knowledge if it subsequently receives appropriate evidential support.
The scientific method is not necessarily a prescribed procedure for making discoveries. It is also a process of critically examining ideas and revising them to develop better explanations.
Induction, Deduction, and Abduction in Scientific Inquiry
Three forms of reasoning are particularly useful for understanding scientific inquiry: induction, deduction, and abduction.
| Type of Reasoning | Basic Idea | Example |
|---|---|---|
| Induction | Infer general patterns from particular observations | Repeated experiments produce similar results, suggesting a general relationship under specified conditions |
| Deduction | Derive specific predictions from general hypotheses or premises | If a theory is correct, a particular phenomenon should occur under defined experimental conditions |
| Abduction | Develop possible explanations for observed phenomena | An unexpected improvement in material performance may result from a change in its microstructure |
Induction involves inferring general regularities from observed instances.
However, repeatedly observing the same result does not logically guarantee that the relationship will always hold in the future. This is the problem of induction, famously discussed by David Hume.
Deduction derives conclusions from premises according to logical rules.
If the reasoning is valid and the premises are true, the conclusion follows. However, deduction alone cannot establish whether the original hypothesis is correct.
Abduction involves developing hypotheses that might explain observed phenomena.
For example, if a material exhibits unexpectedly high durability, researchers may consider possible explanations involving its microstructure, chemical composition, or processing conditions.
However, several different hypotheses may explain the same observation.
Scientific inquiry frequently combines these forms of reasoning.
Researchers identify patterns through observation, develop possible explanations, derive predictions from those explanations, and examine the predictions through further observations or experiments.
What Is Popper’s Falsifiability?
Karl Popper introduced falsifiability as a central criterion for scientific theories in Logik der Forschung, published in 1934 and subsequently translated into English as The Logic of Scientific Discovery.
Popper was particularly concerned with distinguishing scientific theories from claims that appear scientific but cannot be subjected to meaningful empirical criticism.
This is known as the demarcation problem.
Falsifiability means that it must be possible, at least in principle, to specify an observation or experimental result that would count against a theory and demonstrate that it is mistaken.
Consider the statement:
“All swans are white.”
Observing thousands of white swans does not prove that every swan in the world is white.
However, discovering a single black swan, and establishing that it is indeed a swan, is sufficient to refute the universal claim.
This illustrates a logical asymmetry between accumulating confirming observations and identifying a counterexample.
Popper considered this distinction an important characteristic of scientific theories.
For example, consider the hypothesis:
“Under the specified conditions, this material does not undergo a structural change at temperatures up to 100°C.”
An observed structural change under those conditions would count as a potential falsification.
By contrast, a claim such as “This material has some unknown potential under every possible condition” is difficult to evaluate empirically because almost any result could be made compatible with it.
Falsifiability does not mean that a theory is false.
Nor does it mean that a theory becomes true simply because it has not been falsified.
Rather, falsifiability concerns whether a theory is open to empirical evidence that could reveal its errors.
Falsifiability and the Formulation of Scientific Hypotheses
When evaluating a scientific hypothesis, it is important to clarify not only what we expect to happen but also what findings would require us to reconsider the hypothesis.
Suppose an engineer proposes:
“A new material structure might improve impact resistance.”
This is a potentially useful idea, but it does not yet specify what should be measured or compared.
The hypothesis could be made more precise:
“When specimens with the same mass and external dimensions are tested under specified impact conditions, the new structure will exhibit less maximum deformation than the conventional structure.”
This formulation identifies the objects being compared, the experimental conditions, and the outcome to be measured.
The experiment can be made more reliable by considering the number of specimens, material variability, measurement error, and the magnitude of improvement regarded as meaningful.
If the results contradict the prediction, researchers should investigate why and reconsider the proposed structure or hypothesis.
However, failing to observe a difference in a single experiment does not necessarily establish that the new structure has no beneficial effect.
The precision of the experiment, testing conditions, and variability of measurements must also be considered.
The important principle is not to protect the initial hypothesis but to ensure that it can be revised in response to evidence.
Evidence Supporting a Hypothesis Is Not the Same as Proof
When experimental results are consistent with a hypothesis, they are generally treated as supporting evidence.
However, successfully predicting an observation does not establish that the hypothesis is the only correct explanation.
Consider the hypothesis:
“The material became stronger because its microstructure changed.”
Even if increased strength is observed, other factors may explain the result.
Impurities, manufacturing conditions, measurement procedures, or other mechanisms may have contributed.
It is common for several hypotheses to be compatible with the same observations.
Therefore, scientific evaluation requires more than collecting examples consistent with a preferred hypothesis.
Researchers should seek observations or experiments capable of distinguishing between competing explanations.
Popper emphasized corroboration: the idea that a theory has survived demanding tests rather than been established as an absolute truth.
No matter how many observations are consistent with a theory, the theory should remain open to critical examination.
However, contemporary statistics and philosophy of science have also developed approaches that differ from Popper’s, including methods for evaluating degrees of evidential support and updating probabilistic beliefs.
Falsifiability is an important perspective on scientific reasoning, but it does not provide a complete account of every form of scientific inference.
Why Theories Are Not Always Rejected When Predictions Fail
The principle of falsifiability is relatively straightforward, but actual scientific research is more complicated.
A theory is not necessarily abandoned immediately when an observation conflicts with a prediction.
For example, suppose a physical theory predicts a particular measurement, but the observed value is different.
The theory may be wrong.
However, the discrepancy might also result from an instrument malfunction, incorrect experimental conditions, a computational error, or an overlooked external influence.
Experiments and observations generally depend on numerous auxiliary assumptions in addition to the central theory being tested.
Consequently, when a prediction fails, the result does not always identify which assumption should be rejected.
This problem is associated with the work of Pierre Duhem and W. V. O. Quine and is commonly discussed under the name Duhem–Quine thesis. Their philosophical positions were not identical, but both contributed to discussions of the difficulty of testing individual hypotheses in isolation.
A famous historical example concerns irregularities in the observed orbit of Uranus.
These irregularities appeared to conflict with predictions based on Newtonian mechanics.
One possibility was that Newtonian mechanics was incorrect.
Another was that the gravitational influence of an unknown planet had not been included in the calculations.
The latter possibility led to predictions concerning the location of Neptune, which was discovered in 1846.
In this case, the disagreement between prediction and observation did not simply lead to the rejection of an existing theory. It contributed to the discovery of a previously unknown object.
However, modifying a theory by repeatedly adding convenient explanations can also become problematic.
If every unexpected observation is explained away by introducing another untestable assumption, the theory may become resistant to meaningful criticism.
The important question is whether a modification produces new testable predictions, explains previously unknown phenomena, or leads to additional knowledge.
Kuhn and Lakatos — Does Science Progress Only Through Falsification?
After Popper, philosophers and historians of science examined the limitations of treating falsification as the central mechanism of scientific progress.
Thomas S. Kuhn, in The Structure of Scientific Revolutions (1962), argued that scientific research is guided by shared theoretical frameworks, accepted problems, methods of measurement, and exemplary scientific achievements.
Kuhn referred to research conducted within an established framework as normal science.
During normal science, researchers do not continually challenge every foundational assumption. Instead, they use the accepted framework to solve specific problems.
This activity can lead to increasingly precise measurements and theoretical developments.
At the same time, researchers may encounter anomalies that are difficult to explain within the existing framework.
When such anomalies become sufficiently significant and the established framework can no longer address them adequately, a transition to a different framework may occur.
Kuhn’s account highlights that scientific development may involve changes not only in accumulated knowledge but also in the conceptual framework used to define problems and describe the world.
Imre Lakatos proposed another approach, focusing on scientific research programmes rather than isolated hypotheses.
A research programme consists of a series of related theories and methodological commitments.
Lakatos sought to distinguish programmes that generate new predictions and discoveries from those that mainly introduce modifications to accommodate already known difficulties.
These perspectives do not make Popper’s falsifiability irrelevant.
Rather, they suggest that evaluating scientific theories requires consideration of more than a single experimental result.
The history of a research programme, competing theories, measurement procedures, and the new knowledge generated through theoretical revisions may all be relevant.
Scientific Inquiry Is Not Limited to Experiments
The scientific method is frequently explained through laboratory experiments, but scientific disciplines employ many different approaches.
In astronomy, researchers usually cannot manipulate the objects they study. Instead, they infer the properties of celestial bodies from observed light, motion, and other evidence.
In geology and evolutionary biology, scientists reconstruct historical processes using evidence such as geological formations, fossils, and genetic information.
Theoretical physics and mathematical sciences use mathematical models to derive relationships and predictions that may subsequently be compared with observations or experiments.
Computer simulations can also investigate conditions that would be difficult or impossible to reproduce experimentally.
However, simulation results depend on the assumptions and computational procedures used in the model. Their relevance to reality must be evaluated separately.
Rather than defining science through a single procedure, it may be more useful to ask:
- What phenomenon is the investigation attempting to explain?
- What evidence supports the proposed explanation?
- What alternative explanations are possible?
- Which observations or analyses could help distinguish between those explanations?
- Under what conditions and within what limits should the conclusions be expected to hold?
Different scientific disciplines require different kinds of evidence and methods.
Nevertheless, presenting claims in a form that allows others to examine and challenge them is an important feature of scientific inquiry.
Are Falsifiability and Statistical Hypothesis Testing the Same Thing?
Statistics provides methods for formulating hypotheses and evaluating them using observed data.
For example, researchers may test a null hypothesis stating that two manufacturing methods produce the same population mean strength.
A statistical test can then calculate a p-value using the measured data.
However, Popper’s concept of falsifiability and statistical hypothesis testing are not identical.
Falsifiability concerns whether a theory could, in principle, be shown to be mistaken through empirical evidence.
Statistical hypothesis testing is a collection of procedures for evaluating hypotheses while accounting for variability and uncertainty in observed data.
For example, a result that is not statistically significant does not prove that two manufacturing methods have identical effects.
An experiment may have insufficient precision, too few observations, or inadequate statistical power to detect a difference.
Similarly, when a theory makes probabilistic predictions, a single observation that differs from the most likely outcome does not necessarily refute the theory.
The probability distribution implied by the theory and the overall agreement between observations and predictions must be examined using appropriate methods.
It is therefore important to distinguish between the logical concept of falsifiability and the statistical uncertainty involved in practical hypothesis testing.
Exploratory Research and Confirmatory Research
Scientific research includes exploratory investigations intended to discover new phenomena or relationships and confirmatory investigations intended to evaluate previously formulated hypotheses.
Suppose researchers analyze a large dataset of materials and identify a group with unexpected properties.
At this stage, they are exploring the data and discovering potentially interesting relationships.
Such exploration can be scientifically valuable because it generates new questions and hypotheses.
However, using the same dataset both to discover an unexpected relationship and to claim that the relationship has been independently confirmed can lead to overconfidence.
When many possible relationships are examined, some apparently remarkable patterns may arise through random variation.
Therefore, hypotheses discovered through exploratory analysis should ideally be examined using independent data or additional experiments.
In confirmatory research, it is often useful to specify the hypotheses, measurements, evaluation criteria, and analytical methods before examining the results.
Preregistration involves recording a research plan before conducting or analyzing the study. It can help distinguish analyses planned in advance from additional exploratory analyses conducted after examining the data.
However, preregistration does not mean that research plans can never be revised.
Unexpected results may justify new investigations, and exploratory analyses can be valuable.
The important point is to report transparently whether an analysis was planned as a confirmatory test or developed in response to observations.
Exploratory research should not be dismissed simply because it does not begin with a predetermined hypothesis. At the same time, exploratory findings should not automatically be presented as independently verified conclusions.
Reproducibility, Replication, and the Reliability of Scientific Knowledge
Scientific claims should be presented in a form that allows other researchers to examine them.
This requires appropriate documentation of research methods, measurement conditions, data, and analytical procedures.
Modern research methodology sometimes distinguishes between reproducibility and replicability.
For example, a 2019 report by the U.S. National Academies of Sciences, Engineering, and Medicine defines reproducibility in computational research as obtaining consistent results using the same data and computational methods.
Replicability refers to obtaining consistent findings across studies addressing the same scientific question using independently collected data.
However, terminology varies among scientific disciplines.
Reproducing calculations using the original data helps verify analytical procedures, but it does not necessarily establish that the scientific conclusions are correct.
Likewise, if an independent investigation produces different results, this does not automatically mean that the original study was wrong.
Differences in study populations, experimental conditions, measurement procedures, or random variation may need to be considered.
Confidence in scientific knowledge generally increases when claims are examined repeatedly through independent evidence and different investigative methods.
The Limits of Falsifiability — Can Science and Non-Science Be Clearly Separated?
Popper’s concept of falsifiability has had a major influence on philosophy of science.
However, whether it provides a complete criterion for distinguishing science from non-science remains controversial.
First, actual scientific investigations depend on multiple assumptions.
When observations conflict with predictions, it is often difficult to determine which assumptions should be revised.
Second, scientific research includes activities such as description, classification, measurement development, and exploratory data collection.
These activities are not always adequately characterized as attempts to falsify a clearly formulated hypothesis.
Third, a hypothesis being falsifiable does not mean that it is well supported by evidence.
A falsifiable claim may still lack reliable observations or be based on poor-quality research.
Furthermore, claims that are not empirically falsifiable are not necessarily meaningless.
Ethical judgments and philosophical questions may concern matters that cannot be directly settled through experiments.
Falsifiability is therefore better understood not as a mechanical rule capable of classifying every intellectual activity but as an important principle that asks whether we are willing to expose our ideas to evidence that could require us to revise them.
Hypothesis Generation and Testing in Research and Development
Corporate research and development involves several kinds of hypotheses, including scientific hypotheses, hypotheses about technological feasibility, and business hypotheses.
These may be related, but they concern different questions and require different forms of evaluation.
| Type of Hypothesis | Example | Main Evaluation Methods |
|---|---|---|
| Scientific Hypothesis | A particular microstructure increases material strength | Measurements, experiments, comparison with theoretical models |
| Technological Hypothesis | A new processing method can be implemented in mass-production equipment | Prototyping, process testing, durability and quality evaluation |
| Customer Hypothesis | A particular customer segment needs a new function | Interviews, behavioral observation, usage experiments |
| Business Hypothesis | A new product can be supplied sustainably at a particular price and cost | Market research, sales trials, cost and revenue analysis |
For example, the idea that technology from another industry could enable a new product may be a useful starting point for R&D.
However, this statement alone does not identify what should be investigated.
Researchers must identify specific technologies, understand their functions, and consider how those functions might be applied to the target product.
They must then determine whether the required performance can be achieved using actual materials or equipment and whether the proposed approach offers advantages over existing technologies.
The distinction between generating ideas and evaluating hypotheses is particularly important here.
During the exploration of new technologies, restricting the investigation to a narrowly predefined hypothesis may prevent unexpected discoveries.
Once a concrete development candidate has been selected, however, it becomes important to define the performance requirements and specify what results would justify rejecting or reconsidering the proposal.
Thus, open-ended exploration and rigorous testing are not opposing activities.
They serve different purposes at different stages of research and development.
The Scientific Method in the AI Era
Large language models are creating opportunities to support various aspects of scientific inquiry.
For example, AI systems can help organize existing knowledge and unresolved questions from research literature, explore connections between technological fields, and generate candidate hypotheses.
They can also assist in deriving predictions, considering experimental designs, and identifying alternative explanations for analytical results.
However, an LLM generating a plausible explanation does not establish that the explanation is correct.
If an AI system proposes that phenomena from different technological fields share a common principle, the proposed correspondence must be investigated.
Similar terminology or descriptions do not necessarily indicate similar operating mechanisms or constraints.
AI systems may also generate explanations that support a hypothesis supplied by the user, rather than examining it critically.
A rigorous investigation therefore requires searching not only for supporting examples but also for competing explanations, counterexamples, necessary conditions, and circumstances in which the hypothesis might fail.
Popper’s idea of falsifiability remains useful in this context.
Instead of asking only:
“Find examples supporting this hypothesis.”
Researchers can also ask:
“What observations would require us to revise this hypothesis?”
“What alternative hypotheses could explain the same phenomenon?”
“Under what conditions would the proposed mechanism fail?”
However, an AI-generated counterexample is not necessarily a real counterexample.
Sources, measurements, and experimental evidence must be examined before drawing conclusions.
AI can assist with hypothesis generation and critical examination, but the evaluation of hypotheses as scientific knowledge ultimately requires evidence connected to the real world.
The Scientific Method in ThinkNavi
ThinkNavi treats the scientific method not simply as a standalone procedure but as a set of principles underlying many thinking and analytical techniques.
For example, the KJ Method can help reveal new problem structures from observations, while abductive reasoning supports the formation of explanatory hypotheses.
5 Whys and Cause-and-Effect Diagrams can help organize possible causes.
Design of Experiments provides methods for systematically investigating hypotheses through controlled changes in experimental conditions.
Statistical analysis can examine observed relationships, effect sizes, and uncertainty.
These activities do not necessarily follow a single direction.
Testing a hypothesis may reveal an unexpected problem, requiring the original hypothesis to be reformulated or the scope of the investigation to be changed.
ThinkNavi therefore emphasizes several important distinctions:
- Observed facts and interpretations of those facts
- Evidence supporting a hypothesis and evidence contradicting it
- Established statistical relationships and causal relationships that remain unverified
- Hypotheses generated through exploration and hypotheses subjected to independent testing
- Scientific knowledge and judgments about technological implementation or commercial feasibility
In LLM-assisted analysis, a useful approach would be to connect candidate hypotheses with supporting literature or data, competing explanations, proposed verification methods, and experimental findings.
These are potential directions for future development, not descriptions of all functions currently available.
Popper’s falsifiability is not the only principle needed to understand scientific inquiry.
Nevertheless, the willingness to reconsider our own ideas in the light of contradictory evidence has enduring importance, particularly when we have become strongly convinced that those ideas are correct.
The value of the scientific method does not lie in following a fixed procedure that guarantees the correct answer. It lies in freely generating new questions and hypotheses, examining them critically through evidence, and continually revising our understanding when necessary.
Related Topics
- What Is Statistics? — Discovering Patterns in Data and Understanding an Uncertain World
- Epistemology | What Can We Know, and How?
- Abduction | Reasoning Toward Explanations
- Philosophy of Science | What Makes Knowledge Scientific?
- Critical Thinking | Examining Claims and Evidence
- Develop Hypotheses | Explain and Test
- Design of Experiments (DOE) — Investigating Multiple Factors and Their Interactions Through Systematic Experimentation
- Five Whys | Exploring Root Causes by Repeatedly Asking “Why?”
- Cause-and-Effect Diagram (Fishbone Diagram) — Systematically Organizing Factors That May Contribute to a Problem
- The KJ Method | Building Meaningful Structures from Fragmented Information
- Multivariate Analysis — Discovering Relationships and Structures Among Multiple Variables
- Concept Formation and Categorization | Change the Categories, Change What You See
- Concept Exploration
References and Further Reading
Center for Open Science — Preregistration
Karl R. Popper, The Logic of Scientific Discovery, Hutchinson, 1959 (original German edition, Logik der Forschung, 1934).
Karl R. Popper, Conjectures and Refutations: The Growth of Scientific Knowledge, Routledge & Kegan Paul, 1963.
Thomas S. Kuhn, The Structure of Scientific Revolutions, University of Chicago Press, 1962.
Imre Lakatos, The Methodology of Scientific Research Programmes, Cambridge University Press, 1978.
Pierre Duhem, The Aim and Structure of Physical Theory, Princeton University Press, 1954 (original French edition, 1906).
Stanford Encyclopedia of Philosophy — Scientific Method
Stanford Encyclopedia of Philosophy — Karl Popper
Stanford Encyclopedia of Philosophy — Science and Pseudo-Science
Stanford Encyclopedia of Philosophy — Thomas Kuhn
Stanford Encyclopedia of Philosophy — Imre Lakatos
Stanford Encyclopedia of Philosophy — Pierre Duhem
National Academies — Reproducibility and Replicability in Science, 2019.