define unfalsifiable hypothesis

Experiments

Empirical Evidence

Primary Data

Positive control, scientific fact, explanatory power, electromagnetic spectrum.

Logical Thinking

Intelligence examples, rationalism vs empiricism, social constructionism, emotion opposite, true opposite, begging the question, i think therefore i am, new articles.

Karl Popper: Theory of Falsification

Saul McLeod, PhD

Editor-in-Chief for Simply Psychology

BSc (Hons) Psychology, MRes, PhD, University of Manchester

Saul McLeod, PhD., is a qualified psychology teacher with over 18 years of experience in further and higher education. He has been published in peer-reviewed journals, including the Journal of Clinical Psychology.

Learn about our Editorial Process

Olivia Guy-Evans, MSc

Associate Editor for Simply Psychology

BSc (Hons) Psychology, MSc Psychology of Education

Olivia Guy-Evans is a writer and associate editor for Simply Psychology. She has previously worked in healthcare and educational sectors.

Karl Popper’s theory of falsification contends that scientific inquiry should aim not to verify hypotheses but to rigorously test and identify conditions under which they are false. For a theory to be valid according to falsification, it must produce hypotheses that have the potential to be proven incorrect by observable evidence or experimental results. Unlike verification, falsification focuses on categorically disproving theoretical predictions rather than confirming them.

• Karl Popper believed that scientific knowledge is provisional – the best we can do at the moment.
• Popper is known for his attempt to refute the classical positivist account of the scientific method by replacing induction with the falsification principle.
• The Falsification Principle, proposed by Karl Popper, is a way of demarcating science from non-science. It suggests that for a theory to be considered scientific, it must be able to be tested and conceivably proven false.
• For example, the hypothesis that “all swans are white” can be falsified by observing a black swan.
• For Popper, science should attempt to disprove a theory rather than attempt to continually support theoretical hypotheses.

Theory of Falsification

Karl Popper is prescriptive and describes what science should do (not how it actually behaves). Popper is a rationalist and contended that the central question in the philosophy of science was distinguishing science from non-science.

Karl Popper, in ‘The Logic of Scientific Discovery’ emerged as a major critic of inductivism, which he saw as an essentially old-fashioned strategy.

Popper replaced the classical observationalist-inductivist account of the scientific method with falsification (i.e., deductive logic) as the criterion for distinguishing scientific theory from non-science.

All inductive evidence is limited: we do not observe the universe at all times and in all places. We are not justified, therefore, in making a general rule from this observation of particulars.

According to Popper, scientific theory should make predictions that can be tested, and the theory should be rejected if these predictions are shown not to be correct.

He argued that science would best progress using deductive reasoning as its primary emphasis, known as critical rationalism.

Popper gives the following example:

Europeans, for thousands of years had observed millions of white swans. Using inductive evidence, we could come up with the theory that all swans are white.

However, exploration of Australasia introduced Europeans to black swans.  Poppers’ point is this: no matter how many observations are made which confirm a theory, there is always the possibility that a future observation could refute it.  Induction cannot yield certainty.

Karl Popper was also critical of the naive empiricist view that we objectively observe the world. Popper argued that all observation is from a point of view, and indeed that all observation is colored by our understanding. The world appears to us in the context of theories we already hold: it is ‘theory-laden.’

Popper proposed an alternative scientific method based on falsification.  However, many confirming instances exist for a theory; it only takes one counter-observation to falsify it. Science progresses when a theory is shown to be wrong and a new theory is introduced that better explains the phenomena.

For Popper, the scientist should attempt to disprove his/her theory rather than attempt to prove it continually. Popper does think that science can help us progressively approach the truth, but we can never be certain that we have the final explanation.

Critical Evaluation

Popper’s first major contribution to philosophy was his novel solution to the problem of the demarcation of science. According to the time-honored view, science, properly so-called, is distinguished by its inductive method – by its characteristic use of observation and experiment, as opposed to purely logical analysis, to establish its results.

The great difficulty was that no run of favorable observational data, however long and unbroken, is logically sufficient to establish the truth of an unrestricted generalization.

Popper’s astute formulations of logical procedure helped to reign in the excessive use of inductive speculation upon inductive speculation, and also helped to strengthen the conceptual foundation for today’s peer review procedures.

However, the history of science gives little indication of having followed anything like a methodological falsificationist approach.

Indeed, and as many studies have shown, scientists of the past (and still today) tended to be reluctant to give up theories that we would have to call falsified in the methodological sense, and very often, it turned out that they were correct to do so (seen from our later perspective).

The history of science shows that sometimes it is best to ’stick to one’s guns’. For example, “In the early years of its life, Newton’s gravitational theory was falsified by observations of the moon’s orbit”

Also, one observation does not falsify a theory. The experiment may have been badly designed; data could be incorrect.

Quine states that a theory is not a single statement; it is a complex network (a collection of statements). You might falsify one statement (e.g., all swans are white) in the network, but this should not mean you should reject the whole complex theory.

Critics of Karl Popper, chiefly Thomas Kuhn , Paul Feyerabend, and Imre Lakatos, rejected the idea that there exists a single method that applies to all science and could account for its progress.

Popperp, K. R. (1959). The logic of scientific discovery . University Press.

Further Information

• Thomas Kuhn – Paradigm Shift Is Psychology a Science?
• Steps of the Scientific Method
• Positivism in Sociology: Definition, Theory & Examples
• The Scientific Revolutions of Thomas Kuhn: Paradigm Shifts Explained

You Can Know Things

A BLOG ABOUT SCIENCE IN A WORLD OF UNTRUE FACTS

When you can never be wrong: the unfalsifiable hypothesis

• February 9, 2021

By Kristen Panthagani, PhD

If there was one single scientific concept I could teach everyone in the country right now it would be this: what is an  unfalsifiable hypothesis , and why do they confuse everyone.

This concept alone explains a lot of the confusion and conspiracy theories around the COVID pandemic… why many still insist that Bill Gates was involved in planning the pandemic or that there are microchips in vaccines. 

What is a hypothesis?

Before we get to unfalsifiable hypotheses, let’s start with what a hypothesis is. In very simple terms, a hypothesis is a tentative explanation that needs to be tested . It’s an idea formed on the available evidence that is maybe true, but still needs to be explored and verified. For example, at the beginning of the pandemic, many had the hypothesis that hydroxychloroquine is an effective treatment for COVID.  

Hypotheses are the jumping off points of scientific experiments. They define what question we want to test. And that brings us to one of the most important qualities of a valid scientific hypothesis: they must actually be testable. Or said another way,  they must be falsifiable.

What is a falsifiable hypothesis?

What does it mean for a hypothesis to be falsifiable? It means that we can actually design an experiment to test if it’s wrong (false).  For a hypothesis to be falsifiable, we must be able to design a test that provides us with one of three possible outcomes:

1. the results support the hypothesis,* or

2. the results are inconclusive, or 

3. the results reject the hypothesis. 

When the results reject our hypothesis, it tells us our hypothesis is wrong, and we move on.

*If we want to be nitpicky, instead of saying the results ‘support’ our hypothesis we should really say ‘the results fail to disprove our hypothesis.’ But, that’s beyond the scope of what you need to know for this post.

When the results reject our hypothesis, it tells us our hypothesis is wrong, and we move on. Tweet

That is the hallmark of a falsifiable hypothesis: you can find out when you’re wrong. So then, what is an unfalsifiable hypothesis? It is a hypothesis that is impossible to disprove . And it is not impossible to disprove because it’s correct, it’s impossible to disprove because there is no way to conclusively test it. For unfalsifiable hypotheses, every test you run will come up with not three, but two possible outcomes: 

1. the results support the hypothesis or

2. the results are inconclusive. 

‘ Results reject the hypothesis ‘  is missing. No amount of testing will ever lead to data that conclusively rejects the hypothesis, even if the hypothesis is completely wrong.

For unfalsifiable hypotheses that happen to be true (i.e. love exists), this is not a huge issue, because it’s usually pretty obvious that they’re right, despite their unfalsifiability. The problem arises for unfalsifiable hypotheses that are more tenuous claims.

In these cases, people may deeply believe they’re right, in part, because it is impossible to find conclusive evidence that they’re wrong.   Every time they try to test if their claim is true, they only find inconclusive evidence. And again, this is not because the hypothesis is correct, it’s because the hypothesis is set up in a way where a definitive “no that’s wrong” is impossible to find. A great example is the hypothesis that there are microchips in the vaccines. You could say ‘well just look in one and see if it’s there!’ And somebody checks and finds no microchip. End of story? Well no.. someone could argue ‘well the microchips are just too small to detect!’ or ‘They will know to take it out of the vials before they are scanned!’ Excuses are made so that the negative results are no longer negative results, but instead are inconclusive. Thus every possible result from any test we do can be deemed inconclusive by those who believe the hypothesis is correct. This makes the hypothesis, for the sake of the people who believe in it, unfalsifiable. This is why conspiracy theories are so hard to debunk… many of them are unfalsifiable hypotheses.

Why do these trap people so effectively? Two reasons. First, for a believer of the hypothesis, all they see is inconclusive data (which they can usually make fit their narrative). They never see any data disproving it, so it makes it easy for them to believe they’re right. And second, because it’s impossible to conclusively disprove it, we can’t go and… conclusively disprove it. This makes it easy for people to stay trapped in an unfalsifiable hypothesis they want to believe in, even when it’s 100% wrong.

So how do you know if you’ve been trapped into believing an unfalsifiable hypothesis? Ask yourself… how would I know if this was false? What evidence would come forward that would convince me? If the answer is ‘ well, I’m waiting for the results of this study to decide ‘ or ‘ I’m waiting for the outcome of this particular event to know ,’ then that suggests you’re not trapped in an unfalsifiable hypothesis, as you are open to actual evidence showing you that you’re wrong. (But, only if you do actually change your mind if that evidence fails to support your hypothesis, rather than finding an excuse why that event or evidence doesn’t actually disprove it.)

But, if the answer relies not on specific events or outcomes but primarily on the opinion of other believers, then you may be trapped in an unfalsifiable hypothesis, because that isn’t evidence… it’s just group think.

Want to see future posts?

Subscribe   or follow on:

What is going on with bird flu?

Threads, please don’t censor COVID

Dear NIH, please fund health communication research

St. Jude Family of Websites

Explore our cutting edge research, world-class patient care, career opportunities and more.

St. Jude Children's Research Hospital Home

• St. Jude Research
• Progress: A Digital Magazine
• St. Jude Cloud
• Childhood Cancer Survivor Study
• Careers at St. Jude
• Have More in Memphis
• Care & Treatment at St. Jude
• Together by St. Jude™ online resource
• Long-Term Follow-Up Study
• St. Jude LIFE
• Education & Training at St. Jude
• St. Jude Graduate School of Biomedical Sciences
• St. Jude Global

• Get Involved
• Ways to Give

A hypothesis can’t be right unless it can be proven wrong

Charles Rock, PhD, (right) and Jiangwei Yao, PhD, recently reviewed Richard Harris’ book about scientific research, titled "Rigor Mortis: How Sloppy Science Creates Worthless Cures, Crushes Hope, and Wastes Billions." Now, Rock and Yao address specific issues raised in Harris’ book and offer solutions or tips to help avoid the pitfalls identified in the book.

“That (your hypothesis) is not only not right; it is not even wrong.” Wolfgang Pauli (Nobel Prize in Physics, 1945)

A hypothesis is the cornerstone of the scientific method.

It is an educated guess about how the world works that integrates knowledge with observation.

Everyone appreciates that a hypothesis must be testable to have any value, but there is a much stronger requirement that a hypothesis must meet.

A hypothesis is considered scientific only if there is the possibility to disprove the hypothesis.

The proof lies in being able to disprove

A hypothesis or model is called falsifiable if it is possible to conceive of an experimental observation that disproves the idea in question. That is, one of the possible outcomes of the designed experiment must be an answer, that if obtained, would disprove the hypothesis.

Our daily horoscopes are good examples of something that isn’t falsifiable. A scientist cannot disprove that a Piscean may get a surprise phone call from someone he or she hasn’t heard from in a long time. The statement is intentionally vague. Even if our Piscean didn’t get a phone call, the prediction cannot be false because he or she may get a phone call. They may not.

A good scientific hypothesis is the opposite of this. If there is no experimental test to disprove the hypothesis, then it lies outside the realm of science.

Scientists all too often generate hypotheses that cannot be tested by experiments whose results have the potential to show that the idea is false.

Three types of experiments proposed by scientists

• Type 1 experiments are the most powerful. Type 1 experimental outcomes include a possible negative outcome that would falsify, or refute, the working hypothesis. It is one or the other.
• Type 2 experiments are very common, but lack punch. A positive result in a type 2 experiment is consistent with the working hypothesis, but the negative or null result does not address the validity of the hypothesis because there are many explanations for the negative result. These call for extrapolation and semantics.
• Type 3 experiments are those experiments whose results may be consistent with the hypothesis, but are useless because regardless of the outcome, the findings are also consistent with other models. In other words, every result isn’t informative.

Formulate hypotheses in such a way that you can prove or disprove them by direct experiment.

Science advances by conducting the experiments that could potentially disprove our hypotheses.

Increase the efficiency and impact of your science by testing clear hypotheses with well-designed experiments.

For more on the challenges in experimental science , read our review of Richard Harris’  Rigor Mortis: How Sloppy Science Creates Worthless Cures, Crushes Hope, and Wastes Billions.

A researcher’s look at Rigor Mortis: Are motivators and incentives to find a cure hurting scientific research?

St. Jude researchers take a look at Rigor Mortis, Richard Harris’ exposé of how the drive to find results hampers scientific progress.

About the author

Charles Rock, PhD, was a member of the Department of Infectious Diseases and later the Department of Host-Microbe Interactions at St. Jude Children’s Research Hospital until his passing in 2023.  Learn about Dr. Rock's research career .

Related Posts

Math behind the medicine

The long and the short of the cell antenna

Exploring biomolecular condensate networks’ internal structure

Stay ahead of the curve.

• Search Menu
• Sign in through your institution
• Volume 2024, Issue 1, 2024 (In Progress)
• Volume 2023, Issue 1, 2023
• Special Issues
• Author Guidelines
• Journal Policies
• Submission Site
• Open Access
• Why publish with this journal?
• Call For Papers
• About Neuroscience of Consciousness
• About the Association for the Scientific Study of Consciousness
• Editorial Board
• Advertising and Corporate Services
• Journals Career Network
• Self-Archiving Policy
• Journals on Oxford Academic
• Books on Oxford Academic

Article Contents

Introduction, formal description of testing theories, the substitution argument, inference and prediction data are strictly dependent, supplementary materials, acknowledgments.

• < Previous

Falsification and consciousness

• Article contents
• Figures & tables
• Supplementary Data

Johannes Kleiner, Erik Hoel, Falsification and consciousness, Neuroscience of Consciousness , Volume 2021, Issue 1, 2021, niab001, https://doi.org/10.1093/nc/niab001

• Permissions Icon Permissions

The search for a scientific theory of consciousness should result in theories that are falsifiable. However, here we show that falsification is especially problematic for theories of consciousness. We formally describe the standard experimental setup for testing these theories. Based on a theory’s application to some physical system, such as the brain, testing requires comparing a theory’s predicted experience (given some internal observables of the system like brain imaging data) with an inferred experience (using report or behavior). If there is a mismatch between inference and prediction, a theory is falsified. We show that if inference and prediction are independent, it follows that any minimally informative theory of consciousness is automatically falsified. This is deeply problematic since the field’s reliance on report or behavior to infer conscious experiences implies such independence, so this fragility affects many contemporary theories of consciousness. Furthermore, we show that if inference and prediction are strictly dependent, it follows that a theory is unfalsifiable. This affects theories which claim consciousness to be determined by report or behavior. Finally, we explore possible ways out of this dilemma.

Successful scientific fields move from exploratory studies and observations to the point where theories are proposed that can offer precise predictions. Within neuroscience, the attempt to understand consciousness has moved out of the exploratory stage and there are now a number of theories of consciousness capable of predictions that have been advanced by various authors ( Koch et al. 2016 ).

At this point in the field’s development, falsification has become relevant. In general, scientific theories should strive to make testable predictions ( Popper 1968 ). In the search for a scientific theory of consciousness, falsifiability must be considered explicitly as it is commonly assumed that consciousness itself cannot be directly observed, instead it can only be inferred based off of report or behavior.

Contemporary neuroscientific theories of consciousness first began to be proposed in the early 1990s ( Crick 1994 ). Some have been based directly on neurophysiological correlates, such as proposing that consciousness is associated with neurons firing at a particular frequency ( Crick and Koch 1990 ) or activity in some particular area of the brain like the claustrum ( Crick and Koch 2005 ). Other theories have focused more on the dynamics of neural processing, such as the degree of recurrent neural connectivity ( Lamme 2006 ). Others yet have focused on the “global workspace” of the brain, based on how signals are propagated across different brain regions ( Baars 1997 ). Specifically, Global Neuronal Workspace (GNW) theory claims that consciousness is the result of an “avalanche” or “ignition” of widespread neural activity created by an interconnected but dispersed network of neurons with long-range connections ( Sergent and Dehaene 2004 ).

Another avenue of research strives to derive a theory of consciousness from analysis of phenomenal experience. The most promising example thereof is Integrated Information Theory (IIT) ( Tononi 2004 , 2008 ; Oizumi et al. 2014 ). Historically, IIT is the first well-formalized theory of consciousness. It was the first (and arguably may still be the lone) theory that makes precise quantitative predictions about both the contents and level of consciousness ( Tononi 2004 ). Specifically, the theory takes the form of a function, the input of which is data derived from some physical system’s internal observables, while the output of this function is predictions about the contents of consciousness (represented mathematically as an element of an experience space) and the level of consciousness (represented by a scalar value Φ ⁠ ).

Both GNW and IIT have gained widespread popularity, sparked a general interest in consciousness, and have led to dozens if not hundreds of new empirical studies ( Massimini et al. 2005 ; Del Cul et al. 2007 ; Dehaene and Changeux 2011 ; Gosseries et al. 2014 ; Wenzel et al. 2019 ). Indeed, there are already significant resources being spent attempting to falsify either GNW or IIT in the form of a global effort pre-registering predictions from the two theories so that testing can be conducted in controlled circumstances by researchers across the world ( Ball 2019 ; Reardon 2019 ). We therefore often refer to both GNW and IIT as exemplar theories within consciousness research and show how our results apply to both. However, our results and reasoning apply to most contemporary theories, e.g. ( Lau and Rosenthal 2011 ; Chang et al. 2019 ), particularly in their ideal forms. Note that we refer to both “theories” of consciousness and also “models” of consciousness, and use these interchangeably ( Seth 2007 ).

Due to IIT’s level of formalization as a theory, it has triggered the most in-depth responses, expansions, and criticisms ( Cerullo 2015 ; Bayne 2018 ; Mediano et al. 2019 ; Kleiner and Tull 2020 ) since well-formalized theories are much easier to criticize than nonformalized theories. Recently, one criticism levied against IIT was based on how the theory predicts feedforward neural networks have zero Φ and recurrent neural networks have nonzero Φ ⁠ . Since a given recurrent neural network can be “unfolded” into a feedforward one while preserving its output function, this has been argued to render IIT outside the realm of science ( Doerig et al. 2019 ). Replies have criticized the assumptions which underlie the derivation of this argument ( Tsuchiya et al. 2019 ; Kleiner 2020 ).

Here, we frame and expand concerns around testing and falsification of theories by examining a more general question: what are the conditions under which theories of consciousness (beyond IIT alone) can be falsified? We outline a parsimonious description of theory testing with minimal assumptions based on first principles. In this agnostic setup, falsifying a theory of consciousness is the result of finding a mismatch between the inferred contents of consciousness (usually based on report or behavior) and the contents of consciousness as predicted by the theory (based on the internal observables of the system under question).

This mismatch between prediction and inference is critical for an empirically meaningful scientific agenda, because a theory’s prediction of the state and content of consciousness on its own cannot be assessed. For instance, imagine a theory that predicts (based on internal observables like brain dynamics) that a subject is seeing an image of a cat. Without any reference to report or outside information, there can be no falsification of this theory, since it cannot be assessed whether the subject was actually seeing a “dog” rather than “cat.” Falsifying a theory of consciousness is based on finding such mismatches between reported experiences and predictions.

In the following work, we formalize this by describing the prototypical experimental setup for testing a theory of consciousness. We come to a surprising conclusion: a widespread experimental assumption implies that most contemporary theories of consciousness are already falsified.

The assumption in question is the independence of an experimenter’s inferences about consciousness from a theory’s predictions. To demonstrate the problems this independence creates for contemporary theories, we introduce a “substitution argument.” This argument is based on the fact that many systems are equivalent in their reports (e.g. their outputs are identical for the same inputs), and yet their internal observables may differ greatly. This argument constitutes both a generalization and correction of the “unfolding argument” against IIT presented in Doerig et al. (2019) . Examples of such substitutions may involve substituting a brain with a Turing machine or a cellular automaton since both types of systems are capable of universal computation ( Turing 1937 ; Wolfram 1984 ) and hence may emulate the brain’s responses, or replacing a deep neural network with a single-layer neural network, since both types of networks can approximate any given function ( Hornik et al. 1989 ; Schäfer and Zimmermann 2006 ).

Crucially, our results do not imply that falsifications are impossible. Rather, they show that the independence assumption implies that whenever there is an experiment where a theory’s predictions based on internal observables and a system’s reports agree, there exists also an actual physical system that falsifies the theory. One consequence is that the “unfolding argument” concerning IIT ( Doerig et al. 2019 ) is merely a small subset of a much larger issue that affects all contemporary theories which seek to make predictions about experience off of internal observables. Our conclusion shows that if independence holds, all such theories come falsified a priori . Thus, instead of putting the blame of this problem on individual theories of consciousness, we show that it is due to issues of falsification in the scientific study of consciousness, particularly the field’s contemporary usage of report or behavior to infer conscious experiences.

A simple response to avoid this problem is to claim that report and inference are not independent. This is the case, e.g., in behaviorist theories of consciousness, but arguably also in Global Workspace Theory ( Baars 2005 ), the “attention schema” theory of consciousness ( Graziano and Webb 2015 ) or “fame in the brain” (Dennett 1991) proposals. We study this answer in detail and find that making a theory’s predictions and an experimenter’s inferences strictly dependent leads to pathological unfalsifiability.

Our results show that if the independence of prediction and inference holds true, as in contemporary cases where report about experiences is relied upon, it is likely that no current theory of consciousness is correct. Alternatively, if the assumption of independence is rejected, theories rapidly become unfalsifiable. While this dilemma may seem like a highly negative conclusion, we take it to show that our understanding of testing theories of consciousness may need to change to deal with these issues.

Here, we provide a formal framework for experimentally testing a particular class of theories of consciousness. The class we consider makes predictions about the conscious experience of physical systems based on observations or measurements . This class describes many contemporary theories, including leading theories such as IIT ( Oizumi et al. 2014 ), GNW Theory ( Dehaene and Changeux 2004 ), Predictive Processing [when applied to account for conscious experience ( Hohwy 2012 ; Hobson et al. 2014 ; Seth 2014 ; Clark 2019 ; Dolega and Dewhurst 2020 )], or Higher Order Thought Theory ( Rosenthal 2002 ). These theories may be motivated in different ways, or contain different formal structures, such as e.g., the ones of category theory ( Tsuchiya et al. 2016 ). In some cases, contemporary theories in this class may lack the specificity to actually make precise predictions in their current form. Therefore, the formalisms we introduce may sometimes describe a more advanced form of a theory, one that can actually make predictions.

In the following section, we introduce the necessary terms to define how to falsify this class of theories: how the measurement of a physical system’s observables results in datasets (Experiments section), how a theory makes use of those datasets to offer predictions about consciousness (Predictions section), how an experimenter makes inferences about a physical system’s experiences (Inferences section), and finally how falsification of a theory occurs when there is a mismatch between a theory’s prediction and an experimenter’s inference (Falsification section). In Summary section, we give a summary of the introduced terms. In subsequent sections, we explore the consequences of this setup, such as how all contemporary theories are already falsified if the data used by inferences and predictions are independent, and also how theories are unfalsifiable if this is changed to a strict form of dependency.

Experiments

All experimental attempts to either falsify or confirm a member of the class of theories we consider begin by examining some particular physical system which has some specific physical configuration, state, or dynamics, p . This physical system is part of a class P of such systems which could have been realized, in principle, in the experiment. For example, in IIT, the class of systems P may be some Markov chains, set of logic gates, or neurons in the brain, and every p ∈ P denotes that system being in a particular state at some time t . On the other hand, for GNW, P might comprise the set of long-range cortical connections that make up the global workspace of the brain, with p being the activity of that global workspace at that time.

Note that obs describes the experiment, the choice of observables, and all conditions during an experiment that generates the dataset o necessary to apply the theory, which may differ from theory to theory, such as interventions in the case of IIT. In all realistic cases, the correspondence obs is likely quite complicated since it describes the whole experimental setup. For our argument, it simply suffices that this mapping exists, even if it is not known in detail.

It is also worth noting here that all leading neuroscientific theories of consciousness, from IIT to GNW, assume that experiences are not observable or directly measurable when applying the theory to physical systems. That is, experiences themselves are never identified or used in obs but are rather inferred based on some dataset o that contains report or other behavioral indicators.

Next, we explore how the datasets in O are used to make predictions about the experience of a physical system.

Predictions

A theory of consciousness makes predictions about the experience of some physical system in some configuration, state, or dynamics, p , based on some dataset o . To this end, a theory carries within its definition a set or space E whose elements correspond to various different conscious experiences a system could have. The interpretation of this set varies from theory to theory, ranging from descriptions of the level of conscious experience in early versions of IIT, descriptions of the level and content of conscious experience in contemporary IIT ( Kleiner and Tull 2020 ), or the description only of whether a presented stimuli is experienced in GNW or HOT. We sometimes refer to elements e of E simply as experiences .

Shown in Fig. 1 is the full set of terms needed to formally define how most contemporary theories of consciousness make predictions about the experience. So far, what we have said is very general. Indeed, the force and generalizability of our argument comes from the fact that we do not have to define pred explicitly for the various models we consider. It suffices that it exists, in some form or the other, for the models under consideration.

We assume that an experimental setup apt for a particular model of consciousness has been chosen for some class of physical systems P , wherein p ∈ P represents the dynamics or configurations of a particular physical system. O then denotes all datasets that can arise from observations or measurements on P . Measuring the observables of p maps to datasets o ∈ O ⁠ , which is denoted by the obs correspondence. E represents the mathematical description of experience given by the theory or model of consciousness under consideration. In the simplest case, this is just a set whose elements indicate whether a stimulus has been perceived consciously or not, but far more complicated structures can arise (e.g. in IIT). The correspondence pred describes the process of prediction as a map from O to E .

It is crucial to note that predicting states of consciousness alone does not suffice to test a model of consciousness. Some have previously criticized theories of consciousness, IIT in particular, just based off of their counter-intuitive predictions. An example is the criticism that relatively simply grid-like networks have high Φ ( Aaronson 2014 ; Tononi 2014). However, debates about counter-intuitive predictions are not meaningful by themselves, since pred alone does not contain enough information to say whether a theory is true or false. The most a theory could be criticized for is either not fitting our own phenomenology or not being parsimonious enough, neither of which are necessarily violated by counter-intuitive predictions. For example, it may actually be parsimonious to assume that many physical systems have consciousness ( Goff 2017 ). That is, speculation about acceptable predictions by theories of consciousness must implicitly rely on a comparative reference to be meaningful, and speculations that are not explicit about their reference are uninformative.

As discussed in the previous section, a theory is unfalsifiable given just predictions alone, and so pred must be compared to something else. Ideally, this would be the actual conscious experience of the system under investigation. However, as noted previously, the class of theories we focus on here assumes that experience itself is not part of the observables. For this reason, the experience of a system must be inferred separately from a theory’s prediction to create a basis of comparison. Most commonly, such inferences are based on reports . For instance, an inference might be based on an experimental participant reporting on the switching of some perceptually bistable image ( Blake et al. 2014 ) or on reports about seen vs. unseen images in masking paradigms ( Alais et al. 2010 ).

It has been pointed out that report in a trial may interfere with the actual isolation of consciousness, and there has recently been the introduction of so-called “no-report paradigms” ( Tsuchiya et al. 2015 ). In these cases, report is first correlated to some autonomous phenomenon like optokinetic nystagmus (stereotyped eye movement), and then the experimenter can use this instead of the subject’s direct reports to infer their experiences. Indeed, there can even be simpler cases where report is merely assumed: e.g., that in showing a red square, a participant will experience a red square without necessarily asking the participant since previously that participant has proved compos mentis. Similarly, in cases of nonhumans incapable of verbal report, “report” can be broadly construed as behavior or output.

Defining inf as a function means that we assume that for every experimental dataset o , one single experience in E is inferred during the experiment. Here, we use a function instead of a correspondence for technical and formal ease, which does not affect our results: if two correspondences to the same space are given, one of them can be turned into a function. (If inf is a correspondence, one defines a new space E ′ by E ′ : = { inf ( o ) | o ∈ O } ⁠ . Every individual element of this space describes exactly what can be inferred from one dataset o ∈ O ⁠ , so that inf ′ : O → E ′ is a function. The correspondence obs is then redefined, for every e ′ ∈ E ′ ⁠ , by the requirement that e ′ ∈ obs ′ ( o ) iff e ∈ obs ( o ) for some e ∈ e ′ ⁠ .) The inf function is flexible enough to encompass both direct report, no-report, input/output analysis, and also assumed-report cases. It is a mapping that describes the process of inferring the conscious experience of a system from data recorded in the experiments. Both inf and pred are depicted in Fig. 2 .

Two maps are necessary for a full experimental setup, one that describes a theory’s predictions about experience ( pred ), another that describes the experimenter’s inference about it ( inf ). Both map from a dataset o ∈ O collected in an experimental trail to some subset of experiences described by the model, E .

It is worth noting that we have used here the same class O as in the definition of the prediction mapping pred above. This makes sense because the inference process also uses data obtained in experimental trials, such as reports by a subject. So both pred and inf can be described to operate on the same total dataset measured, even though they usually use different parts of this dataset (cf. below).

Falsification

This definition can be spelled out in terms of individual components of E . To this end, for any given dataset o ∈ O ⁠ , let e r : = inf ( o ) denote the experience that is being inferred, and let e p ∈ obs ( o ) be one of the experiences that is predicted based off of some dataset. Then (2) simply states that we have e p ≠ e r for all possible predictions e p   ∈ obs ( o ) ⁠ . None of the predicted states of experience is equal to the inferred experience.

What does Equation (2) mean? There are two cases which are possible. Either, the prediction based on the theory of consciousness is correct, and the inferred experience is wrong. Or the prediction is wrong, so that in this case the model would be falsified. In short: either the prediction process or the inference process is wrong.

We remark that if there is a dataset o on which the inference procedure inf or the prediction procedure pred cannot be used, then this dataset cannot be used in falsifying a model of consciousness. Thus, when it comes to falsifications, we can restrict to datasets o for which both procedures are defined.

In order to understand in more detail what is going on if (2) holds, we have to look into a single dataset o ∈ O ⁠ . This will be of use later.

o i For a chosen dataset o ∈ O ⁠ , we denote the part of the dataset which is used for the prediction process by o i (for “internal” data). This can be thought of as data about the internal workings of the system. We call o i the prediction data in o . o r For a chosen dataset o ∈ O ⁠ , we denote the part of the dataset which is used for inferring the state of experience by o r (for “report” data). We call it the inference data in o .

Note that in both cases, the subscript can be read similarly as the notation for restricting a set. We remark that a different kind of prediction could be considered as well, where one makes use of the inverse of pred ⁠ . In Appendix B, we prove that this is in fact equivalent to the case considered here, so that Definition 2.1 indeed covers the most general situation.

This figure represents the same setup as Fig. 2 . The left circle depicts one single dataset o . o i (orange) is the part of the dataset used for prediction. o r (green) is the part of the dataset used for inferring the state of experience. Usually the green area comprises verbal reports or button presses, whereas the orange area comprises the data obtained from brain scans. The right circle depicts the experience space E of a theory under consideration. e p denotes a predicted experience while e r denotes the inferred experience. Therefore, in total, to represent some specific experimental trial we use p ∈ P ,   o ∈ O ,   e r ∈ E and e p ∈ E ⁠ , where e p ∈ pred ( o ) ⁠ .

P denotes a class of physical systems that could have been tested, in principle, in the experiment under consideration, each in various different configurations. In most cases, every p ∈ P thus describes a physical system in a particular state, dynamical trajectory, or configuration. obs is a correspondence which contains all details on how the measurements are set up and what is measured. It describes how measurement results (datasets) are determined by a system configuration under investigation. This correspondence is given, though usually not explicitly known, once a choice of measurement scheme has been made. O is the class of all possible datasets that can result from observations or measurements of the systems in the class P . Any single experimental trial results in a single dataset o ∈ O ⁠ , whose data are used for making predictions based on the theory of consciousness and for inference purposes. pred describes the process of making predictions by applying some theory of consciousness to a dataset o . It is therefore a mapping from O to E . E denotes the space of possible experiences specified by the theory under consideration. The result of the prediction is a subset of this space, denoted as pred ( o ) ⁠ . Elements of this subset are denoted by e i and describe predicted experiences. inf describes the process of inferring a state of experience from some observed data, e.g., verbal reports, button presses or using no-report paradigms. Inferred experiences are denoted by e r .

Substitutions are changes of physical systems (i.e. the substitution of one for another) that leave the inference data invariant, but may change the result of the prediction process. A specific case of substitution, the unfolding of a reentrant neural network to a feedforward one, was recently applied to IIT to argue that IIT cannot explain consciousness ( Doerig et al. 2019 ).

Here, we show that, in general, the contemporary notion of falsification in the science of consciousness exhibits this fundamental flaw for almost all contemporary theories, rather than being a problem for a particular theory. This flaw is based on the independence between the data used for inferences about consciousness (like reports) and the data used to make predictions about consciousness. We discuss various responses to this flaw in Objections section.

We begin by defining what a substitution is in Substitutions section, show that it implies falsifications in Substitutions imply falsifications section and analyze the particularly problematic case of universal substitutions in Universal substitutions imply complete falsification section. In When does a universal substitution exist? section, we prove that universal substitutions exist if prediction and inference data are independent and give some examples of already-known cases.

Substitutions

In words, a substitution requires there to be a transformation S which keeps the inference data constant but changes the prediction of the system. So much in fact that the prediction of the original configuration p and of the transformed configuration S ( p ) are fully incompatible, i.e. there is no single experience e which is contained in both predictions. Given some inference data o r , an o r -substitution then requires this to be the case for at least one system configuration p that gives this inference data. In other words, the transformation S is such that for at least one p , the predictions change completely , while the inference content o r is preserved.

A pictorial definition of substitutions is given in Fig. 4 . We remark that if pred and obs were functions, so that pred   ○   obs ( p ) only contained one element, Equation (4) would be equivalent to pred ( obs ( p ) ) ≠ pred ( obs ( S ( p ) ) ) ⁠ .

We will find below that the really problematic case arises if there is an o r -substitution for every possible inference content o r . We refer to this case as a universal substitution.

This picture illustrates substitutions. Assume that some dataset o with inference content o r is given. A substitution is a transformation S of physical systems which leaves the inference content o r invariant but which changes the result of the prediction process. Thus whereas p and S ( p ) have the same inference content o r , the prediction content of experimental datasets is different; different in fact to such an extent that the predictions of consciousness based on these datasets are incompatible (illustrated by the nonoverlapping gray circles on the right). Here, we have used that by definition of P o r ⁠ , every p ˜ ∈ P o r yields at least one dataset o ′ with the same inference content as o and have identified o and o ′ in the drawing.

There is a universal substitution if there is an o r -substitution   S o r : P o r → P o r for every o r .

We recall that according to the notation introduced in Falsification section, the inference content of any dataset o ∈ O is denoted by o r (adding the subscript r ). Thus, the requirement is that there is an o r -substitution S o r : P o r → P o r for every inference data that can pertain in the experiment under consideration (for every inference data that is listed in O ⁠ ). The subscript o r of S o r indicates that the transformation S in Definition 3.1 can be chosen differently for different o r . Definition 3.2 does not require there to be one single transformation that works for all o r .

Substitutions imply falsifications

If there is a o r -substitution, there is a falsification at some   o ∈ O ⁠ .

Since, however, o r = o r ′ ⁠ , we have inf ( o ) = inf ( o ′ ) ⁠ . Thus we have either inf ( o ) ∈ pred ( o ) or inf ( o ′ ) ∈ pred ( o ′ ) ⁠ , or both. Thus there is either a falsification at o , a falsification at o ′ ⁠ , or both. □

The last lemma shows that if there are substitutions, then there are necessarily falsifications. This might, however, not be considered too problematic, since it could always be the case that the model is right whereas the inferred experience is wrong. Inaccessible predictions are not unusual in science. A fully problematic case only pertains for universal substitutions, i.e., if there is an o r -substitution for every inference content o r that can arise in an experiment under consideration.

Universal substitutions imply complete falsification

There is an o r -falsification if there is a falsification for some   o ∈ O which has inference content o r .

This definition is weaker than the original definition, because among all datasets which have inference content o r , only one needs to exhibit a falsification. Using this notion, the next lemma specifies the exact relation between substitutions and falsifications.

If there is an o r -substitution, there is an o r -falsification.

Proof. This lemma follows directly from the proof of Lemma 3.3 because the datasets o and o ′ used in that proof both have inference content o r . □

This finally allows us to show our first main result. It shows that if a universal substitution exists, the theory of consciousness under consideration is falsified. We explain the meaning of this proposition after the proof.

If there is a universal substitution, there is an o r -falsification for all possible inference contents o r .

Proof. By definition of universal substitution, there is an o r -substitution for every o r . Thus, the claim follows directly from Lemma 3.5. □

where we have slightly abused notation in writing inf ( o r ) instead of inf ( o ) for clarity. This implies that one of two cases needs to pertain: either at least one of the inferred experiences inf ( o r ) is correct, in which case the corresponding prediction is wrong and the theory needs to be considered falsified. The only other option is that for all inference contents o r , the prediction pred ( o ) is correct, which qua (6) implies that no single inference inf ( o r ) points at the correct experience, so that the inference procedure is completely wrong. This shows that Proposition 3.6 can equivalently be stated as follows.

If there is a universal substitution, either every single inference operation is wrong or the theory under consideration is already falsified.

Next, we discuss under which circumstances a universal substitution exists.

When does a universal substitution exist?

In the last section, we have seen that if a universal substitution exists, this has strong consequences. In this section, we discuss under what conditions universal substitutions exist.

Theories need to be minimally informative

We have taken great care above to make sure that our notion of prediction is compatible with incomplete or noisy datasets. This is the reason why pred is a correspondence, the most general object one could consider. For the purpose of this section, we add a gentle assumption which restricts pred slightly: we assume that every prediction carries at least a minimal amount of information. In our case, this means that for every prediction pred ( o ) ⁠ , there is at least one other prediction pred ( o ′ ) which is different from pred ( o ) ⁠ . Put in simple terms, this means that we do not consider theories of consciousness which have only a single prediction.

Using this, we can state our minimal information assumption in a way that is compatible with the general setup displayed in Fig. 2 :

Inference and prediction data are independent

Here, we use the same shorthand as in (3). For example, the requirement o i ∈ obs ( p ) is a shorthand for there being an o ∈ obs ( p ) which has prediction content o i . The variation ν in this definition is a variation in P , which describes physical systems which could, in principle, have been realized in an experiment (cf. Summary section). We note that a weaker version of this definition can be given which still implies our results below, cf. Appendix A. Note that if inference and prediction data are not independent, e.g., because they have a common cause, problems of tautologies loom large, cf. Objections section. Throughout the text, we often refer to Definition 3.8 simply as “independence.”

Universal substitutions exist

If inference and prediction data are independent, universal substitutions exist.

Proof. To show that a universal substitution exists, we need to show that for every o ∈ O ⁠ , an o r -substitution exists (Definition 3.1). Thus assume that an arbitrary o ∈ O is given. The minimal information assumption guarantees that there is an o ′ such that Equation (8) holds. As before, we denote the prediction content of o and o ′ by o i and o i ′ ⁠ , respectively, and the inference content of o by o r .

The intuition behind this proof is very simple. In virtue of our assumption that theories of consciousness need to be minimally informative, for any dataset o , there is another dataset o ′ which makes a nonoverlapping prediction. But in virtue of inference and prediction data being independent, we can find a variation that changes the prediction content as prescribed by o and o ′ but keeps the inference content constant. This suffices to show that there exists a transformation S as required by the definition of a substitution.

Combining this result with Proposition 3.7, we finally can state our main theorem.

If inference and prediction data are independent, either every single inference operation is wrong or the theory under consideration is already falsified.

Proof. The theorem follows by combining Propositions 3.9 and 3.7. □

In the next section, we give several examples of universal substitutions, before discussing various possible responses to our result in Objections section.

Examples of data independence

Our main theorem shows that testing a theory of consciousness will necessarily lead to its falsification if inference and prediction data are independent (Definition 3.8), and if at least one single inference can be trusted (Theorem 3.10). In this section, we give several examples that illustrate the independence of inference and prediction data. We take report to mean output, behavior, or verbal report itself and assume that prediction data derives from internal measurements.

Artificial neural networks . ANNs, particularly those trained using deep learning, have grown increasingly powerful and capable of human-like performance ( LeCun et al. 2015 ; Bojarski et al. 2016 ). For any ANN, report (output) is a function of node states. Crucially, this function is noninjective, i.e., some nodes are not part of the output. For example, in deep learning, the report is typically taken to consist of the last layer of the ANN, while the hidden layers are not taken to be part of the output. Correspondingly, for any given inference data, one can construct a ANN with arbitrary prediction data by adding nodes, changing connections and changing those nodes which are not part of the output. Put differently, one can always substitute a given ANN with another with different internal observables but identical or near-identical reports. From a mathematical perspective, it is well-known that both feedforward ANNs and recurrent ANNs can approximate any given function ( Hornik et al. 1989 ; Schäfer and Zimmermann 2007 ). Since reports are just some function, it follows that there are viable universal substitutions.

A special case thereof is the unfolding transformation considered in Doerig et al. (2019) in the context of IIT. The arguments in this article constitute a proof of the fact that for ANNs, inference and prediction data are independent (Definition 3.8). Crucially, our main theorem shows that this has implications for all minimally informative theories of consciousness. A similar result (using a different characterization of theories of consciousness than minimally informative) has been shown in Kleiner (2020) .

Universal computers . Turing machines are extremely different in architecture than ANNs. Since they are capable of universal computation ( Turing 1937 ), they should provide an ideal candidate for a universal substitution. Indeed, this is exactly the reasoning behind the Turing test of conversational artificial intelligence ( Turing 1950 ). Therefore, if one believes it is possible for a sufficiently fast Turing machine to pass the Turing test, one needs to accept that substitutions exist. Notably, Turing machines are just one example of universal computation, and there are other instances of different parameter spaces or physical systems that are capable thereof, such as cellular automata ( Wolfram 1984 ).

Universal intelligences . There are models of universal intelligence that allow for maximally intelligent behavior across any set of tasks ( Hutter 2003 ). For instance, consider the AIXI model, the gold-standard for universal intelligence, which operates via Solomonoff induction ( Solomonoff 1964 ; Hutter 2004 ). The AIXI model generates an optimal decision making over some class of problems, and methods linked to it have already been applied to a range of behaviors, such as creating “AI physicists” ( Wu and Tegmark 2019 ). Its universality indicates it is a prime candidate for universal substitutions. Notably, unlike a Turing machine, it avoids issues of precisely how it is accomplishing universal substitution of report, since the algorithm that governs the AIXI model behavior is well-described and “relatively” simple.

The above are all real and viable classes of systems that are used everyday in science and engineering which all provide different viable universal substitutions if inferences are based on reports or outputs. They show that in normal experimental setups such as the ones commonly used in neuroscientific research into consciousness ( Frith et al. 1999 ), inference and prediction data are indeed independent, and dependency is not investigated nor properly considered. It is always possible to substitute the physical system under consideration with another that has different internal observables, and therefore different predictions, but similar or identical reports. Indeed, recent research in using the work introduced in this work shows that even different spatiotemporal models of a system can be substituted for one another, leading to falsification ( Hanson and Walker 2020 ). We have not considered possible but less reasonable examples of universal substitutions, like astronomically large look-up ledgers of reports.

As an example of our Main Theorem 3.10, we consider the case of IIT. Since the theory is normally applied in Boolean networks, logic gates, or artificial neural networks, one usually takes report to mean “output.” In this case, it has already been proven that systems with different internal structures and hence different predicted experiences, can have identical input/output (and therefore identical reports or inferences about report) ( Albantakis and Tononi 2019 ). To take another case: within IIT it has already been acknowledged that a Turing machine may have a wildly different predicted contents of consciousness for the same behavior or reports ( Koch 2019 ). Therefore, data independence during testing has already been shown to apply to IIT under its normal assumptions.

An immediate response to our main result showing that many theories suffer from a priori falsification would be to claim that it offers support of theories which define conscious experience in terms of what is accessible to report. This is the case, e.g., for behaviorist theories of consciousness but might arguably also be the case for some interpretations of global workspace theory or fame in the brain proposals. In this section, we show that this response is not valid, as theories of this kind, where inference and prediction data are strictly dependent, are unfalsifiable.

In order to analyze this case, we first need to specifically outline how theories can be pathologically unfalsifiable. Clearly, the goal of the scientific study as a whole is to find, eventually, a theory of consciousness that are empirically adequate and therefore corroborated by all experimental evidence. Therefore, not being falsified in experiments is a necessary condition (though not sufficient) any purportedly “true” theory of consciousness needs to satisfy. Therefore, not being falsifiable by the set of possible experiments per se is not a bad thing. We seek to distinguish this from cases of unfasifiability due to pathological assumptions that underlie a theory of consciousness, assumptions which render an experimental investigation meaningless. Specifically, a pathological dependence between inferences and predictions leads to theories which are unfalsifiable.

Intuitively, in terms of possible worlds semantics, O describes the datasets which could appear, for the type of experiment under consideration, in the actual world. O ¯ ⁠ , in contrast, describes the datasets which could appear in this type of experiment in any possible world. For example, O ¯ contains datasets which can only occur if consciousness attaches to the physical in a different way than it actually does in the actual word.

By construction, O is a subset of O ¯ ⁠ , which describes which among the possible datasets actually arises across experimental trials. Hence, O also determines which theory of consciousness is compatible with (i.e. not falsified by) experimental investigation. However, O ¯ defines all possible datasets independent of any constraint by real empirical results, i.e., all possible imaginable datasets.

A theory of consciousness which does not have a falsification over   O ¯ is empirically unfalsifiable.

Here, we use the term “empirically unfalsifiable” to highlight and refer to the pathological notion of unfalsifiability. Intuitively speaking, a theory which satisfies this definition appears to be true independently of any experimental investigation, and without the need for any such investigation. Using O ¯ ⁠ , we can also define the notion of strict dependence in a useful way.

Inference and prediction data are strictly dependent if there is a function f such that for any   o ∈ O ¯ , we have   o i = f ( o r ) ⁠ .

This definition says that there exists a function f which for every possible inference data o r allows to deduce the prediction data o i . We remark that the definition refers to O ¯ and not O ⁠ , as the dependence of inference and prediction considered here holds by assumption and is not simply asserting a contingency in nature.

Here, f is simply the restriction function. This arguably applies to global workspace theory ( Baars 2005 ), the “attention schema” theory of consciousness ( Graziano and Webb 2015 ) or “fame in the brain” (Dennett 1991) proposals.

In all these cases, consciousness is generated by—and hence needs to be predicted via—what is accessible to report or output. In terms of Block’s distinction between phenomenal consciousness and access consciousness ( Block 1996 ), Equation (10) holds true whenever a theory of consciousness is under investigation where access consciousness determines phenomenal consciousness.

Our second main theorem is the following.

If a theory of consciousness implies that inference and prediction data are strictly dependent, then it is either already falsified or empirically unfalsifiable.

Crucially, here, o i and o r ′ do not have to be part of the same dataset o . Combined with Definition 2.1, we conclude that there is a falsification over O ¯ if for some ( o i , o r ′ ) ∈ O all ⁠ , we have inf ( o ) ∉ pred ( o ′ ) ⁠ , and there is a falsification over O if for some ( o i , o r ) ∈ O   exp   ⁠ , we have inf ( o ) ∉ pred ( o ) ⁠ .

This, however, is exactly O   exp   as defined in (11). Thus we conclude that if inference and prediction data are strictly dependent, O all = O   exp   necessarily holds.

Returning to the characterization of falsification in terms of O   exp   and O all above, what we have just found implies that there is a falsification over O if and only if there is a falsification over O ¯ ⁠ . Thus either there is a falsification over O ⁠ , in which case the theory is already falsified or there is no falsification over O ¯ ⁠ , in which case the theory under consideration is empirically unfalsifiable. □

The gist of this proof is that if inference and prediction data are strictly dependent, then as far as the inference and prediction contents go, O and O ¯ are the same. That is, the experiment does not add anything to the evaluation of the theory. It is sufficient to know only all possible datasets to decide whether there is a falsification. In practise, this would mean that knowledge of the experimental design (which reports are to be collected, on the one hand, which possible data a measurement device can produce, one the other) is sufficient to evaluate the theory, which is clearly at odds with the role of empirical evidence required in any scientific investigation. Thus, such theories are empirically unfalsifiable.

To give an intuitive example of the theorem, let us examine a theory that uses the information accessible to report in a system to predict conscious experience (perhaps this information is “famous” in the brain or is within some accessible global workspace). In terms of our notation, we can assume that o r denotes everything that is accessible to report, and o i denotes that part which is used by the theory to predict conscious experience. Thus, in this case we have o i ⊆ o r ⁠ . Since the predicted contents are always part of what can be reported, there can never be any mismatch between reports and predictions. However, this is not only the case for O   exp   but also, in virtue of the theory’s definition, for all possible datasets, i.e., O all ⁠ . Therefore, such theories are empirically unfalsifiable. Experiments add no information to whether the theory is true or not, and such theories are empirically uninformative or tautological.

In this section, we discuss a number of possible objections to our results.

Restricting inferences to humans only

The examples given in section 3.4.4 show that data independence holds during the usual testing setups. This is because prima facie it seems reasonable to base inferences either on report capability or intelligent behavior in a manner agnostic of the actual physical makeup of the system. Yet this entails independence, so in these cases our conclusions apply.

One response to our results might be to restrict all testing of theories of consciousness solely to humans. In our formalisms, this is equivalent to making the strength of inferences based not on reports themselves but on an underlying biological homology. Such an inf function may still pick out specific experiences via reports, but the weight of the inference is carried by homology rather than report or behavior. This would mean that the substitution argument does not significantly affect consciousness research, as reports of nonhuman systems would simply not count. Theories of consciousness, so this idea goes, would be supported by abductive reasoning from testing in humans alone.

Overall, there are strong reasons to reject this restriction of inferences. One significant issue is that this objection is equivalent to saying that reports or behavior in nonhumans carry no information about consciousness, an incredibly strong claim. Indeed, this is highly problematic for consciousness research which often uses nonhuman animal models ( Boly et al. 2013 ). For instance, cephalopods are among the most intelligent animals yet are quite distant on the tree of life from humans and have a distinct neuroanatomy, and still are used for consciousness research ( Mather 2008 ). Even in artificial intelligence research, there is increasing evidence that deep neural networks produced brain-like structures such as grid cells, shape tuning, and visual illusions, and many others ( Richards et al. 2019 ). Given these similarities, it becomes questionable why the strength of inferences should be based on homology instead of capability of report or intelligence.

What is more, restricting inferences to humans alone is unlikely to be sufficient to avoid our results. Depending on the theory under consideration, data independence might exist just in human brains alone. That is, it is probable that there are transformations (as in Equation (9) ) available within the brain wherein o r is fixed but o i varies. This is particularly true once one allows for interventions on the human brain by experimenters, such as perturbations like transcranial magnetic stimulation, which is already used in consciousness research ( Rounis et al. 2010 ; Napolitani et al. 2014 ).

For these reasons this objection does not appear viable. At minimum, it is clear that if the objection were taken seriously, it would imply significant changes to consciousness research which would make the field extremely restricted with strong a priori assumptions.

Reductio ad absurdum

Another hypothetical objection to our results is to argue that they could just as well be applied to scientific theories in other fields. If this turned out to be true, this would not imply our argument is necessarily incorrect. But, the fact that other scientific theories do not seem especially problematic with regard to falsification would generate the question of whether some assumption is illegitimately strong. In order to address this, we explain which of our assumptions is specific to theories of consciousness and would not hold when applied to other scientific theories. Subsequently, we give an example to illustrate this point.

The assumption in question is that O ⁠ , the class of all datasets that can result from observations or measurements of a system, is determined by the physical configurations in P alone. That is, every single dataset o , including both its prediction content o i and its inference content o r , is determined by p , and not by a conscious experience in E . In Fig. 2 , this is reflected in the fact that there is an arrow from P to O ⁠ , but no arrow from E to O ⁠ .

This assumption expresses the standard paradigm of testing theories of consciousness in neuroscience, according to which both the data used to predict a state of consciousness and the reports of a system are determined by its physical configuration alone. This, in turn, may be traced back to consciousness’ assumed subjective and private nature, which implies that any empirical access to states of consciousness in scientific investigations is necessarily mediated by a subject’s reports, and to general physicalist assumptions.

This is different from experiments in other natural sciences. If there are two quantities of interest whose relation is to be modeled by a scientific theory, then in all reasonable cases there are two independent means of collecting information relevant to a test of the theory, one providing a dataset that is determined by the first quantity, and one providing a dataset that is determined by the second quantity.

Theories could be based on phenomenology

Another response to the issue of independence/dependence identified here is to propose that a theory of consciousness may not have to be falsified but can be judged by other characteristics. This is reminiscent of ideas put forward in connection with String Theory, which some have argued can be judged by elegance or parsimony alone ( Carroll 2018) .

In addition to elegance and parsimony, in consciousness science, one could in particular consider a theory’s fit with phenomenology, i.e., how well a theory describes the general structure of conscious experience. Examples of theories that are constructed based on a fit with phenomenology are recent versions of IIT ( Oizumi et al. 2014 ) or any view that proposes developing theories based on isomorphisms between the structure of experiences and the structure of physical systems or processes ( Tsuchiya et al. 2019 ).

It might be suggested that phenomenological theories might be immune to aspects of the issues we outline in our results ( Negro 2020 ). We emphasize that in order to avoid our results, and indeed the need for any experimental testing at all, a theory constructed from phenomenology has to be uniquely derivable from conscious experience. However, to date, no such derivation exists, as phenomenology seems to generally underdetermine the postulates of IIT ( Bayne 2018 ; Barrett and Mediano 2019 ), and because it is unknown what the scope and nature of nonhuman experience is. Therefore, theories based on phenomenology can only confidently identify systems with human-like conscious experiences and cannot currently do so uniquely. Thus they cannot avoid the need for testing.

As long as no unique and correct derivation exists across the space of possible conscious experiences, the use of experimental tests to assess theories of consciousness, and hence our results, cannot be avoided.

Rejecting falsifiability

Another response to our findings might be to deny the importance of falsifications within the scientific methodology. Such responses may reference a Lakatosian conception of science, according to which science does not proceed by discarding theories immediately upon falsification, but instead consists of research programs built around a family of theories ( Lakatos 1980 ). These research programs have a protective belt which consists of nonessential assumptions that are required to make predictions, and which can easily be modified in response to falsifications, as well as a hard core that is immune to falsifications. Within the Lakatosian conception of science research programs are either progressive or degenerating based on whether they can “anticipate theoretically novel facts in its growth” or not ( Lakatos 1980 ).

It is important to note, however, that Lakatos does not actually break with falsificationism. This is why Lakatos description of science is often called “refined falsificationism” in philosophy of science ( Radnitzky 1991 ). Thus cases of testing theories’ predictions remain relevant in a Lakatosian view in order to distinguish between progressive and degenerating research programs. Therefore, our results generally translate into this view of scientific progress. In particular, Theorem 3.10 shows that for every single inference procedure that is taken to be valid, there exists a system for which the theory makes a wrong prediction. This implies necessarily that a research program is degenerating. That is, independence implies that there is always an available substitution that can falsify any particular prediction coming from the research program.

In this article, we have subjected the usual scheme for testing theories of consciousness to a thorough formal analysis. We have shown that there appear to be deep problems inherent in this scheme which need to be addressed.

Crucially, in contrast to other similar results ( Doerig et al. 2019 ), we do not put the blame on individual theories of consciousness, but rather show that a key assumption that is usually being made is responsible for the problems: an experimenter’s inference about consciousness and a theory’s predictions are generally implicitly assumed to be independent during testing across contemporary theories. As we formally prove, if this independence holds, substitutions or changes to physical systems are possible that falsify any given contemporary theory. Whenever there is an experimental test of a theory of consciousness on some physical system which does not lead to a falsification, there necessary exists another physical system which, if it had been tested, would have produced a falsification of that theory. We emphasize that this problem does not only affect one particular type of theory, e.g., those based on causal interactions like IIT; theorems apply to all contemporary neuroscientific theories of consciousness if independence holds.

In the second part of our results, we examine the case where independence does not hold. We show that if an experimenter’s inferences about consciousness and a theory’s predictions are instead considered to be strictly dependent, empirical unfalsifiability follows, which renders any type of experiment to test a theory uninformative. This affects all theories wherein consciousness is predicted off of reports or behavior (such as behaviorism), theories based off of input/output functions, and also theories that equate consciousness with on accessible or reportable information.

Thus, theories of consciousness seem caught between Scylla and Charybdis, requiring delicate navigation. In our opinion, there may only be two possible paths forward to avoid these dilemmas, which we briefly outline below. Each requires a revision of the current scheme of testing or developing theories of consciousness.

Lenient dependency

When combined, our main theorems show that both independence and strict dependence of inference and prediction data are problematic and thus neither can be assumed in an experimental investigation. This raises the question of whether there are reasonable cases where inference and prediction are dependent, but not strictly dependent.

A priori , in the space of possible relationships between inference and prediction data, there seems to be room for relationships that are neither independent (The substitution argument section) nor strictly dependent (Inference and prediction data are strictly dependent section). We define this relationships of this kind as cases of lenient dependency . No current theory or testing paradigm that we know of satisfies this definition. Yet cases of lenient dependency cannot be excluded to exist. Such cases would technically not be beholden to either Theorem 3.10 or Theorem 4.3.

There seems to be two general possibilities of how lenient dependencies could be built. On the one hand, one could hope to find novel forms of inference that allow to surpass the problems we have identified here. This would likely constitute a major change in the methodologies of experimental testing of theories of consciousness. On the other hand, another possibility to attain lenient dependence would be to construct theories of consciousness which yield prediction functions that are designed to explicitly have a leniently dependent link to inference functions. This would likely constitute a major change in constructing theories of consciousness.

Physics is not causally closed

Another way to avoid our conclusion is to only consider theories of consciousness which do not describe the physical as causally closed ( Kim 1998 ). That is, the presence or absence of a particular experience itself would have to make a difference to the configuration, dynamics, or states of physical systems above and beyond what would be predicted with just information about the physical system itself. If a theory of consciousness does not describe the physical as closed, a whole other range of predictions are possible: predictions which concern the physical domain itself, e.g., changes in the dynamics of the system which depend on the dynamics of conscious experience. These predictions are not considered in our setup and may serve to test a theory of consciousness without the problems we have explored here.

(A) Weak Independence

In this section, we show how Definition 3.8 can be substantially relaxed while still ensuring our results to hold. To this end, we need to introduce another bit of formalism: We assume that predictions can be compared to establish how different they are. This is the case, e.g., in IIT where predictions map to the space of maximally irreducible conceptual structures (MICS), sometimes also called the space of Q-shapes, which carries a distance function analogous to a metric ( Kleiner and Tull, 2020 ). We assume that for any given prediction, one can determine which of all those predictions that do not overlap with the given one is most similar to the latter, or equivalently which is least different . We calls this a minimally differing prediction and use it to induce a notion of minimally differing datasets below. Uniqueness is not required.

We denote by   min ( o ) those datasets in   o ⊥ whose prediction is least different from the prediction of o.

In many cases min ( o ) will only contain one dataset, but here we treat the general case where this is not so. We emphasize that the minimal information assumption guarantees that min ( o ) exists. We can now specify a much weaker version of Definition 3.8.

The difference between Definitions A.2 and 3.8 is that for a given o ∈ O ⁠ , the latter requires the transformation ν to exist for any o ′ ∈ O ⁠ , wheres the former only requires it to exist for minimally different datasets o ′ ∈ min ( o ) ⁠ . The corresponding proposition is the following.

If inference and prediction data are weakly independent, universal substitutions exist.

Proof. To show that a universal substitution exists, we need to show that for every o ∈ O ⁠ , an o r -substitution exists (Definition 3.1). Thus assume that an arbitrary o ∈ O is given and pick an o ′ ∈ min ( o ) ⁠ . As before, we denote the prediction content of o and o ′ by o i and o i ′ ⁠ , respectively, and the inference content of o by o r .

The following theorem shows that Definition A.2 is sufficient to establish the claim of Theorem 3.10.

If inference and prediction data are weakly independent, either every single inference operation is wrong or the theory under consideration is already falsified.

Proof. The theorem follows by combining Propositions A.3 and 3.7. □

(B) Inverse Predictions

When defining falsification, we have considered predictions that take as input data about the physical configuration of a system and yield as output a state of consciousness. An alternative would be to consider the inverse procedure: a prediction which takes as input a reported stated of consciousness and yields as output some constraint on the physical configuration of the system that is having the conscious experience. In this section, we discuss the second case in detail.

As before, we assume that some dataset o has been measured in an experimental trail, which contains both the inference data o r (which includes report and behavioral indicators of consciousness used in the experiment under consideration) as well as some data o i that provides information about the physical configuration of the system under investigation. For simplicity, we will also call this prediction data here. Also as before, we take into account that the state of consciousness of the system has to be inferred from o r , and again denote this inference procedure by inf ⁠ .

The case of an inverse prediction. Rather than comparing the inferred and predicted state of consciousness, one predicts the physical configuration of a system based on the system’s report and compares this with measurement results.

In terms of the notion introduced in Summary section, Equation (20) could equivalently be written as o i ∈ pred − 1 ( inf ( o r ) ) i ⁠ . The following lemma shows that there is a type-2 falsification if and only if there is a type-1 falsification. Hence all of our previous results apply as well to type-2 falsifications.

There is a type-2 falsification at o if and only if there is a type-1 falsification at o.

The former is the definition of a type-2 falsification. The latter is Equation (2) in the definition of a type-1 falsification. Hence the claim follows. □

We would like to thank David Chalmers, Ned Block, and the participants of the NYU philosophy of mind discussion group for valuable comments and discussion. Thanks also to Ryota Kanai, Jake Hanson, Stephan Sellmaier, Timo Freiesleben, Mark Wulff Carstensen, and Sofiia Rappe for feedback on early versions of the manuscript.

Conflict of interest statement . None declared.

Aaronson S . Why I am not an integrated information theorist (or, the unconscious expander). In: Shtetl-Optimized: The Blog of Scott Aaronson . eds. Radin D. , Richard D. , & Karim T. , 2014 .

Google Scholar

Google Preview

Alais D , Cass J , O’Shea RP , et al.  Visual sensitivity underlying changes in visual consciousness . Curr Biol 2010 ; 20 : 1362 – 7 .

Albantakis L , Tononi G. Causal composition: Structural differences among dynamically equivalent systems . Entropy 2019 ; 21 : 989 .

Baars BJ. In the theatre of consciousness. global workspace theory, a rigorous scientific theory of consciousness . J Conscious Stud 1997 ; 4 : 292 – 309 .

Baars BJ. Global workspace theory of consciousness: toward a cognitive neuroscience of human experience . Progr Brain Res 2005 ; 150 : 45 – 53 .

Ball P . Neuroscience readies for a showdown over consciousness ideas . Quanta Mag 2019 ; 6 .

Barrett AB , Mediano PA. The phi measure of integrated information is not well-defined for general physical systems . J Conscious Stud 2019 ; 26 : 11 – 20 .

Bayne TJ On the axiomatic foundations of the integrated information theory of consciousness . Neurosci Conscious 2018 ; 4 : 1 –8.

Blake R , Brascamp J , Heeger DJ. Can binocular rivalry reveal neural correlates of consciousness? Philos Trans R Soc B 2014 ; 369 : 20130211 .

Block N. How can we find the neural correlate of consciousness? Trends Neurosci 1996 ; 19 : 456 – 9 .

Bojarski M , Del Testa D , Dworakowski D , et al.  End to end learning for self-driving cars. arXiv preprint arXiv:1604.07316, 2016 .

Boly M , Seth AK , Wilke M , et al.  Consciousness in humans and non-human animals: recent advances and future directions . Front Psychol 2013 ; 4 : 625 .

Carroll SM . Beyond falsifiability: normal science in a multiverse. In: Why Trust a Theory? 2018 , 300 .

Cerullo MA. The problem with phi: a critique of integrated information theory . PLoS Comput Biol 2015 ; 11 : e1004286 .

Chang AY , Biehl M , Yu Y , et al.  Information closure theory of consciousness. Frontiers in Psychology 2020 ;11.

Clark A. Consciousness as generative entanglement . J Philos 2019 ; 116 : 645 – 62 .

Crick F. Astonishing Hypothesis: The Scientific Search for the Soul . New York : Simon and Schuster , 1994 .

Crick F , Koch C. Towards a neurobiological theory of consciousness. In: Seminars in the Neurosciences , Vol. 2 . Saunders Scientific Publications , 1990 , 263 – 75 .

Crick FC , Koch C. What is the function of the claustrum? Philos Trans R Soc B 2005 ; 360 : 1271 – 9 .

Dehaene S , Changeux J-P. Neural mechanisms for access to consciousness . Cogn Neurosci 2004 ; 3 : 1145 – 58 .

Dehaene S , Changeux J-P. Experimental and theoretical approaches to conscious processing . Neuron 2011 ; 70 : 200 – 27 .

Del Cul A , Baillet S , Dehaene S. Brain dynamics underlying the nonlinear threshold for access to consciousness . PLoS Biol 2007 ; 5 : e260 .

Dennett DC. Consciousness Explained . Boston : Little, Brown and Co, 1991 .

Doerig A , Schurger A , Hess K , et al.  The unfolding argument: why iit and other causal structure theories cannot explain consciousness . Conscious Cogn 2019 ; 72 : 49 – 59 .

Dołęga K , Dewhurst JE. Fame in the predictive brain: a deflationary approach to explaining consciousness in the prediction error minimization framework . Synthese 2020 ; 1 – 26 .

Frith C , Perry R , Lumer E. The neural correlates of conscious experience: an experimental framework . Trends Cogn Sci 1999 ; 3 : 105 – 14 .

Gilmore RO , Diaz MT , Wyble BA , et al.  Progress toward openness, transparency, and reproducibility in cognitive neuroscience . Ann N Y Acad Sci 2017 ; 1396 : 5 .

Goff P. Consciousness and Fundamental Reality . Oxford University Press , 2017 .

Gosseries O , Di H , Laureys S , et al.  Measuring consciousness in severely damaged brains . Annu Rev Neurosci 2014 ; 37 : 457 – 78 .

Graziano MS , Webb TW. The attention schema theory: a mechanistic account of subjective awareness . Front Psychol 2015 ; 6 : 500 .

Hanson JR , Walker SI . Formalizing falsification of causal structure theories for consciousness across computational hierarchies. arXiv preprint arXiv:2006.07390, 2020 .

Hobson AJ , et al.  Consciousness, dreams, and inference: the Cartesian theatre revisited . J Conscious Stud 2014 ; 21 : 6 – 32 .

Hohwy J. Attention and conscious perception in the hypothesis testing brain . Front Psychol 2012 ; 3 : 96 .

Hornik K , Stinchcombe M , White H. Multilayer feedforward networks are universal approximators . Neural Netw 1989 ; 2 : 359 – 66 .

Hutter M. A gentle introduction to the universal algorithmic agent {AIXI}. In Artificial General Intelligence , eds. Goertzel B., and Pennachin C., Springer. 2003 .

Hutter M. Universal Artificial Intelligence: Sequential Decisions Based on Algorithmic Probability . Berlin : Springer Science & Business Media , 2005 .

Kim J. Mind in a Physical World: An Essay on the Mind-Body Problem and Mental Causation . Cambridge : MIT Press , 1998 .

Kleiner J . Brain states matter. a reply to the unfolding argument . Conscious Cogn 2020 ; 85 : 102981 .

Kleiner J , Tull S . The mathematical structure of integrated information theory. arXiv preprint arXiv:2002.07655, 2020 .

Koch C. The Feeling of Life Itself: Why Consciousness is Widespread but can’t be Computed . Cambidge MA Boston : MIT Press , 2019 .

Koch C , Massimini M , Boly M , et al.  Neural correlates of consciousness: progress and problems . Nat Rev Neurosci 2016 ; 17 : 307 .

Lakatos I. The Methodology of Scientific Research Programmes: Volume 1: Philosophical Papers , Vol. 1 . London : Cambridge University Press, UK , 1980 .

Lamme VA. Towards a true neural stance on consciousness . Trends Cogn Sci 2006 ; 10 : 494 – 501 .

Lau H , Rosenthal D. Empirical support for higher-order theories of conscious awareness . Trends Cogn Sci 2011 ; 15 : 365 – 73 .

LeCun Y , Bengio Y , Hinton G. Deep learning . Nature 2015 ; 521 : 436 – 44 .

Massimini M , Ferrarelli F , Huber R , et al.  Breakdown of cortical effective connectivity during sleep . Science 2005 ; 309 : 2228 – 32 .

Mather JA. Cephalopod consciousness: behavioural evidence . Conscious Cogn 2008 ; 17 : 37 – 48 .

Mediano P , Seth A , Barrett A. Measuring integrated information: comparison of candidate measures in theory and simulation . Entropy 2019 ; 21 : 17 .

Napolitani M , Bodart O , Canali P , et al.  Transcranial magnetic stimulation combined with high-density EEG in altered states of consciousness . Brain Injury 2014 ; 28 : 1180 – 9 .

Negro N. Phenomenology-first versus third-person approaches in the science of consciousness: the case of the integrated information theory and the unfolding argument . Phenomenol Cogn Sci 2020 ; 19:979 – 96 .

Oizumi M , Albantakis L , Tononi G. From the phenomenology to the mechanisms of consciousness: integrated information theory 3.0 . PLoS Comput Biol 2014 ; 10 : e1003588 .

Popper K. The Logic of Scientific Discovery . New York : Harper & Row, 1968 .

Putnam H . Minds and machines. In Dimensions of Mind , ed. Hook S. , New York : New York University Press , 1960 , pp. 57 – 80 .

Radnitzky G . Review: Refined falsificationism meets the challenge from the relativist philosophy of science . Br J Philos Sci 1991 ; 42 : 273 – 284 .

Reardon S . Rival theories face off over brain’s source of consciousness . Science 2019 ; 366 : 293 – 293 .

Richards BA , Lillicrap TP , Beaudoin P , et al.  A deep learning framework for neuroscience . Nat Neurosci 2019 ; 22 : 1761 – 70 .

Rosenthal DM. How many kinds of consciousness? Conscious Cogn 2002 ; 11 : 653 – 65 .

Rounis E , Maniscalco B , Rothwell JC , et al.  Theta-burst transcranial magnetic stimulation to the prefrontal cortex impairs metacognitive visual awareness . Cogn Neurosci 2010 ; 1 : 165 – 75 .

Schäfer AM , Zimmermann HG. Recurrent neural networks are universal approximators. In: International journal of neural systems . Springer , 2007 ; 17 :253– 63 .

Sergent C , Dehaene S. Neural processes underlying conscious perception: experimental findings and a global neuronal workspace framework . J Physiol Paris 2004 ; 98 : 374 – 84 .

Seth AK. Models of consciousness . Scholarpedia 2007 ; 2 : 1328 .

Seth AK. A predictive processing theory of sensorimotor contingencies: explaining the puzzle of perceptual presence and its absence in synesthesia . Cogn Neurosci 2014 ; 5 : 97 – 118 .

Skinner BF . The behavior of organisms: an experimental analysis. Appleton-Century , Cambridge, Massachusetts : B.F. Skinner Foundation . 1938 .

Solomonoff RJ. A formal theory of inductive inference . Part I. Inform Control 1964 ; 7 : 1 – 22 .

Tononi G. An information integration theory of consciousness . BMC Neurosci 2004 ; 5 : 42 .

Tononi G. Consciousness as integrated information: a provisional manifesto . Biol Bull 2008 ; 215 : 216 – 42 .

Giulio T . Why Scott should stare at a blank wall and reconsider (or, the conscious grid). In: Shtetl-Optimized: The Blog of Scott Aaronson. Available online: http://www.scottaaronson.com/blog , 2014 .

Tsuchiya N , Wilke M , Frässle S , et al.  No-report paradigms: extracting the true neural correlates of consciousness . Trends Cogn Sci 2015 ; 19 : 757 – 70 .

Tsuchiya N , Taguchi S , Saigo H. Using category theory to assess the relationship between consciousness and integrated information theory . Neurosci Res 2016 ; 107 : 1 – 7 .

Tsuchiya N , Andrillon T , Haun A. A reply to “the unfolding argument”: beyond functionalism/behaviorism and towards a truer science of causal structural theories of consciousness. PsyArXiv, 2019 .

Turing AM. Computing machinery and intelligence. Mind 1950 ; 59 :433– 60 .

Turing AM. On computable numbers, with an application to the entscheidungsproblem . Proc Lond Math Soc 1937 ; 2 : 230 – 65 .

Wenzel M , Han S , Smith EH , et al.  Reduced repertoire of cortical microstates and neuronal ensembles in medically induced loss of consciousness . Cell Syst 2019 ; 8 : 467 – 74 .

Wolfram S. Cellular automata as models of complexity . Nature 1984 ; 311 : 419 .

Wu T , Tegmark M. Toward an artificial intelligence physicist for unsupervised learning . Phys Rev E 2019 ; 100 : 033311 .

Email alerts

Citing articles via.

• About Association for the Scientific Study of Consciousness

Affiliations

• Online ISSN 2057-2107
• Copyright © 2024 Oxford University Press
• About Oxford Academic
• Publish journals with us
• University press partners
• What we publish
• New features  
• Open access
• Institutional account management
• Rights and permissions
• Get help with access
• Accessibility
• Advertising
• Media enquiries
• Oxford University Press
• Oxford Languages
• University of Oxford

Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide

• Copyright © 2024 Oxford University Press
• Cookie settings
• Cookie policy
• Privacy policy
• Legal notice

This Feature Is Available To Subscribers Only

Sign In or Create an Account

This PDF is available to Subscribers Only

For full access to this pdf, sign in to an existing account, or purchase an annual subscription.

Does Science Need Falsifiability?

Scientists are rethinking the fundamental principle that scientific theories must make testable predictions.

Can String Theory Be Tested?

String Theory: A Theory of Everything Essay

String Theory: Science or Philosophy?

If a theory doesn’t make a testable prediction, it isn’t science.

It’s a basic axiom of the scientific method, dubbed “falsifiability” by the 20th century philosopher of science Karl Popper. General relativity passes the falsifiability test because, in addition to elegantly accounting for previously-observed phenomena like the precession of Mercury’s orbit, it also made predictions about as-yet-unseen effects—how light should bend around the Sun, the way clocks should seem to run slower in a strong gravitational field, and others that have since been borne out by experiment. On the other hand, theories like Marxism and Freudian psychoanalysis failed the falsifiability test—in Popper’s mind, at least—because they could be twisted to explain nearly any “data” about the world. As Wolfgang Pauli is said to have put it, skewering one student’s apparently unfalsifiable idea, “This isn’t right. It’s not even wrong.”

Now, some physicists and philosophers think it is time to reconsider the notion of falsifiability. Could a theory that provides an elegant and accurate account of the world around us—even if its predictions can’t be tested by today’s experiments, or tomorrow’s—still “count” as science?

“We are in various ways hitting the limits of what will ever be testable.”

As theory pulls further and further ahead of the capabilities of experiment, physicists are taking this question seriously. “We are in various ways hitting the limits of what will ever be testable, unless we have misunderstood some essential point about the nature of reality,” says theoretical cosmologist George Ellis. “We have now seen all the visible universe (i.e back to the visual horizon) and only gravitational waves remain to test further; and we are approaching the limits of what particle colliders it will ever be feasible to build, for economic and technical reasons.”

Case in point: String theory. The darling of many theorists, string theory represents the basic building blocks of matter as vibrating strings. The strings take on different properties depending on their modes of vibration, just as the strings of a violin produce different notes depending on how they are played. To string theorists, the whole universe is a boisterous symphony performed upon these strings.

It’s a lovely idea. Lovelier yet, string theory could unify general relativity with quantum mechanics, solving what is perhaps the most stubborn problem in fundamental physics. The trouble? To put string theory to the test, we may need experiments that operate at energies far higher than any modern collider. It’s possible that experimental tests of the predictions of string theory will never be within our reach.

Meanwhile, cosmologists have found themselves at a similar impasse. We live in a universe that is, by some estimations, too good to be true. The fundamental constants of nature and the cosmological constant, which drives the accelerating expansion of the universe, seem “ fine-tuned ” to allow galaxies and stars to form. As Anil Ananthaswamy wrote elsewhere on this blog , “Tweak the charge on an electron, for instance, or change the strength of the gravitational force or the strong nuclear force just a smidgen, and the universe would look very different, and likely be lifeless.”

Why do these numbers, which are essential features of the universe and cannot be derived from more fundamental quantities, appear to conspire for our comfort?

One answer goes: If they were different, we wouldn’t be here to ask the question.

This is called the “anthropic principle,” and if you think it feels like a cosmic punt, you’re not alone. Researchers have been trying to underpin our apparent stroke of luck with hard science for decades. String theory suggests a solution: It predicts that our universe is just one among a multitude of universes, each with its own fundamental constants. If the cosmic lottery has played out billions of times, it isn’t so remarkable that the winning numbers for life should come up at least once.

In fact, you can reason your way to the “multiverse” in at least four different ways, according to MIT physicist Max Tegmark’s accounting . The tricky part is testing the idea. You can’t send or receive messages from neighboring universes, and most formulations of multiverse theory don’t make any testable predictions. Yet the theory provides a neat solution to the fine-tuning problem. Must we throw it out because it fails the falsifiability test?

Falsifiability is “just a simple motto that non-philosophically-trained scientists have latched onto.” “It would be completely non-scientific to ignore that possibility just because it doesn’t conform with some preexisting philosophical prejudices,” says Sean Carroll, a physicist at Caltech, who called for the “retirement” of the falsifiability principle in a controversial essay for Edge last year. Falsifiability is “just a simple motto that non-philosophically-trained scientists have latched onto,” argues Carroll. He also bristles at the notion that this viewpoint can be summed up as “elegance will suffice,” as Ellis put it in a stinging Nature comment written with cosmologist Joe Silk.

“Elegance can help us invent new theories, but does not count as empirical evidence in their favor,” says Carroll. “The criteria we use for judging theories are how good they are at accounting for the data, not how pretty or seductive or intuitive they are.”

Support Provided By

But Ellis and Silk worry that if physicists abandon falsifiability, they could damage the public’s trust in science and scientists at a time when that trust is critical to policymaking. “This battle for the heart and soul of physics is opening up at a time when scientific results—in topics from climate change to the theory of evolution—are being questioned by some politicians and religious fundamentalists,” Ellis and Silk wrote in Nature.

“The fear is that it would become difficult to separate such ‘science’ from New Age thinking, or science fiction,” says Ellis. If scientists backpedal on falsifiability, Ellis fears, intellectual disputes that were once resolved by experiment will devolve into never-ending philosophical feuds, and both the progress and the reputation of science will suffer.

But Carroll argues that he is simply calling for greater openness and honesty about the way science really happens. “I think that it’s more important than ever that scientists tell the truth. And the truth is that in practice, falsifiability is not a good criterion for telling science from non-science,” he says.

Perhaps “falsifiability” isn’t up to shouldering the full scientific and philosophical burden that’s been placed on it. “Sean is right that ‘falsifiability’ is a crude slogan that fails to capture what science really aims at,” argues MIT computer scientist Scott Aaronson, writing on his blog Shtetl Optimized . Yet, writes Aaronson, “falsifiability shouldn’t be ‘retired.’ Instead, falsifiability’s portfolio should be expanded , with full-time assistants (like explanatory power) hired to lighten falsifiability’s load.”

“I think falsifiability is not a perfect criterion, but it’s much less pernicious than what’s being served up by the ‘post-empirical’ faction,” says Frank Wilczek, a physicist at MIT. “Falsifiability is too impatient, in some sense,” putting immediate demands on theories that are not yet mature enough to meet them. “It’s an important discipline, but if it is applied too rigorously and too early, it can be stifling.”

So, where do we go from here?

“We need to rethink these issues in a philosophically sophisticated way that also takes the best interpretations of fundamental science, and it's limitations, seriously,” says Ellis. “Maybe we have to accept uncertainty as a profound aspect of our understanding of the universe in cosmology as well as particle physics.”

Go Deeper Editor’s picks for further reading

Edge: What scientific idea is ready for retirement? Falsifiability Sean Carroll calls for rethinking the falsifiability principle.

Nature: Scientific method: Defend the integrity of physics George Ellis and Joe Silk’s defense of falsifiability.

Philosophy of Science: Underdetermination and Theory Succession from the Perspective of String Theory Richard Dawid, a philosopher of science, argues that string theory is ushering in new paradigm of scientific thinking.

Quarterly Journal of the Royal Astronomical Society: Cosmology, A Brief Review In this 1963 address, cosmologist William McCrea surveyed the state of cosmology and suggested that it may be impossible to overcome uncertainty in our knowledge of the fundamental laws of the universe. (Hat tip to George Ellis.)

The Trouble with Physics Physicist Lee Smolin offers a biting critique of string theory in this popular 2006 book.

Receive emails about upcoming NOVA programs and related content, as well as featured reporting about current events through a science lens.

National Corporate funding for NOVA is provided by Carlisle Companies. Major funding for NOVA is provided by the NOVA Science Trust, the Corporation for Public Broadcasting, and PBS viewers.

Law of Falsifiability

The Law of Falsifiability is a rule that a famous thinker named Karl Popper came up with. In simple terms, for something to be called scientific, there must be a way to show it could be incorrect. Imagine you’re saying you have an invisible, noiseless, pet dragon in your room that no one can touch or see. If no one can test to see if the dragon is really there, then it’s not scientific. But if you claim that water boils at 100 degrees Celsius at sea level, we can test this. If it turns out water does not boil at this temperature under these conditions, then the claim would be proven false. That’s what Karl Popper was getting at – science is about making claims that can be tested, possibly shown to be false, and that’s what keeps it trustworthy and moving forward.

Examples of Law of Falsifiability

• Astrology – Astrology is like saying certain traits or events will happen to you based on star patterns. But because its predictions are too general and can’t be checked in a clear way, it doesn’t pass the test of falsifiability. This means astrology cannot be considered a scientific theory since you can’t show when it’s wrong with specific tests.
• The Theory of Evolution – In contrast, the theory of evolution is something we can test. It says that different living things developed over a very long time. If someone were to find an animal’s remains in a rock layer where it should not be, such as a rabbit in rock that’s 500 million years old, that would challenge the theory. Since we can test it by looking for evidence like this, evolution is considered falsifiable.

Why is it Important?

The Law of Falsifiability matters a lot because it separates what’s considered scientific from what’s not. When an idea can’t be tested or shown to be wrong, it can lead people down the wrong path. By focusing on theories we can test, science gets stronger and we learn more about the world for real. For everyday people, this is key because it means we can rely on science for things like medicine, technology, and understanding our environment. If scientists didn’t use this rule, we might believe in things that aren’t true, like magic potions or the idea that some stars can predict your future.

Implications and Applications

The rule of being able to test if something is false is basic in the world of science and is used in all sorts of subjects. For example, in an experiment, scientists try really hard to see if their guess about something can be shown wrong. If their guess survives all the tests, it’s a good sign; if not, they need to think again or throw it out. This is how science gets better and better.

Comparison with Related Axioms

• Verifiability : This means checking if a statement or idea is true. Both verifiability and falsifiability have to do with testing, but falsifiability is seen as more important because things that can be proven wrong are usually also things we can check for truth.
• Empiricism : This is the belief that knowledge comes from what we can sense – like seeing, hearing, or touching. Falsifiability and empiricism go hand in hand because both involve using real evidence to test out ideas.
• Reproducibility : This idea says that doing the same experiment in the same way should give you the same result. To show something is falsifiable, you should be able to repeat a test over and over, with the chance that it might fail.

Karl Popper brought the Law of Falsifiability into the world in the 1900s. He didn’t like theories that seemed to answer everything because, to him, they actually explained nothing. By making this law, he aimed to make a clear line between what could be taken seriously in science and what could not. It was his way of making sure scientific thinking stayed sharp and clear.

Controversies

Not everyone agrees that falsifiability is the only way to tell if something is scientific. Some experts point out areas in science, like string theory from physics, which are really hard to test and so are hard to apply this law to. Also, in science fields that look at history, like how the universe began or how life changed over time, it’s not always about predictions that can be tested, but more about understanding special events. These differences in opinion show that while it’s a strong part of scientific thinking, falsifiability might not work for every situation or be the only thing that counts for scientific ideas.

Related Topics

• Scientific Method : This is the process scientists use to study things. It involves asking questions, making a hypothesis, running experiments, and seeing if the results support the hypothesis. Falsifiability is part of this process because scientists have to be able to test their hypotheses.
• Peer Review : When scientists finish their work, other experts check it to make sure it was done right. This involves reviewing if the experiments and tests were set up in a way that they could have shown the work was false if it wasn’t true.
• Logic and Critical Thinking : These are skills that help us make good arguments and decisions. Understanding falsifiability helps people develop these skills because it teaches them to always look for ways to test ideas.

In conclusion, the Law of Falsifiability, as brought up by Karl Popper, is like a key part of a scientist’s toolbox. It makes sure that ideas need to be able to be tested and possibly shown to be not true. By using this rule, we avoid believing in things without good evidence, and we make the stuff we learn about the world through science stronger and more reliable.

• Foundations
• Write Paper

Search form

• Experiments
• Anthropology
• Self-Esteem
• Social Anxiety
• Foundations >
• Reasoning >

Falsifiability

Karl popper's basic scientific principle, karl popper's basic scientific principle.

Falsifiability, according to the philosopher Karl Popper, defines the inherent testability of any scientific hypothesis.

This article is a part of the guide:

• Inductive Reasoning
• Deductive Reasoning
• Hypothetico-Deductive Method
• Scientific Reasoning
• Testability

Browse Full Outline

• 1 Scientific Reasoning
• 2.1 Falsifiability
• 2.2 Verification Error
• 2.3 Testability
• 2.4 Post Hoc Reasoning
• 3 Deductive Reasoning
• 4.1 Raven Paradox
• 5 Causal Reasoning
• 6 Abductive Reasoning
• 7 Defeasible Reasoning

Science and philosophy have always worked together to try to uncover truths about the universe we live in. Indeed, ancient philosophy can be understood as the originator of many of the separate fields of study we have today, including psychology, medicine, law, astronomy, art and even theology.

Scientists design experiments and try to obtain results verifying or disproving a hypothesis, but philosophers are interested in understanding what factors determine the validity of scientific endeavors in the first place.

Whilst most scientists work within established paradigms, philosophers question the paradigms themselves and try to explore our underlying assumptions and definitions behind the logic of how we seek knowledge. Thus there is a feedback relationship between science and philosophy - and sometimes plenty of tension!

One of the tenets behind the scientific method is that any scientific hypothesis and resultant experimental design must be inherently falsifiable. Although falsifiability is not universally accepted, it is still the foundation of the majority of scientific experiments. Most scientists accept and work with this tenet, but it has its roots in philosophy and the deeper questions of truth and our access to it.

What is Falsifiability?

Falsifiability is the assertion that for any hypothesis to have credence, it must be inherently disprovable before it can become accepted as a scientific hypothesis or theory.

For example, someone might claim "the earth is younger than many scientists state, and in fact was created to appear as though it was older through deceptive fossils etc.” This is a claim that is unfalsifiable because it is a theory that can never be shown to be false. If you were to present such a person with fossils, geological data or arguments about the nature of compounds in the ozone, they could refute the argument by saying that your evidence was fabricated to appeared that way, and isn’t valid.

Importantly, falsifiability doesn’t mean that there are currently arguments against a theory, only that it is possible to imagine some kind of argument which would invalidate it. Falsifiability says nothing about an argument's inherent validity or correctness. It is only the minimum trait required of a claim that allows it to be engaged with in a scientific manner – a dividing line between what is considered science and what isn’t. Another important point is that falsifiability is not any claim that has yet to be proven true. After all, a conjecture that hasn’t been proven yet is just a hypothesis.

The idea is that no theory is completely correct , but if it can be shown both to be falsifiable  and supported with evidence that shows it's true, it can be accepted as truth.

For example, Newton's Theory of Gravity was accepted as truth for centuries, because objects do not randomly float away from the earth. It appeared to fit the data obtained by experimentation and research , but was always subject to testing.

However, Einstein's theory makes falsifiable predictions that are different from predictions made by Newton's theory, for example concerning the precession of the orbit of Mercury, and gravitational lensing of light. In non-extreme situations Einstein's and Newton's theories make the same predictions, so they are both correct. But Einstein's theory holds true in a superset of the conditions in which Newton's theory holds, so according to the principle of Occam's Razor , Einstein's theory is preferred. On the other hand, Newtonian calculations are simpler, so Newton's theory is useful for almost any engineering project, including some space projects. But for GPS we need Einstein's theory. Scientists would not have arrived at either of these theories, or a compromise between both of them, without the use of testable, falsifiable experiments. 

Popper saw falsifiability as a black and white definition; that if a theory is falsifiable, it is scientific , and if not, then it is unscientific. Whilst some "pure" sciences do adhere to this strict criterion, many fall somewhere between the two extremes, with  pseudo-sciences  falling at the extreme end of being unfalsifiable. 

Pseudoscience

According to Popper, many branches of applied science, especially social science, are not truly scientific because they have no potential for falsification.

Anthropology and sociology, for example, often use case studies to observe people in their natural environment without actually testing any specific hypotheses or theories.

While such studies and ideas are not falsifiable, most would agree that they are scientific because they significantly advance human knowledge.

Popper had and still has his fair share of critics, and the question of how to demarcate legitimate scientific enquiry can get very convoluted. Some statements are logically falsifiable but not practically falsifiable – consider the famous example of “it will rain at this location in a million years' time.” You could absolutely conceive of a way to test this claim, but carrying it out is a different story.

Thus, falsifiability is not a simple black and white matter. The Raven Paradox shows the inherent danger of relying on falsifiability, because very few scientific experiments can measure all of the data, and necessarily rely upon generalization . Technologies change along with our aims and comprehension of the phenomena we study, and so the falsifiability criterion for good science is subject to shifting.

For many sciences, the idea of falsifiability is a useful tool for generating theories that are testable and realistic. Testability is a crucial starting point around which to design solid experiments that have a chance of telling us something useful about the phenomena in question. If a falsifiable theory is tested and the results are significant , then it can become accepted as a scientific truth.

The advantage of Popper's idea is that such truths can be falsified when more knowledge and resources are available. Even long accepted theories such as Gravity, Relativity and Evolution are increasingly challenged and adapted.

The major disadvantage of falsifiability is that it is very strict in its definitions and does not take into account the contributions of sciences that are observational and descriptive .

• Psychology 101
• Flags and Countries
• Capitals and Countries

Martyn Shuttleworth , Lyndsay T Wilson (Sep 21, 2008). Falsifiability. Retrieved Sep 12, 2024 from Explorable.com: https://explorable.com/falsifiability

You Are Allowed To Copy The Text

The text in this article is licensed under the Creative Commons-License Attribution 4.0 International (CC BY 4.0) .

This means you're free to copy, share and adapt any parts (or all) of the text in the article, as long as you give appropriate credit and provide a link/reference to this page.

That is it. You don't need our permission to copy the article; just include a link/reference back to this page. You can use it freely (with some kind of link), and we're also okay with people reprinting in publications like books, blogs, newsletters, course-material, papers, wikipedia and presentations (with clear attribution).

Want to stay up to date? Follow us!

Save this course for later.

Don't have time for it all now? No problem, save it as a course and come back to it later.

Footer bottom

• Privacy Policy

• Subscribe to our RSS Feed
• Like us on Facebook
• Follow us on Twitter

• To save this word, you'll need to log in. Log In

unfalsifiable

Definition of unfalsifiable

Examples of unfalsifiable in a sentence.

These examples are programmatically compiled from various online sources to illustrate current usage of the word 'unfalsifiable.' Any opinions expressed in the examples do not represent those of Merriam-Webster or its editors. Send us feedback about these examples.

Word History

circa 1934, in the meaning defined above

Dictionary Entries Near unfalsifiable

unfaltering

Cite this Entry

“Unfalsifiable.” Merriam-Webster.com Dictionary , Merriam-Webster, https://www.merriam-webster.com/dictionary/unfalsifiable. Accessed 12 Sep. 2024.

Subscribe to America's largest dictionary and get thousands more definitions and advanced search—ad free!

Can you solve 4 words at once?

Word of the day.

See Definitions and Examples »

Get Word of the Day daily email!

Popular in Grammar & Usage

Plural and possessive names: a guide, 31 useful rhetorical devices, more commonly misspelled words, absent letters that are heard anyway, how to use accents and diacritical marks, popular in wordplay, 8 words for lesser-known musical instruments, it's a scorcher words for the summer heat, 7 shakespearean insults to make life more interesting, 10 words from taylor swift songs (merriam's version), 9 superb owl words, games & quizzes.

• Table of Contents
• Random Entry
• Chronological
• Editorial Information
• About the SEP
• Editorial Board
• How to Cite the SEP
• Special Characters
• Advanced Tools
• Support the SEP
• PDFs for SEP Friends
• Make a Donation
• SEPIA for Libraries
• Entry Contents

Bibliography

Academic tools.

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

Science and Pseudo-Science

The demarcation between science and pseudoscience is part of the larger task of determining which beliefs are epistemically warranted. This entry clarifies the specific nature of pseudoscience in relation to other categories of non-scientific doctrines and practices, including science denial(ism) and resistance to the facts. The major proposed demarcation criteria for pseudo-science are discussed and some of their weaknesses are pointed out. There is much more agreement on particular cases of demarcation than on the general criteria that such judgments should be based upon. This is an indication that there is still much important philosophical work to be done on the demarcation between science and pseudoscience.

1. The purpose of demarcations

2. the “science” of pseudoscience, 3.1 non-, un-, and pseudoscience, 3.2 non-science posing as science, 3.3 the doctrinal component, 3.4 a wider sense of pseudoscience, 3.5 the objects of demarcation, 3.6 a time-bound demarcation, 4.1 the logical positivists, 4.2 falsificationism, 4.3 the criterion of puzzle-solving, 4.4 criteria based on scientific progress, 4.5 epistemic norms, 4.6 multi-criterial approaches, 5. two forms of pseudo-science, 6.1 scepticism, 6.2 resistance to facts, 6.3 conspiracy theories, 6.4 bullshit, 6.5 epistemic relativism, 7. unity in diversity, cited works, philosophically-informed literature on pseudosciences and contested doctrines, other internet resources, related entries.

Demarcations of science from pseudoscience can be made for both theoretical and practical reasons (Mahner 2007, 516). From a theoretical point of view, the demarcation issue is an illuminating perspective that contributes to the philosophy of science in much the same way that the study of fallacies contributes to our knowledge of informal logic and rational argumentation. From a practical point of view, the distinction is important for decision guidance in both private and public life. Since science is our most reliable source of knowledge in a wide range of areas, we need to distinguish scientific knowledge from its look-alikes. Due to the high status of science in present-day society, attempts to exaggerate the scientific status of various claims, teachings, and products are common enough to make the demarcation issue pressing in many areas. The demarcation issue is therefore important in practical applications such as the following:

Climate policy : The scientific consensus on ongoing anthropogenic climate change leaves no room for reasonable doubt (Cook et al. 2016; Powell 2019). Science denial has considerably delayed climate action, and it is still one of the major factors that impede efficient measures to reduce climate change (Oreskes and Conway 2010; Lewandowsky et al. 2019). Decision-makers and the public need to know how to distinguish between competent climate science and science-mimicking disinformation on the climate.

Environmental policies : In order to be on the safe side against potential disasters it may be legitimate to take preventive measures when there is valid but yet insufficient evidence of an environmental hazard. This must be distinguished from taking measures against an alleged hazard for which there is no valid evidence at all. Therefore, decision-makers in environmental policy must be able to distinguish between scientific and pseudoscientific claims.

Healthcare : Medical science develops and evaluates treatments according to evidence of their effectiveness and safety. Pseudoscientific activities in this area give rise to ineffective and sometimes dangerous interventions. Healthcare providers, insurers, government authorities and – most importantly – patients need guidance on how to distinguish between medical science and medical pseudoscience.

Expert testimony : It is essential for the rule of law that courts get the facts right. The reliability of different types of evidence must be correctly determined, and expert testimony must be based on the best available knowledge. Sometimes it is in the interest of litigants to present non-scientific claims as solid science. Therefore courts must be able to distinguish between science and pseudoscience. Philosophers have often had prominent roles in the defence of science against pseudoscience in such contexts. (Pennock 2011)

Science education : The promoters of some pseudosciences (notably creationism) try to introduce their teachings in school curricula. Teachers and school authorities need to have clear criteria of inclusion that protect students against unreliable and disproved teachings.

Journalism : When there is scientific uncertainty, or relevant disagreement in the scientific community, this should be covered and explained in media reports on the issues in question. Equally importantly, differences of opinion between on the one hand legitimate scientific experts and on the other hand proponents of scientifically unsubstantiated claims should be described as what they are. Public understanding of topics such as climate change and vaccination has been considerably hampered by organised campaigns that succeeded in making media portray standpoints that have been thoroughly disproved in science as legitimate scientific standpoints (Boykoff and Boykoff 2004; Boykoff 2008). The media need tools and practices to distinguish between legitimate scientific controversies and attempts to peddle pseudoscientific claims as science.

Attempts to define what we today call science have a long history, and the roots of the demarcation problem have sometimes been traced back to Aristotle’s Posterior Analytics (Laudan 1983). Cicero’s arguments for dismissing certain methods of divination in his De divinatione has considerable similarities with modern criteria for the demarcation of science (Fernandez-Beanato 2020). However it was not until the 20th century that influential definitions of science have contrasted it against pseudoscience. Philosophical work on the demarcation problem seems to have waned after Laudan’s (1983) much noted death certificate according to which there is no hope of finding a necessary and sufficient criterion of something as heterogeneous as scientific methodology. In more recent years, the problem has been revitalized. Philosophers attesting to its vitality maintain that the concept can be clarified by other means than necessary and sufficient criteria (Pigliucci 2013; Mahner 2013) or that such a definition is indeed possible although it has to be supplemented with discipline-specific criteria in order to become fully operative (Hansson 2013).

The Latin word “pseudoscientia” was used already in the first half of the 17th century in discussions about the relationship between religion and empirical investigations (Guldentops 2020, 288n). The oldest known use of the English word “pseudoscience” dates from 1796, when the historian James Pettit Andrew referred to alchemy as a “fantastical pseudo-science” (Oxford English Dictionary). The word has been in frequent use since the 1880s (Thurs and Numbers 2013). Throughout its history the word has had a clearly defamatory meaning (Laudan 1983, 119; Dolby 1987, 204). It would be as strange for someone to proudly describe her own activities as pseudoscience as to boast that they are bad science. Since the derogatory connotation is an essential characteristic of the word “pseudoscience”, an attempt to extricate a value-free definition of the term would not be meaningful. An essentially value-laden term has to be defined in value-laden terms. This is often difficult since the specification of the value component tends to be controversial.

This problem is not specific to pseudoscience, but follows directly from a parallel but somewhat less conspicuous problem with the concept of science. The common usage of the term “science” can be described as partly descriptive, partly normative. When an activity is recognized as science this usually involves an acknowledgement that it has a positive role in our strivings for knowledge. On the other hand, the concept of science has been formed through a historical process, and many contingencies influence what we call and do not call science. Whether we call a claim, doctrine, or discipline “scientific” depends both on its subject area and its epistemic qualities. The former part of the delimitation is largely conventional, whereas the latter is highly normative, and closely connected with fundamental epistemological and metaphysical issues.

Against this background, in order not to be unduly complex a definition of science has to go in either of two directions. It can focus on the descriptive contents, and specify how the term is actually used. Alternatively, it can focus on the normative element, and clarify the more fundamental meaning of the term. The latter approach has been the choice of most philosophers writing on the subject, and will be at focus here. It involves, of necessity, some degree of idealization in relation to common usage of the term “science”, in particular concerning the delimitation of the subject-area of science.

The English word “science” is primarily used about the natural sciences and other fields of research that are considered to be similar to them. Hence, political economy and sociology are counted as sciences, whereas studies of literature and history are usually not. The corresponding German word, “Wissenschaft”, has a much broader meaning and includes all the academic specialties, including the humanities. The German term has the advantage of more adequately delimiting the type of systematic knowledge that is at stake in the conflict between science and pseudoscience. The misrepresentations of history presented by Holocaust deniers and other pseudo-historians are very similar in nature to the misrepresentations of natural science promoted by creationists and homeopaths.

More importantly, the natural and social sciences and the humanities are all parts of the same human endeavour, namely systematic and critical investigations aimed at acquiring the best possible understanding of the workings of nature, people, and human society. The disciplines that form this community of knowledge disciplines are increasingly interdependent. Since the second half of the 20th century, integrative disciplines such as astrophysics, evolutionary biology, biochemistry, ecology, quantum chemistry, the neurosciences, and game theory have developed at dramatic speed and contributed to tying together previously unconnected disciplines. These increased interconnections have also linked the sciences and the humanities closer to each other, as can be seen for instance from how historical knowledge relies increasingly on advanced scientific analysis of archaeological findings.

The conflict between science and pseudoscience is best understood with this extended sense of science. On one side of the conflict we find the community of knowledge disciplines that includes the natural and social sciences and the humanities. On the other side we find a wide variety of movements and doctrines, such as creationism, astrology, homeopathy, and Holocaust denialism that are in conflict with results and methods that are generally accepted in the community of knowledge disciplines.

Another way to express this is that the demarcation problem has a deeper concern than that of demarcating the selection of human activities that we have for various reasons chosen to call “sciences”. The ultimate issue is “how to determine which beliefs are epistemically warranted” (Fuller 1985, 331). In a wider approach, the sciences are fact-finding practices , i.e., human practices aimed at finding out, as far as possible, how things really are (Hansson 2018). Other examples of fact-finding practices in modern societies are journalism, criminal investigations, and the methods used by mechanics to search for the defect in a malfunctioning machine. Fact-finding practices are also prevalent in indigenous societies, for instance in the forms of traditional agricultural experimentation and the methods used for tracking animal prey (Liebenberg 2013). In this perspective, the demarcation of science is a special case of the delimitation of accurate fact-finding practices. The delimitation between science and pseudoscience has much in common with other delimitations, such as that between accurate and inaccurate journalism and between properly and improperly performed criminal investigations (Hansson 2018).

3. The “pseudo” of pseudoscience

The phrases “demarcation of science” and “demarcation of science from pseudoscience” are often used interchangeably, and many authors seem to have regarded them as equal in meaning. In their view, the task of drawing the outer boundaries of science is essentially the same as that of drawing the boundary between science and pseudoscience.

This picture is oversimplified. All non-science is not pseudoscience, and science has non-trivial borders to other non-scientific phenomena, such as metaphysics, religion, and various types of non-scientific systematized knowledge. (Mahner (2007, 548) proposed the term “parascience” to cover non-scientific practices that are not pseudoscientific.) Science also has the internal demarcation problem of distinguishing between good and bad science.

A comparison of the negated terms related to science can contribute to clarifying the conceptual distinctions. “Unscientific” is a narrower concept than “non-scientific” (not scientific), since the former but not the latter term implies some form of contradiction or conflict with science. “Pseudoscientific” is in its turn a narrower concept than “unscientific”. The latter term differs from the former in covering inadvertent mismeasurements and miscalculations and other forms of bad science performed by scientists who are recognized as trying but failing to produce good science.

Etymology provides us with an obvious starting-point for clarifying what characteristics pseudoscience has in addition to being merely non- or un-scientific. “Pseudo-” (ψευδο-) means false. In accordance with this, the Oxford English Dictionary (OED) defines pseudoscience as follows:

“A pretended or spurious science; a collection of related beliefs about the world mistakenly regarded as being based on scientific method or as having the status that scientific truths now have.”

Many writers on pseudoscience have emphasized that pseudoscience is non-science posing as science. The foremost modern classic on the subject (Gardner 1957) bears the title Fads and Fallacies in the Name of Science . According to Brian Baigrie (1988, 438), “[w]hat is objectionable about these beliefs is that they masquerade as genuinely scientific ones.” These and many other authors assume that to be pseudoscientific, an activity or a teaching has to satisfy the following two criteria (Hansson 1996):

The former of the two criteria is central to the concerns of the philosophy of science. Its precise meaning has been the subject of important controversies among philosophers, to be discussed below in Section 4. The second criterion has been less discussed by philosophers, but it needs careful treatment not least since many discussions of pseudoscience (in and out of philosophy) have been confused due to insufficient attention to it. Proponents of pseudoscience often attempt to mimic science by arranging conferences, journals, and associations that share many of the superficial characteristics of science, but do not satisfy its quality criteria. Naomi Oreskes (2019) called this phenomenon “facsimile science”. Blancke and coworkers (2017) called it “cultural mimicry of science”.

An immediate problem with the definition based on (1) and (2) is that it is too wide. There are phenomena that satisfy both criteria but are not commonly called pseudoscientific. One of the clearest examples of this is fraud in science. This is a practice that has a high degree of scientific pretence and yet does not comply with science, thus satisfying both criteria. Nevertheless, fraud in otherwise legitimate branches of science is seldom if ever called “pseudoscience”. The reason for this can be clarified with the following hypothetical examples (Hansson 1996).

Case 1 : A biochemist performs an experiment that she interprets as showing that a particular protein has an essential role in muscle contraction. There is a consensus among her colleagues that the result is a mere artefact, due to experimental error.

Case 2 : A biochemist goes on performing one sloppy experiment after the other. She consistently interprets them as showing that a particular protein has a role in muscle contraction not accepted by other scientists.

Case 3 : A biochemist performs various sloppy experiments in different areas. One is the experiment referred to in case 1. Much of her work is of the same quality. She does not propagate any particular unorthodox theory.

According to common usage, 1 and 3 are regarded as cases of bad science, and only 2 as a case of pseudoscience. What is present in case 2, but absent in the other two, is a deviant doctrine . Isolated breaches of the requirements of science are not commonly regarded as pseudoscientific. Pseudoscience, as it is commonly conceived, involves a sustained effort to promote standpoints different from those that have scientific legitimacy at the time.

This explains why fraud in science is not usually regarded as pseudoscientific. Such practices are not in general associated with a deviant or unorthodox doctrine. To the contrary, the fraudulent scientist is usually anxious that her results be in conformity with the predictions of established scientific theories. Deviations from these would lead to a much higher risk of disclosure.

The term “science” has both an individuated and an unindividuated sense. In the individuated sense, biochemistry and astronomy are different sciences, one of which includes studies of muscle proteins and the other studies of supernovae. The Oxford English Dictionary (OED) defines this sense of science as “a particular branch of knowledge or study; a recognized department of learning”. In the unindividuated sense, the study of muscle proteins and that of supernovae are parts of “one and the same” science. In the words of the OED, unindividuated science is “the kind of knowledge or intellectual activity of which the various ‘sciences‘ are examples”.

Pseudoscience is an antithesis of science in the individuated rather than the unindividuated sense. There is no unified corpus of pseudoscience corresponding to the corpus of science. For a phenomenon to be pseudoscientific, it must belong to one or the other of the particular pseudosciences. In order to accommodate this feature, the above definition can be modified by replacing (2) by the following (Hansson 1996):

Most philosophers of science, and most scientists, prefer to regard science as constituted by methods of inquiry rather than by particular doctrines. There is an obvious tension between (2′) and this conventional view of science. This, however, may be as it should since pseudoscience often involves a representation of science as a closed and finished doctrine rather than as a methodology for open-ended inquiry.

Sometimes the term “pseudoscience” is used in a wider sense than that which is captured in the definition constituted of (1) and (2′). Contrary to (2′), doctrines that conflict with science are sometimes called “pseudoscientific” in spite of not being advanced as scientific. Hence, Grove (1985, 219) included among the pseudoscientific doctrines those that “purport to offer alternative accounts to those of science or claim to explain what science cannot explain.” Similarly, Lugg (1987, 227–228) maintained that “the clairvoyant’s predictions are pseudoscientific whether or not they are correct”, despite the fact that most clairvoyants do not profess to be practitioners of science. In this sense, pseudoscience is assumed to include not only doctrines contrary to science proclaimed to be scientific but doctrines contrary to science  tout court, whether or not they are put forward in the name of science. Arguably, the crucial issue is not whether something is called “science” but whether it is claimed to have the function of science, namely to provide the most reliable information about its subject-matter. To cover this wider sense of pseudoscience, (2′) can be modified as follows (Hansson 1996, 2013):

Common usage seems to vacillate between the definitions (1)+(2′) and (1)+(2″); and this in an interesting way: In their comments on the meaning of the term, critics of pseudoscience tend to endorse a definition close to (1)+(2′), but their actual usage is often closer to (1)+(2″).

The following examples serve to illustrate the difference between the two definitions and also to clarify why clause (1) is needed:

• A creationist book gives a correct account of the structure of DNA.
• An otherwise reliable chemistry book gives an incorrect account of the structure of DNA.
• A creationist book denies that the human species shares common ancestors with other primates.
• A preacher who denies that science can be trusted also denies that the human species shares common ancestors with other primates.

(a) does not satisfy (1), and is therefore not pseudoscientific on either account. (b) satisfies (1) but neither (2′) nor (2″) and is therefore not pseudoscientific on either account. (c) satisfies all three criteria, (1), (2′), and (2″), and is therefore pseudoscientific on both accounts. Finally, (d) satisfies (1) and (2″) and is therefore pseudoscientific according to (1)+(2″) but not according to (1)+(2′). As the last two examples illustrate, pseudoscience and anti-science are sometimes difficult to distinguish. Promoters of some pseudosciences (notably homeopathy) tend to be ambiguous between opposition to science and claims that they themselves represent the best science.

Various proposals have been put forward on exactly what elements in science or pseudoscience criteria of demarcation should be applied to. Proposals include that the demarcation should refer to a research program (Lakatos 1974a, 248–249), an epistemic field or cognitive discipline, i.e. a group of people with common knowledge aims, and their practices (Bunge 1982, 2001; Mahner 2007), a theory (Popper 1962, 1974), a practice (Lugg 1992; Morris 1987), a scientific problem or question (Siitonen 1984), and a particular inquiry (Kuhn 1974; Mayo 1996). It is probably fair to say that demarcation criteria can be meaningfully applied on each of these levels of description. A much more difficult problem is whether one of these levels is the fundamental level to which assessments on the other levels are reducible. However, it should be noted that appraisals on different levels may be interdefinable. For instance, it is not an unreasonable assumption that a pseudoscientific doctrine is one that contains pseudoscientific statements as its core or defining claims. Conversely, a pseudoscientific statement may be characterized in terms of being endorsed by a pseudoscientific doctrine but not by legitimate scientific accounts of the same subject area.

Derksen (1993) differs from most other writers on the subject in placing the emphasis in demarcation on the pseudoscientist, i.e. the individual person conducting pseudoscience. His major argument for this is that pseudoscience has scientific pretensions, and such pretensions are associated with a person, not a theory, practice or entire field. However, as was noted by Settle (1971), it is the rationality and critical attitude built into institutions, rather than the personal intellectual traits of individuals, that distinguishes science from non-scientific practices such as magic. The individual practitioner of magic in a pre-literate society is not necessarily less rational than the individual scientist in modern Western society. What she lacks is an intellectual environment of collective rationality and mutual criticism. “It is almost a fallacy of division to insist on each individual scientist being critically-minded” (Settle 1971, 174).

Some authors have maintained that the demarcation between science and pseudoscience must be timeless. If this were true, then it would be contradictory to label something as pseudoscience at one but not another point in time. Hence, after showing that creationism is in some respects similar to some doctrines from the early 18th century, one author maintained that “if such an activity was describable as science then, there is a cause for describing it as science now” (Dolby 1987, 207). This argument is based on a fundamental misconception of science. It is an essential feature of science that it methodically strives for improvement through empirical testing, intellectual criticism, and the exploration of new terrain. A standpoint or theory cannot be scientific unless it relates adequately to this process of improvement, which means as a minimum that well-founded rejections of previous scientific standpoints are accepted. The practical demarcation of science cannot be timeless, for the simple reason that science itself is not timeless.

Nevertheless, the mutability of science is one of the factors that renders the demarcation between science and pseudoscience difficult. Derksen (1993, 19) rightly pointed out three major reasons why demarcation is sometimes difficult: science changes over time, science is heterogenous, and established science itself is not free of the defects characteristic of pseudoscience.

4. Alternative demarcation criteria

Philosophical discussions on the demarcation of pseudoscience have usually focused on the normative issue, i.e. the missing scientific quality of pseudoscience (rather than on its attempt to mimic science. One option is to base the demarcation on the fundamental function that science shares with other fact-finding processes, namely to provide us with the most reliable information about its subject-matter that is currently available. This could lead to the specification of critierion (1) from Section 3.2 as follows:

This definition has the advantages of (i) being applicable across disciplines with highly different methodologies and (ii) allowing for a statement to be pseudoscientific at present although it was not so in an earlier period (or, although less commonly, the other way around). (Hansson 2013) At the same time it removes the practical determination whether a statement or doctrine is pseudoscientific from the purview of armchair philosophy to that of scientists specialized in the subject-matter that the statement or doctrine relates to. Philosophers have usually opted for demarcation criteria that appear not to require specialized knowledge in the pertinent subject area.

Around 1930, the logical positivists of the Vienna Circle developed various verificationist approaches to science. The basic idea was that a scientific statement could be distinguished from a metaphysical statement by being at least in principle possible to verify. This standpoint was associated with the view that the meaning of a proposition is its method of verification (see the section on Verificationism in the entry on the Vienna Circle ). This proposal has often been included in accounts of the demarcation between science and pseudoscience. However, this is not historically quite accurate since the verificationist proposals had the aim of solving a distinctly different demarcation problem, namely that between science and metaphysics.

Karl Popper described the demarcation problem as the “key to most of the fundamental problems in the philosophy of science” (Popper 1962, 42). He rejected verifiability as a criterion for a scientific theory or hypothesis to be scientific, rather than pseudoscientific or metaphysical. Instead he proposed as a criterion that the theory be falsifiable, or more precisely that “statements or systems of statements, in order to be ranked as scientific, must be capable of conflicting with possible, or conceivable observations” (Popper 1962, 39).

Popper presented this proposal as a way to draw the line between statements belonging to the empirical sciences and “all other statements – whether they are of a religious or of a metaphysical character, or simply pseudoscientific” (Popper 1962, 39; cf. Popper 1974, 981). This was both an alternative to the logical positivists’ verification criteria and a criterion for distinguishing between science and pseudoscience. Although Popper did not emphasize the distinction, these are of course two different issues (Bartley 1968). Popper conceded that metaphysical statements may be “far from meaningless” (1974, 978–979) but showed no such appreciation of pseudoscientific statements.

Popper’s demarcation criterion has been criticized both for excluding legitimate science (Hansson 2006) and for giving some pseudosciences the status of being scientific (Agassi 1991; Mahner 2007, 518–519). Strictly speaking, his criterion excludes the possibility that there can be a pseudoscientific claim that is refutable. According to Larry Laudan (1983, 121), it “has the untoward consequence of countenancing as ‘scientific’ every crank claim which makes ascertainably false assertions”. Astrology, rightly taken by Popper as an unusually clear example of a pseudoscience, has in fact been tested and thoroughly refuted (Culver and Ianna 1988; Carlson 1985). Similarly, the major threats to the scientific status of psychoanalysis, another of his major targets, do not come from claims that it is untestable but from claims that it has been tested and failed the tests.

Defenders of Popper have claimed that this criticism relies on an uncharitable interpretation of his ideas. They claim that he should not be interpreted as meaning that falsifiability is a sufficient condition for demarcating science. Some passages seem to suggest that he takes it as only a necessary condition (Feleppa 1990, 142). Other passages suggest that for a theory to be scientific, Popper requires (in addition to falsifiability) that energetic attempts are made to put the theory to test and that negative outcomes of the tests are accepted (Cioffi 1985, 14–16). A falsification-based demarcation criterion that includes these elements will avoid the most obvious counter-arguments to a criterion based on falsifiability alone.

However, in what seems to be his last statement of his position, Popper declared that falsifiability is a both necessary and a sufficient criterion. “A sentence (or a theory) is empirical-scientific if and only if it is falsifiable.” Furthermore, he emphasized that the falsifiability referred to here “only has to do with the logical structure of sentences and classes of sentences” (Popper [1989] 1994, 82). A (theoretical) sentence, he says, is falsifiable if and only if it logically contradicts some (empirical) sentence that describes a logically possible event that it would be logically possible to observe (Popper [1989] 1994, 83). A statement can be falsifiable in this sense although it is not in practice possible to falsify it. It would seem to follow from this interpretation that a statement’s status as scientific or non-scientific does not shift with time. On previous occasions he seems to have interpreted falsifiability differently, and maintained that “what was a metaphysical idea yesterday can become a testable scientific theory tomorrow; and this happens frequently” (Popper 1974, 981, cf. 984).

Logical falsifiability is a much weaker criterion than practical falsifiability. However, even logical falsifiability can create problems in practical demarcations. Popper once adopted the view that natural selection is not a proper scientific theory, arguing that it comes close to only saying that “survivors survive”, which is tautological. “Darwinism is not a testable scientific theory, but a metaphysical research program” (Popper 1976, 168). This statement has been criticized by evolutionary scientists who pointed out that it misrepresents evolution. The theory of natural selection has given rise to many predictions that have withstood tests both in field studies and in laboratory settings (Ruse 1977; 2000).

In a lecture in Darwin College in 1977, Popper retracted his previous view that the theory of natural selection is tautological. He now admitted that it is a testable theory although “difficult to test” (Popper 1978, 344). However, in spite of his well-argued recantation, his previous standpoint continues to be propagated in defiance of the accumulating evidence from empirical tests of natural selection.

Thomas Kuhn is one of many philosophers for whom Popper’s view on the demarcation problem was a starting-point for developing their own ideas. Kuhn criticized Popper for characterizing “the entire scientific enterprise in terms that apply only to its occasional revolutionary parts” (Kuhn 1974, 802). Popper’s focus on falsifications of theories led to a concentration on the rather rare instances when a whole theory is at stake. According to Kuhn, the way in which science works on such occasions cannot be used to characterize the entire scientific enterprise. Instead it is in “normal science”, the science that takes place between the unusual moments of scientific revolutions, that we find the characteristics by which science can be distinguished from other activities (Kuhn 1974, 801).

In normal science, the scientist’s activity consists in solving puzzles rather than testing fundamental theories. In puzzle-solving, current theory is accepted, and the puzzle is indeed defined in its terms. In Kuhn’s view, “it is normal science, in which Sir Karl’s sort of testing does not occur, rather than extraordinary science which most nearly distinguishes science from other enterprises”, and therefore a demarcation criterion must refer to the workings of normal science (Kuhn 1974, 802). Kuhn’s own demarcation criterion is the capability of puzzle-solving, which he sees as an essential characteristic of normal science.

Kuhn’s view of demarcation is most clearly expressed in his comparison of astronomy with astrology. Since antiquity, astronomy has been a puzzle-solving activity and therefore a science. If an astronomer’s prediction failed, then this was a puzzle that he could hope to solve for instance with more measurements or adjustments of the theory. In contrast, the astrologer had no such puzzles since in that discipline “particular failures did not give rise to research puzzles, for no man, however skilled, could make use of them in a constructive attempt to revise the astrological tradition” (Kuhn 1974, 804). Therefore, according to Kuhn, astrology has never been a science.

Popper disapproved thoroughly of Kuhn’s demarcation criterion. According to Popper, astrologers are engaged in puzzle solving, and consequently Kuhn’s criterion commits him to recognize astrology as a science. (Contrary to Kuhn, Popper defined puzzles as “minor problems which do not affect the routine”.) In his view Kuhn’s proposal leads to “the major disaster” of a “replacement of a rational criterion of science by a sociological one” (Popper 1974, 1146–1147).

Popper’s demarcation criterion concerns the logical structure of theories. Imre Lakatos described this criterion as “a rather stunning one. A theory may be scientific even if there is not a shred of evidence in its favour, and it may be pseudoscientific even if all the available evidence is in its favour. That is, the scientific or non-scientific character of a theory can be determined independently of the facts” (Lakatos 1981, 117).

Instead, Lakatos (1970; 1974a; 1974b; 1981) proposed a modification of Popper’s criterion that he called “sophisticated (methodological) falsificationism”. On this view, the demarcation criterion should not be applied to an isolated hypothesis or theory, but rather to a whole research program that is characterized by a series of theories successively replacing each other. In his view, a research program is progressive if the new theories make surprising predictions that are confirmed. In contrast, a degenerating research programme is characterized by theories being fabricated only in order to accommodate known facts. Progress in science is only possible if a research program satisfies the minimum requirement that each new theory that is developed in the program has a larger empirical content than its predecessor. If a research program does not satisfy this requirement, then it is pseudoscientific.

According to Paul Thagard (1978, 228), a theory or discipline is pseudoscientific if it satisfies two criteria. One of these is that the theory fails to progress, and the other that “the community of practitioners makes little attempt to develop the theory towards solutions of the problems, shows no concern for attempts to evaluate the theory in relation to others, and is selective in considering confirmations and disconfirmations”. A major difference between this approach and that of Lakatos is that Lakatos would classify a nonprogressive discipline as pseudoscientific even if its practitioners work hard to improve it and turn it into a progressive discipline. (In later work, Thagard has abandoned this approach and instead promoted a form of multi-criterial demarcation (Thagard 1988, 157-173).)

In a somewhat similar vein, Daniel Rothbart (1990) emphasized the distinction between the standards to be used when testing a theory and those to be used when determining whether a theory should at all be tested. The latter, the eligibility criteria, include that the theory should encapsulate the explanatory success of its rival, and that it should yield testable implications that are inconsistent with those of the rival. According to Rothbart, a theory is unscientific if it is not testworthy in this sense.

George Reisch proposed that demarcation could be based on the requirement that a scientific discipline be adequately integrated into the other sciences. The various scientific disciplines have strong interconnections that are based on methodology, theory, similarity of models etc. Creationism, for instance, is not scientific because its basic principles and beliefs are incompatible with those that connect and unify the sciences. More generally speaking, says Reisch, an epistemic field is pseudoscientific if it cannot be incorporated into the existing network of established sciences (Reisch 1998; cf. Bunge 1982, 379).

Paul Hoyninengen-Huene (2013) identifies science with systematic knowledge, and proposes that systematicity can be used as a demarcation criterion. However as shown by Naomi Oreskes, this is a problematic criterion, not least since some pseudosciences seem to satisfy it (Oreskes 2019).

A different approach, namely to base demarcation criteria on the value base of science, was proposed by sociologist Robert K. Merton ([1942] 1973). According to Merton, science is characterized by an “ethos”, i.e. spirit, that can be summarized as four sets of institutional imperatives. The first of these, universalism , asserts that whatever their origins, truth claims should be subjected to preestablished, impersonal criteria. This implies that the acceptance or rejection of claims should not depend on the personal or social qualities of their protagonists.

The second imperative, communism , says that the substantive findings of science are the products of social collaboration and therefore belong to the community, rather than being owned by individuals or groups. This is, as Merton pointed out, incompatible with patents that reserve exclusive rights of use to inventors and discoverers. The term “communism” is somewhat infelicitous; “communality” probably captures better what Merton aimed at.

His third imperative, disinterestedness , imposes a pattern of institutional control that is intended to curb the effects of personal or ideological motives that individual scientists may have. The fourth imperative, organized scepticism , implies that science allows detached scrutiny of beliefs that are dearly held by other institutions. This is what sometimes brings science into conflicts with religions and ideologies.

Merton described these criteria as belonging to the sociology of science, and thus as empirical statements about norms in actual science rather than normative statements about how science should be conducted (Merton [1942] 1973, 268). His criteria have often been dismissed by sociologists as oversimplified, and they have only had limited influence in philosophical discussions on the demarcation issue (Dolby 1987; Ruse 2000). Their potential in the latter context does not seem to have been sufficiently explored.

Popper’s method of demarcation consists essentially of the single criterion of falsifiability (although some authors have wanted to combine it with the additional criteria that tests are actually performed and their outcomes respected, see Section 4.2). Most of the other criteria discussed above are similarly mono-criterial, of course with Merton’s proposal as a major exception.

Most authors who have proposed demarcation criteria have instead put forward a list of such criteria. A large number of lists have been published that consist of (usually 5–10) criteria that can be used in combination to identify a pseudoscience or pseudoscientific practice. This includes lists by Langmuir ([1953] 1989), Gruenberger (1964), Dutch (1982), Bunge (1982), Radner and Radner (1982), Kitcher (1982, 30–54), Grove (1985), Thagard (1988, 157–173), Glymour and Stalker (1990), Derksen (1993, 2001), Vollmer (1993), Ruse (1996, 300–306) and Mahner (2007). Many of the criteria that appear on such lists relate closely to criteria discussed above in Sections 4.2 and 4.4. One such list reads as follows:

• Belief in authority : It is contended that some person or persons have a special ability to determine what is true or false. Others have to accept their judgments.
• Unrepeatable experiments : Reliance is put on experiments that cannot be repeated by others with the same outcome.
• Handpicked examples : Handpicked examples are used although they are not representative of the general category that the investigation refers to.
• Unwillingness to test : A theory is not tested although it is possible to test it.
• Disregard of refuting information : Observations or experiments that conflict with a theory are neglected.
• Built-in subterfuge : The testing of a theory is so arranged that the theory can only be confirmed, never disconfirmed, by the outcome.
• Explanations are abandoned without replacement . Tenable explanations are given up without being replaced, so that the new theory leaves much more unexplained than the previous one.

Some of the authors who have proposed multicriterial demarcations have defended this approach as being superior to any mono-criterial demarcation. Hence, Bunge (1982, 372) asserted that many philosophers have failed to provide an adequate definition of science since they have presupposed that a single attribute will do; in his view the combination of several criteria is needed. Dupré (1993, 242) proposed that science is best understood as a Wittgensteinian family resemblance concept. This would mean that there is a set of features that are characteristic of science, but although every part of science will have some of these features, we should not expect any part of science to have all of them. Irzik and Nola (2011) proposed the use of this approach in science education.

However, a multicriterial definition of science is not needed to justify a multicriterial account of how pseudoscience deviates from science. Even if science can be characterized by a single defining characteristic, different pseudoscientific practices may deviate from science in widely divergent ways.

Some forms of pseudoscience have as their main objective the promotion of a particular theory of their own, whereas others are driven by a desire to fight down some scientific theory or branch of science. The former type of pseudoscience has been called pseudo-theory promotion , and the latter science denial(ism) (Hansson 2017). Pseudo-theory promotion is exemplified by homeopathy, astrology, and ancient astronaut theories. The term “denial” was first used about the pseudo-scientific claim that the Nazi holocaust never took place. The phrase “holocaust denial” was in use already in the early 1980s (Gleberzon 1983). The term “climate change denial” became common around 2005 (e.g. Williams 2005). Other forms of science denial are relativity theory denial, tobacco disease denial, hiv denialism, and vaccination denialism.

Many forms of pseudoscience combine pseudo-theory promotion with science denialism. For instance, creationism and its skeletal version “intelligent design” are constructed to support a fundamentalist interpretation of Genesis. However, as practiced today, creationism has a strong focus on the repudiation of evolution, and it is therefore predominantly a form of science denialism.

The most prominent difference between pseudo-theory promotion and science denial is their different attitudes to conflicts with established science. Science denialism usually proceeds by producing false controversies with legitimate science, i.e. claims that there is a scientific controversy when there is in fact none. This is an old strategy, applied already in the 1930s by relativity theory deniers (Wazeck 2009, 268–269). It has been much used by tobacco disease deniers sponsored by the tobacco industry (Oreskes and Conway 2010; Dunlap and Jacques 2013), and it is currently employed by climate science denialists (Boykoff and Boykoff 2004; Boykoff 2008). However, whereas the fabrication of fake controversies is a standard tool in science denial, it is seldom if ever used in pseudo-theory promotion. To the contrary, advocates of pseudosciences such as astrology and homeopathy tend to describe their theories as conformable to mainstream science.

6. Some related terms

The term scepticism (skepticism) has at least three distinct usages that are relevant for the discussion on pseudoscience. First, scepticism is a philosophical method that proceeds by casting doubt on claims usually taken to be trivially true, such as the existence of the external world. This has been, and still is, a highly useful method for investigating the justification of what we in practice consider to be certain beliefs. Secondly, criticism of pseudoscience is often called scepticism. This is the term most commonly used by organisations devoted to the disclosure of pseudoscience. Thirdly, opposition to the scientific consensus in specific areas is sometimes called scepticism. For instance, climate science deniers often call themselves “climate sceptics”.

To avoid confusion, the first of these notions can be specified as “philosophical scepticism”, the second as “scientific scepticism” or “defence of science”, and the third as “science denial(ism)”. Adherents of the first two forms of scepticism can be called “philosophical sceptics”, respectively “science defenders”. Adherents of the third form can be called “science deniers” or “science denialists”. Torcello (2016) proposed the term “pseudoscepticism” for so-called climate scepticism.

Unwillingness to accept strongly supported factual statements is a traditional criterion of pseudoscience. (See for instance item 5 on the list of seven criteria cited in Section 4.6.) The term “fact resistance” or “resistance to facts” was used already in the 1990s, for instance by Arthur Krystal (1999, p. 8), who complained about a “growing resistance to facts”, consisting in people being “simply unrepentant about not knowing things that do not reflect their interests”. The term “fact resistance” can refer to unwillingness to accept well-supported factual claims whether or not that support originates in science. It is particularly useful in relation to fact-finding practices that are not parts of science. (Cf. Section 2.)

Generally speaking, conspiracy theories are theories according to which there exists some type of secret collusion for any type of purpose. In practice, the term mostly refers to implausible such theories, used to explain social facts that have other, considerably more plausible explanations. Many pseudosciences are connected with conspiracy theories. For instance, one of the difficulties facing anti-vaccinationists is that they have to explain the overwhelming consensus among medical experts that vaccines are efficient. This is often done by claims of a conspiracy:

At the heart of the anti-vaccine conspiracy movement [lies] the argument that large pharmaceutical companies and governments are covering up information about vaccines to meet their own sinister objectives. According to the most popular theories, pharmaceutical companies stand to make such healthy profits from vaccines that they bribe researchers to fake their data, cover up evidence of the harmful side effects of vaccines, and inflate statistics on vaccine efficacy. (Jolley and Douglas 2014)

Conspiracy theories have peculiar epistemic characteristics that contribute to their pervasiveness. (Keeley 1999) In particular, they are often associated with a type of circular reasoning that allows evidence against the conspiracy to be interpreted as evidence for it.

The term “bullshit” was introduced into philosophy by Harry Frankfurt, who first discussed it in a 1986 essay ( Raritan Quarterly Review ) and developed the discussion into a book (2005). Frankfurt used the term to describe a type of falsehood that does not amount to lying. A person who lies deliberately chooses not to tell the truth, whereas a person who utters bullshit is not interested in whether what (s)he says is true or false, only in its suitability for his or her purpose. Moberger (2020) has proposed that pseudoscience should be seen as a special case of bullshit, understood as “a culpable lack of epistemic conscientiousness”.

Epistemic relativism is a term with many meanings; the meaning most relevant in discussions on pseudoscience is denial of the common assumption that there is intersubjective truth in scientific matters, which scientists can and should try to approach. Epistemic relativists claim that (natural) science has no special claim to knowledge, but should be seen “as ordinary social constructions or as derived from interests, political-economic relations, class structure, socially defined constraints on discourse, styles of persuasion, and so on” (Buttel and Taylor 1992, 220). Such ideas have been promoted under different names, including “social constructivism”, the “strong programme”, “deconstructionism”, and “postmodernism”. The distinction between science and pseudoscience has no obvious role in epistemic relativism. Some academic epistemic relativists have actively contributed to the promotion of doctrines such as AIDS denial, vaccination denial, creationism, and climate science denial (Hansson 2020, Pennock 2010). However, the connection between epistemic relativism and pseudoscience is controversial. Some proponents of epistemic relativism have maintained that that relativism “is almost always more useful to the side with less scientific credibility or cognitive authority” (Scott et al. 1990, 490). Others have denied that epistemic relativism facilitates or encourages standpoints such as denial of anthropogenic climate change or other environmental problems (Burningham and Cooper 1999, 306).

Kuhn observed that although his own and Popper’s criteria of demarcation are profoundly different, they lead to essentially the same conclusions on what should be counted as science respectively pseudoscience (Kuhn 1974, 803). This convergence of theoretically divergent demarcation criteria is a quite general phenomenon. Philosophers and other theoreticians of science differ widely in their views on what science is. Nevertheless, there is virtual unanimity in the community of knowledge disciplines on most particular issues of demarcation. There is widespread agreement for instance that creationism, astrology, homeopathy, Kirlian photography, dowsing, ufology, ancient astronaut theory, Holocaust denialism, Velikovskian catastrophism, and climate change denialism are pseudosciences. There are a few points of controversy, for instance concerning the status of Freudian psychoanalysis, but the general picture is one of consensus rather than controversy in particular issues of demarcation.

It is in a sense paradoxical that so much agreement has been reached in particular issues in spite of almost complete disagreement on the general criteria that these judgments should presumably be based upon. This puzzle is a sure indication that there is still much important philosophical work to be done on the demarcation between science and pseudoscience.

Philosophical reflection on pseudoscience has brought forth other interesting problem areas in addition to the demarcation between science and pseudoscience. Examples include related demarcations such as that between science and religion, the relationship between science and reliable non-scientific knowledge (for instance everyday knowledge), the scope for justifiable simplifications in science education and popular science, the nature and justification of methodological naturalism in science (Boudry et al 2010), and the meaning or meaninglessness of the concept of a supernatural phenomenon. Several of these problem areas have as yet not received much philosophical attention.

• Agassi, Joseph, 1991. “Popper’s demarcation of science refuted”, Methodology and Science , 24: 1–7.
• Baigrie, B.S., 1988. “Siegel on the Rationality of Science”, Philosophy of Science , 55: 435–441.
• Bartley III, W. W., 1968. “Theories of demarcation between science and metaphysics”, pp. 40–64 in Imre Lakatos and Alan Musgrave (eds.), Problems in the Philosophy of Science, Proceedings of the International Colloquium in the Philosophy of Science, London 1965 (Volume 3), Amsterdam: North-Holland Publishing Company.
• Blancke, Stefaan, Maarten Boudry and Massimo Pigliucci, 2017. “Why do irrational beliefs mimic science? The cultural evolution of pseudoscience”, Theoria , 83(1): 78–97.
• Boudry, Maarten, Stefaan Blancke, and Johan Braeckman, 2010. “How not to attack intelligent design creationism: Philosophical misconceptions about methodological naturalism.” Foundations of Science , 153: 227–244.
• Boykoff, M. T., 2008. “Lost in translation? United States television news coverage of anthropogenic climate change, 1995–2004”, Climatic Change , 86: 1–11.
• Boykoff, M. T. and J. M. Boykoff, 2004. “Balance as bias: global warming and the U.S. prestige press”, Global Environmental Change , 14: 125–136.
• Bunge, Mario, 1982. “Demarcating Science from Pseudoscience”, Fundamenta Scientiae , 3: 369–388.
• –––, 2001. “Diagnosing pseudoscience”, in Mario Bunge, Philosophy in Crisis. The Need for Reconstruction , Amherst, N.Y.; Prometheus Books, pp. 161–189.
• Burningham, K., and G. Cooper, 1999. “Being constructive: Social constructionism and the environment”, Sociology , 33(2): 297–316.
• Buttel, Frederick H. and Peter J. Taylor, 1992. “Environmental sociology and global environmental change: A critical assessment”, Society and Natural Resources , 5(3): 211–230.
• Carlson, Shawn, 1985. “A Double Blind Test of Astrology”, Nature , 318: 419–425.
• Cioffi, Frank, 1985. “Psychoanalysis, pseudoscience and testability”, pp 13–44 in Gregory Currie and Alan Musgrave, (eds.) Popper and the Human Sciences , Dordrecht: Martinus Nijhoff Publishers.
• Cook, John, Naomi Oreskes, Peter T. Doran, William RL Anderegg, Bart Verheggen, Ed W. Maibach, J. Stuart Carlton, et al., 2016. “Consensus on consensus: A synthesis of consensus estimates on human-caused global warming”, Environmental Research Letters , 11: 048002.
• Culver, Roger and Ianna, Philip, 1988. Astrology: True or False , Buffalo: Prometheus Books.
• Derksen, A.A., 1993. “The seven sins of pseudoscience”, Journal for General Philosophy of Science , 24: 17–42.
• –––, 2001. “The seven strategies of the sophisticated pseudoscience: a look into Freud’s rhetorical tool box”, Journal for General Philosophy of Science , 32: 329–350.
• Dolby, R.G.A., 1987. “Science and pseudoscience: the case of creationism”, Zygon , 22: 195–212.
• Dunlap, Riley E., and Peter J. Jacques, 2013. “Climate change denial books and conservative think tanks: exploring the connection”, American Behavioral Scientist , 57(6): 699–731.
• Dupré, John, 1993. The Disorder of Things: Metaphysical Foundations of the Disunity of Science , Cambridge, MA: Harvard University Press.
• Dutch, Steven I, 1982. “Notes on the nature of fringe science”, Journal of Geological Education , 30: 6–13.
• Feleppa, Robert, 1990. “Kuhn, Popper, and the Normative Problem of Demarcation”, pp. 140–155 in Patrick Grim (ed.) Philosophy of Science and the Occult , 2 nd edition, Albany: State University of New York Press.
• Fernandez-Beanato, Damian, 2020. “Cicero’s demarcation of science: A report of shared criteria”, Studies in History and Philosophy of Science (Part A), 83: 97–102.
• Frankfurt, Harry G., 2005. On Bullshit , Princeton: Princeton University Press; see also the essay with the same title in Raritan Quarterly Review , 6(2): 81–100.
• Fuller, Steve, 1985. “The demarcation of science: a problem whose demise has been greatly exaggerated”, Pacific Philosophical Quarterly , 66: 329–341.
• Gardner, Martin, 1957. Fads and Fallacies in the Name of Science , Dover 1957; expanded version of his In the Name of Science , 1952.
• Gleberzon, William, 1983. “Academic freedom and Holocaust denial literature: Dealing with infamy”, Interchange , 14(4): 62–69.
• Glymour, Clark and Stalker, Douglas, 1990. “Winning through Pseudoscience”, pp 92–103 in Patrick Grim (ed.) Philosophy of Science and the Occult , 2 nd edition, Albany: State University of New York Press.
• Grove , J.W., 1985. “Rationality at Risk: Science against Pseudoscience”, Minerva , 23: 216–240.
• Gruenberger, Fred J., 1964. “A measure for crackpots”, Science , 145: 1413–1415.
• Guldentops, Guy, 2020. “Nicolaus Ellenbog’s ‘Apologia for the Astrologers’: A Benedictine’s View on Astral Determinism”, Bulletin de Philosophie Médiévale , 62: 251–334.
• Hansson, Sven Ove, 1996. “Defining Pseudoscience”, Philosophia Naturalis , 33: 169–176.
• –––, 2006. “Falsificationism Falsified”, Foundations of Science , 11: 275–286.
• –––, 2013. “Defining pseudoscience and science”, pp. 61–77 in Pigliucci and Boudry (eds.) 2013.
• –––, 2017. “Science denial as a form of pseudoscience”, Studies in History and Philosophy of Science , 63: 39–47.
• –––, 2018. “How connected are the major forms of irrationality? An analysis of pseudoscience, science denial, fact resistance and alternative facts”, Mètode Science Study Journal , 8: 125–131.
• –––, 2020. “Social constructivism and climate science denial”, European Journal for Philosophy of Science , 10: 37.
• Hoyninengen-Huene, Paul, 2013. Systematicity. The nature of science , Oxford: Oxford University Press.
• Irzik, Gürol, and Robert Nola, 2011. “A family resemblance approach to the nature of science for science education”, Science and Education , 20(7): 591–607.
• Jolley, Daniel, and Karen M. Douglas, 2014. “The effects of anti-vaccine conspiracy theories on vaccination intentions”, PloS One , 9(2): e89177.
• Keeley, Brian L., 1999. “Of Conspiracy Theories”, The Journal of Philosophy , 96(3): 109–126.
• Kitcher, Philip, 1982. Abusing Science. The Case Against Creationism , Cambridge, MA: MIT Press.
• Krystal, Arthur, 1999. “At Large and at Small: What Do You Know?”, American Scholar , 68(2): 7–13.
• Kuhn, Thomas S., 1974. “Logic of Discovery or Psychology of Research?”, pp. 798–819 in P.A. Schilpp, The Philosophy of Karl Popper , The Library of Living Philosophers, vol xiv, book ii. La Salle: Open Court.
• Lakatos, Imre, 1970. “Falsification and the Methodology of Research program”, pp 91–197 in Imre Lakatos and Alan Musgrave (eds.) Criticism and the Growth of Knowledge . Cambridge: Cambridge University Press.
• –––, 1974a. “Popper on Demarcation and Induction”, pp. 241–273 in P.A. Schilpp, The Philosophy of Karl Popper (The Library of Living Philosophers, Volume 14, Book 1). La Salle: Open Court.
• –––, 1974b. “Science and pseudoscience”, Conceptus , 8: 5–9.
• –––, 1981. “Science and pseudoscience”, pp. 114–121 in S. Brown, et al . (eds.) Conceptions of Inquiry: A Reader , London: Methuen.
• Langmuir, Irving, [1953] 1989. “Pathological Science”, Physics Today , 42(10): 36–48.
• Laudan, Larry, 1983. “The demise of the demarcation problem”, in R.S. Cohan and L. Laudan (eds.), Physics, Philosophy, and Psychoanalysis , Dordrecht: Reidel, pp. 111–127.
• Lewandowsky, Stephan, Toby D. Pilditch, Jens K. Madsen, Naomi Oreskes, and James S. Risbey, 2019. “Influence and seepage: An evidence-resistant minority can affect public opinion and scientific belief formation”, Cognition , 188: 124–139.
• Liebenberg, L., 2013. The Origin of Science. The evolutionary roots of scientific reasoning and its implications for citizen science , Cape Town: CyberTracker.
• Lugg, Andrew, 1987. “Bunkum, Flim-Flam and Quackery: Pseudoscience as a Philosophical Problem”, Dialectica , 41: 221–230.
• –––, 1992. “Pseudoscience as nonsense”, Methodology and Science , 25: 91–101.
• Mahner, Martin, 2007. “Demarcating Science from Non-Science”, pp 515-575 in Theo Kuipers (ed.) Handbook of the Philosophy of Science: General Philosophy of Science – Focal Issues , Amsterdam: Elsevier.
• –––, 2013. “Science and pseudoscience. How to demarcate after the (alleged) demise of the demarcation problem”, pp. 29–43 in Pigliucci and Boudry (eds.) 2013.
• Mayo, Deborah G., 1996. “Ducks, rabbits and normal science: Recasting the Kuhn’s-eye view of Popper’s demarcation of science”, British Journal for the Philosophy of Science , 47: 271–290.
• Merton, Robert K., [1942] 1973. “Science and Technology in a Democratic Order”, Journal of Legal and Political Sociology , 1: 115–126, 1942; reprinted as “The Normative Structure of Science”, in Robert K. Merton, The Sociology of Science. Theoretical and Empirical Investigations , Chicago: University of Chicago Press, pp. 267–278.
• Moberger, Victor, 2020. “Bullshit, Pseudoscience and Pseudophilosophy”, Theoria , 86(5): 595–611.
• Morris, Robert L., 1987. “Parapsychology and the Demarcation Problem”, Inquiry , 30: 241–251.
• Oreskes, Naomi, 2019. “Systematicity is necessary but not sufficient: on the problem of facsimile science”, Synthese , 196(3): 881–905.
• Oreskes, Naomi and Erik M. Conway, 2010. Merchants of doubt: how a handful of scientists obscured the truth on issues from tobacco smoke to global warming , New York: Bloomsbury Press.
• Pennock, Robert T., 2010. “The postmodern sin of intelligent design creationism” Science and Education , 19(6–8): 757–778.
• –––, 2011. “Can’t philosophers tell the difference between science and religion?: Demarcation revisited”, Synthese , 178(2): 177–206.
• Pigliucci, Massimo, 2013. “The demarcation problem. A (belated) response to Laudan”, in Pigliucci and Boudry (eds.) 2013, pp. 9–28.
• Pigliucci, Massimo and Maarten Boudry (eds.), 2013. Philosophy of Pseudoscience. Reconsidering the demarcation problem. Chicago: Chicago University Press.
• Popper, Karl, 1962. Conjectures and refutations. The growth of scientific knowledge , New York: Basic Books.
• –––, 1974 “Reply to my critics”, in P.A. Schilpp, The Philosophy of Karl Popper (The Library of Living Philosophers, Volume XIV, Book 2), La Salle: Open Court, pp. 961–1197.
• –––, 1976. Unended Quest London: Fontana.
• –––, 1978. “Natural Selection and the Emergence of the Mind”, Dialectica , 32: 339–355.
• –––, [1989] 1994. “Falsifizierbarkeit, zwei Bedeutungen von”, pp. 82–86 in Helmut Seiffert and Gerard Radnitzky, Handlexikon zur Wissenschaftstheorie , 2 nd edition, München: Ehrenwirth GmbH Verlag.
• Powell, James, 2019. “Scientists reach 100% consensus on anthropogenic global warming”, Bulletin of Science, Technlogy and Society , 37(4): 183–184.
• Radner, Daisie and Michael Radner, 1982. Science and Unreason , Belmont CA: Wadsworth.
• Reisch, George A., 1998. “Pluralism, Logical Empiricism, and the Problem of Pseudoscience”, Philosophy of Science , 65: 333–348.
• Rothbart, Daniel, 1990 “Demarcating Genuine Science from Pseudoscience”, in Patrick Grim, ed, Philosophy of Science and the Occult , 2nd edition, Albany: State University of New York Press, pp. 111–122.
• Ruse, Michael, 1977. “Karl Popper’s Philosophy of Biology”, Philosophy of Science , 44: 638–661.
• –––, 2000. “Is evolutionary biology a different kind of science?”, Aquinas , 43: 251–282.
• Ruse, Michael (ed.), (1996). But is it science? The philosophical question in the creation/evolution controversy , Amherst, NY: Prometheus Books.
• Scott, P., Richards, E., and Martin, B., 1990. “Captives of controversy. The Myth of the Neutral Social Researcher in Contemporary Scientific Controversies”, Science, Technology, and Human Values , 15(4): 474–494.
• Settle, Tom, 1971. “The Rationality of Science versus the Rationality of Magic”, Philosophy of the Social Sciences , 1: 173–194.
• Siitonen, Arto, 1984. “Demarcation of science from the point of view of problems and problem-stating”, Philosophia Naturalis , 21: 339–353.
• Thagard, Paul R., 1978. “Why Astrology Is a Pseudoscience”, Philosophy of Science Association ( PSA 1978 ), 1: 223–234.
• –––, 1988. Computational Philosophy of Science , Cambridge, MA: MIT Press.
• Thurs, Daniel P. and Ronald L. Numbers, 2013. “Science, pseudoscience and science falsely so-called”, in Pigliucci and Boudry (eds.) 2013, pp. 121–144.
• Torcello, Lawrence, 2016. “The ethics of belief, cognition, and climate change pseudoskepticism: implications for public discourse”, Topics in Cognitive Science , 8: 19–48.
• Vollmer, Gerhard, 1993. Wissenschaftstheorie im Einsatz, Beiträge zu einer selbstkritischen Wissenschaftsphilosophie Stuttgart: Hirzel Verlag.
• Wazeck, Milena, 2009. Einsteins Gegner. Die öffentliche Kontroverse um die Relativitätstheorie in den 1920er Jahren . Frankfurt: campus.
• Williams, Nigel, 2005. “Heavyweight attack on climate-change denial”, Current Biology , 15(4): R109–R110.

Anthroposophy

• Hansson, Sven Ove, 1991. “Is Anthroposophy Science?”, Conceptus 25: 37–49.
• Staudenmaier, Peter, 2014. Between Occultism and Nazism. Anthroposophy and the Politics of Race in the Fascist Era , Leiden: Brill.
• James, Edward W, 1990. “On Dismissing Astrology and Other Irrationalities”, in Patrick Grim (ed.) Philosophy of Science and the Occult , 2nd edition, State University of New York Press, Albany, pp. 28–36.
• Kanitscheider, Bernulf, 1991. “A Philosopher Looks at Astrology”, Interdisciplinary Science Reviews , 16: 258–266.

Climate science denialism

• McKinnon, Catriona, 2016. “Should We Tolerate Climate Change Denial?”, Midwest Studies in Philosophy , 40(1): 205–216.
• Torcello, Lawrence, 2016. “The Ethics of Belief, Cognition, and Climate Change Pseudoskepticism: Implications for Public Discourse”, Topics in Cognitive Science , 8(1): 19–48.

Creationism

• Lambert, Kevin, 2006. “Fuller’s folly, Kuhnian paradigms, and intelligent design”, Social Studies of Science , 36(6): 835–842.
• Pennock, Robert T., 2010. “The postmodern sin of intelligent design creationism”, Science and Education , 19(6–8): 757–778.
• Ruse, Michael (ed.), 1996. But is it science? The philosophical question in the creation/evolution controversy , Prometheus Books.
• Matthews, Michael R., 2019. Feng Shui: Teaching about science and pseudoscience , Springer.

Holocaust denial

• Lipstadt, Deborah E., 1993. Denying the Holocaust: the growing assault on truth and memory , New York : Free Press.

Parapsychology

• Edwards, Paul, 1996. Reincarnation: A Critical Examination , Amherst NY: Prometheus.
• Flew, Antony, 1980. “Parapsychology: Science or Pseudoscience”, Pacific Philosophical Quarterly , 61: 100–114.
• Hales, Steven D., 2001. “Evidence and the afterlife”, Philosophia , 28(1–4): 335–346.

Psychoanalysis

• Boudry, Maarten, and Filip Buekens, 2011. “The epistemic predicament of a pseudoscience: Social constructivism confronts Freudian psychoanalysis”, Theoria , 77(2): 159–179.
• Cioffi, Frank, 1998. Freud and the Question of Pseudoscience . Chigago: Open Court.
• –––, 2013. “Pseudoscience. The case of Freud’s sexual etiology of the neuroses”, in Pigliucci and Boudry (eds.) 2013, pp. 321–340.
• Grünbaum, Adolf, 1979. “Is Freudian psychoanalytic theory pseudoscientific by Karl Popper’s criterion of demarcation?”, American Philosophical Quarterly , 16: 131–141.

Quackery and non–scientific medicine

• Jerkert, Jesper, 2013. “Why alternative medicine can be scientifically evaluated. Countering the evasions of pseudoscience”, in Pigliucci and Boudry (eds.) 2013, pp. 305–320.
• Smith, Kevin, 2012a. “Against homeopathy–a utilitarian perspective”, Bioethics , 26(8): 398–409.
• –––, 2012b. “Homeopathy is unscientific and unethical”, Bioethics , 26(9): 508–512.

How to cite this entry . Preview the PDF version of this entry at the Friends of the SEP Society . Look up topics and thinkers related to this entry at the Internet Philosophy Ontology Project (InPhO). Enhanced bibliography for this entry at PhilPapers , with links to its database.

• The Skeptic’s Dictionary , contains information, links and references about a wide variety of contested claims and phenomena.
• Committee for Skeptical Inquiry , the major international organization promoting scientific investigations of contested phenomena.
• Quackwatch , devoted to critical assessment of scientifically unvalidated health claims.
• Views of modern philosophers , a summary of the views that modern philosophers have taken on astrology, expanded from an article published in Correlation: Journal of Research into Astrology , 14/2 (1995): 33–34.

creationism | evolution | -->Freud, Sigmund --> | Kuhn, Thomas | Lakatos, Imre | -->logical positivism --> | natural selection | Popper, Karl | skepticism | Vienna Circle

Copyright © 2021 by Sven Ove Hansson < soh @ kth . se >

• Accessibility

Support SEP

Mirror sites.

View this site from another server:

• Info about mirror sites

The Stanford Encyclopedia of Philosophy is copyright © 2023 by The Metaphysics Research Lab , Department of Philosophy, Stanford University

Library of Congress Catalog Data: ISSN 1095-5054

Stack Exchange Network

Stack Exchange network consists of 183 Q&A communities including Stack Overflow , the largest, most trusted online community for developers to learn, share their knowledge, and build their careers.

Q&A for work

Connect and share knowledge within a single location that is structured and easy to search.

What is the value of unfalsifiable beliefs?

I understand the value of falsifiable beliefs. Often they make predictions, and are classified as scientific ideas, useful in churning out predictions.

But do unfalsifiable beliefs have any value? Philosophically or mathematically or otherwise? I know value is a little vague, but consider for example the concept of eternal return. If true it would have a large impact. But, then it's unfalsifiable. Can't be verified at all.

So a sentence A, if true would have a large impact, but it's unfalsifiable at the same moment, would you assume any importance to such sentences?

• falsifiability

• 4 Falsifiable and verifiable are two different things. "There is a unicorn" is verifiable but not falsifiable, "All unicorns are white" is falsifiable but not verifiable. Neither verifiable nor falsifiable beliefs can have pragmatic value, if they hold in most cases encountered in practice, heuristic value in developing new theories, ethical value in guiding behavior, etc. And if something does have an impact then that impact makes it testable in some way, even if not in the narrow verifiable/falsifiable sense. –  Conifold Commented Apr 25, 2018 at 1:51
• 1 I can see that an unfalsifiable belief may have some value, perhaps therapeutic or psychological. But I'm not sure there are any unfalsifiable beliefs. It would depend on how we define 'unfalsifiable'. if it means 'not demonstrably false' then there are many of them. If it means 'unfalsifiable by any means' then I'd suggest there are no such beliefs. . –  user20253 Commented Apr 25, 2018 at 9:38
• 1 Another word for "unfalsifiable beliefs" is axioms . The question then is equivalent to asking "do axioms have any theoretical value?". –  Steve Commented Jun 6, 2020 at 12:43

7 Answers 7

In many epistemologies, unfalsifiable beliefs are going to have great value. In large part, this is because of the following:

universal skepticism fails.

A good way to grasp this is to think about Descartes' project (or perhaps better stated his supposed project). A common claim is that Descartes is this radical skeptic bent on doubting everything, but if you read the Meditations carefully, he's already said this is not going to work by the end of the first meditation. In fact, Descartes' project turns into one where he believes that things which are "clear and distinct" are unfalsifiable.

Now, I mention this not because I think you need to agree with Descartes but to point out a pattern -- for many types of epistemologies, including Humeanism empiricism, Cartesian rationalism, and Kant's, you are going to have some beliefs that of critical importance that are not falsifiable, because these are the beliefs that make other beliefs both verifiable, possible, and falsifiable.

So for instance, Kant has a theory of mind for which he argues, but the main value of it is that this theory enables him to move past skepticism about objects outside the self and to explain how the self comprehends objects. Then we can argue about particular objects and about sensibles and other things. But if we don't get that off the ground, then we can't do much.

A later example might be James' critique of Clifford (from what I understand James' critique is rather unfair but that's not the point). The basic thrust is that you cannot have a principle of universal doubt that stands up to itself -- so you have to have at least something you take to be non-falsifiable to do anything.

Are there ways of trying to avoid this? Yes, you can try to compose a set of things where every particular is falsifiable but the set as a whole (or even if some fail) provides this support.

I understand the value of falsifiable beliefs. Often they make predictions, and are classified as scientific ideas, useful in churning out predictions. But do unfalsifiable beliefs have any value? Philosophically or mathematically or otherwise? I know value is a little vague, but consider for example the concept of eternal return. If true it would have a large impact. But, then it's unfalsifiable. Can't be verified at all.

Verifiability and falsfiability are not the same. An idea is falsifiable if some experiment can be performed whose results could contradict that idea. The idea is verifiable if some experiment can be performed whose results could prove the idea, or make it more probable. For more explanation, see

Is falsificationism a reliable scientific methodology?

Now, you ask if unfalsifiable ideas are useful. Suppose that you say unfalisifiable ideas are useless. Is that a falsifiable idea? No. It's an idea about what you should do, not about what people actually do. Also, what about the standard by which you judge the success of falsification? More generally, all methodological or moral ideas are unfalsifiable. Since these ideas are useful, unfalsifiable ideas are useful.

But do unfalsifiable beliefs have any value

Religion is often take to be a body of unverifiable and unfalsifiable beliefs; William James in the final chapter of his The Varieties of Religious Experience offers a defence of such in the context of the human community. I'm not going to be able to summarise his argument, but I'd urge you to have a look at it.

While I really like Virmaior's answer, I would like to add a few points, mainly regarding the ideas of verifiability and falsifiability as they relate to scientific theories.

Prior to the turn of the last century, it was nearly universally believed that scientific theories could be proven in the same way as theorems in mathematics could. The surprising collapse of Newtonian mechanics, Newtonian gravity, and classical electrodynamics caused philosophers to realize that no amount of verification can lead to a guarantee that the next experiment won't produce a null result.

Karl Popper recognized that theories were indeed not provable, but maintained that they (or at least the good ones) were falsifiable. This idea is still deeply entrenched in popular opinion and is still commonly offered as a solution to the problem of demarcation. However, the claim that scientific theories are falsifiable does not hold up under scrutiny.

Supporting Details

As Thomas Kuhn points out, in his book ' The Structure of Scientific Revolutions ,' the scientific community will go to great lengths to prevent an accepted theory from being falsified. Take the famous example of the observation of perturbations in planet Uranus' orbit , which were not predicted by Newtonian mechanics. These discrepancies between observation and theory were known for nearly 70 years. As Kuhn points out, the scientific community does not reject a theory the first time there is a null result. In the case of the orbit of Uranus, the scientific community, rather than rejecting Newtonian mechanics and Newton's law of gravity, instead hypothesized the existence of a yet undiscovered planet: Neptune. They even calculated the exact location where the new planet would have to exist in order to explain the inconsistencies, which lead to the discovery of Neptune—a great achievement for Newtonian physics. However, never during that 70 year period (between the discovery of the discrepancy and the discovery of Neptune) did the scientific community ever consider rejecting Newtonian mechanics.

A few years after that, the existence of the planet Vulcan was hypothesized to explain persistent irregularities in the orbit of Mercury (irregularities that had first been observed a century earlier). However, unlike Neptune, the planet Vulcan was never discovered, and Mercury's orbit was only explained after Einstein published his theory of general relativity in 1916.

A third example, from antiquity, was the inclusion of epicycles by Ptolemy in the Aristotelian system of astronomy to account for irregularities in the observed orbit of the planets. All three of these examples highlight the scientific communities ability to defend a theory in the face of inconsistent experimental data.

The point of these three examples is to show that the scientific community can always tweak the theory or tweak auxiliary hypotheses (i.e posit a yet undiscovered planet) in the face of incongruent data.

Moreover, even when a theory is 'falsified,' there is no guarantee that it won't come back to life a century later. Take, for example, the particle (corpuscular) theory of light, which was supposedly falsified in 1819 when the French physicist Dominique-François-Jean Arago observed a bright spot at the center of the shadow of a circular disk (a bizarre prediction of Fresnel's wave-theory put forth to discredit the theory). Resistance to the wave theory collapsed and by all accounts, the particle theory of light was completely and utterly destroyed. However, fast-forward to 1905, and the particle theory is resurrected by Albert Einstein to explaining the photoelectric effect and ultraviolet catastrophe.

The point is that the scientific community can always explain away a null result by challenging one or more of the auxiliary hypotheses, and/or by adjusting the theory to account for the results. And even if the scientific community agrees that a theory is falsified, it still might be resurrected at some point in the future new and unforeseen reasons.

• The only conclusion you can draw is that just as you can not prove scientific theories, you also cannot disprove or falsify them either. A 'proven' theory might turn out to be wrong just as a 'falsified' theory might turn out to be correct in some way we could never imagine at the time.

• I think you're misusing the term "falsified", it does not mean to be proved wrong, rather the existence of some hypothetical scenario under which it(the theory) might be proved wrong. –  novice Commented Apr 28, 2018 at 23:09
• @novice, your comment makes no sense to me. It sounds like you are saying 'falsified' does not mean proven wrong, but rather has the potential to be proven wrong? Is this what you are saying? –  njs Commented Apr 29, 2018 at 0:14
• Yes. Correct me if I am wrong. –  novice Commented Apr 29, 2018 at 1:35
• 'Falsified' means to prove false. 'Falsifiable' means that it could be proven false. I'm using these terms the way that Popper used them. According to Popper, good scientific theories are falsifiable and are often falsified (e.g. Newtonian mechanics and classical electrodynamics). He claimed that falsifiability was the key to the problem of demarcation: good theories, make bold (and risky) claims about the world, claims that are testable—special relativity was the perfect example for him. Pseudoscientific theories, on the other hand, explain everything but are never falsified by any evidence. –  njs Commented Apr 29, 2018 at 2:20
• 2 I may edit my original answer to make it more readable to someone not familiar with the subject matter. Basically, the point is that scientific claims (or beliefs) are neither provable nor falsifiable. These terms are really only applicable to math, logic, and geometry. In science, it's a lot messier: evidence strengthens or weakens a theory but (in general) never proves or disproves it. This had to be learned the hard way: theories that were once thought to be proven turned out to be false, and theories that were considered disproven turned out to be true at a later point. –  njs Commented Apr 29, 2018 at 3:34

The idea of eternal return might be falsifiable:

While the big bang theory in the framework of relativistic cosmology seems to be at odds with eternal return, there are now many different speculative big bang scenarios in quantum cosmology which actually imply eternal return... (Wikipedia, “Eternal return”)

So, observations available today leave the question open. But future data, gathered by improved instruments, might make eternal return an untenable idea; such a process eventually sank the theories of phlogiston and interplanetary ether. Or future data might show that eternal return is indeed verifiable, and, like relativity in 1905, waits only for the experiment that will confirm its truth.

But, then it's unfalsifiable. Can't be verified at all.

Those are not the same things. Falsifiability only matters for claims that are false. Verifiability only matters for claims that are true. See the following truth table, with truth values making the columns and falsifiability or verifiability in the rows:

No one has ever come up with a criterion that would falsify that humans can fly. "Scientific" consensus used to be that man would never fly. But it turns out we can fly. Here we have one example of a true statement that is unfalsifiable, and has been thoroughly verified and reduced to practice.

So yes, there is great value in millions of propositions that cannot be falsified. There are open research questions right now that are worth millions of dollars each.

And falsifiability does not tell the whole story of testability--not by a long shot.

Unfalsifiable beliefs are common in science. Often scientific theories don't start out by being falsifiable. For instance Wegner's theory of continental drift , which initially lacked a physical mechanism, which makes it hard to falsify. Science often starts out with a hypothesis generation phase, which is based on the scientists belief about the way things work, which they then refine and try to develop falsifiable predictions. An example of science where this is currently the case is Eternal Inflation (multiverses) for which we may not be able to obtain any evidence. Similarly we can have beliefs for what exists outside the observable universe, but that will be unfalsifiable.

Science is a search for the best explanation of the universe, we can't ignore unfalsifiable explanations if they have e.g. strong consilience with accepted falsifiable and tested theories. It is one way in which the boundaries of science are extended by plausible extrapolation of what we "know". Hardline naive falsificationism is not a good approach to science.

I think the idea that we can have completely certain verifiable knowledge of the real world is also a non-starter - there will always be uncertainty.

So unfalsifiable theories can be useful in science, the field that supposedly is most hostile to them.

You must log in to answer this question.

Not the answer you're looking for browse other questions tagged falsifiability ..

• Featured on Meta
• User activation: Learnings and opportunities
• Site maintenance - Mon, Sept 16 2024, 21:00 UTC to Tue, Sept 17 2024, 2:00...
• 2024 Community Moderator Election Results

Hot Network Questions

• Is this a misstatement of Euclid in Halmos' Naive Set Theory book?
• Big Transition of Binary Counting in perspective of IEEE754 floating point
• Getting lost on a Circular Track
• How to make conditions work in Which?
• What would the natural diet of Bigfoot be?
• Should I write an email to a Latino teacher working in the US in English or Spanish?
• Two sisters live alone in a house after the rest of their family died
• Is this map real?
• Why are my empty files not being assigned the correct mimetype?
• Why does documentation take a large storage?
• What is the unit for 'magnitude' in terms of the Isophotal diameter of a galaxy?
• For glacier winds to exist, are circulation cells needed?
• LaTeX labels propositions as Theorems in text instead of Propositions
• Is Produce Flame a spell that the caster casts upon themself?
• What does the phrase 'sons of God'/בני אלוהים mean throughout the Hebrew bible?
• Is it a correct rendering of Acts 1,24 when the New World Translation puts in „Jehovah“ instead of Lord?
• Concerns with newly installed floor tile
• Does any row of Pascal's triangle contain a Pythagorean triple?
• Why would the GPL be viral, while EUPL isn't, according to the EUPL authors?
• HHL Phase estimation step on non-eigenvector
• When deleting attribute from GDB file all the fields in the remaining attributes get deleted as well in QGIS
• What came of the Trump campaign's complaint to the FEC that Harris 'stole' (or at least illegally received) Biden's funding?
• How can a microcontroller (such as an Arduino Uno) that requires 7-21V input voltage be powered via USB-B which can only run 5V?
• What prevents indoor climbing gyms from making a v18 boulder even if one hasn't been found outside?

UNFALSIFIABLE

designating the quality of a hypothesis, proposition , or theory such that no empirical test can mandate that it is untrue.

Leave a Reply

Your email address will not be published. Required fields are marked *

Latest Posts

What Happens At An ADHD Assessment

A Quick Look at the History Behind Hypnosis

A Brief History of Brainwashing: The Science of Thought Control

A Deep Dive into the Social Psychology of Leadership

Counseling Approaches to Client Care: Theories to Apply in Practice

The Future Of Education: Can You Earn A Psychology Degree Online?

Insomnia & Mental Illness: What is the Correlation?

Stop Guessing: Here Are 3 Steps to Data-Driven Psychological Decisions

Getting Help with Grief: Understanding Therapy & How It Can Help

Exploring the Psychology of Risk and Reward

Understanding ADHD in Women: Symptoms, Treatment & Support

Meeting the Milestones: A Guide to Piaget's Child Developmental Stages

Popular psychology terms, medical model, hypermnesia, affirmation, backup reinforcer, brainwashing, affiliative behavior, message-learning approach, kinesthetic feedback, sensory adaptation, spontaneous neural activity.

Falsifiability in medicine: what clinicians can learn from Karl Popper

• From the Inside
• Published: 22 May 2021
• Volume 47 , pages 1054–1056, ( 2021 )

Cite this article

• Shaurya Taran   ORCID: orcid.org/0000-0001-7639-0365 1 ,
• Neill K. J. Adhikari   ORCID: orcid.org/0000-0003-4038-5382 2 , 3 &
• Eddy Fan   ORCID: orcid.org/0000-0002-1210-9914 3 , 4  

4860 Accesses

6 Citations

44 Altmetric

Explore all metrics

A Correction to this article was published on 17 June 2021

This article has been updated

Avoid common mistakes on your manuscript.

This isn’t right. It’s not even wrong!

-Wolfgang Pauli.

In the early twentieth century, the philosopher Karl Popper became intrigued by a basic question in the philosophy of science: how does one distinguish true science from non-science? For Popper, the distinction hinged on the essential ingredient of falsifiability [ 1 ]. True science was falsifiable: it could be proven incorrect by an experiment that contradicted its predictions. Non-science, on the other hand, was unfalsifiable: it made no predictions that could be disproven by experimental methods. Popper highlighted the difference using Einstein’s theory of general relativity and Freud’s theory of psychoanalysis as examples. Einstein’s theory inferred specific claims about the natural world. It invited experimentation and set itself up to be either corroborated or falsified by experiments. By contrast, Freud’s theory used observations to posit a general theory about human nature, but for a given patient, it made no specific predictions. Since no experiment could be put forth to contradict it, Popper regarded it as unfalsifiable . This concept was also famously highlighted by the physicist Wolfgang Pauli, who, when asked to review a paper that he deemed to be unfalsifiable, lamented, “This isn’t right. It’s not even wrong!”

Although Popper’s theories about the essence of science have been challenged, the core concept of falsifiability to assess whether a claim is scientific has endured. For clinicians, Popper’s notion might be used to evaluate new ideas and decide how much weight to give them. This framework is important because clinicians today, more than ever before, face an explosion of ideas of variable quality, making it difficult to know where to place one’s trust. There are so many provocative conjectures and expert opinions being circulated that the task of carefully assessing their credibility has never been more necessary. The current coronavirus disease 2019 (COVID-19) pandemic has offered several examples of conjectures widely adopted without rigorous evaluation, sometimes leading to patient harm. In light of this chaos, what lessons might Popper’s notion of falsifiability hold for clinicians, and how can these lessons help clinicians become better judges of science?

First, there is a useful difference between conjectures and theories. Conjectures about medicine may stem from uncontrolled clinical observations in patients, physiological experiments, or animal models—sources of evidence whose trustworthiness is often downgraded because of high risk of bias, indirectness to actual clinical problems in patients, or imprecision of the estimated treatment effect [ 2 ]. On their own, conjectures may be difficult to falsify, because they generate no new predictions, or because they rely on a series of linked conjectures to generate testable hypotheses. By contrast, tentative theories emerge from a preponderance of data from internally valid studies whose results point in a consistent direction. Theories can be corroborated or falsified by high-quality tests, such as randomized clinical trials (see Fig. 1 for additional differences between conjectures and theories). This distinction bears reminding in the context of the ongoing pandemic, where conjectures have often shifted clinical practice in a manner out of proportion to the certainty of evidence. Consider the example of hydroxychloroquine, which was touted as a breakthrough treatment for COVID-19 on the basis of in vitro studies demonstrating anti-viral activity [ 3 ] and studies of fewer than 30 patients that showed reduced viral nasopharyngeal carriage [ 4 ]. In France, where one of these studies was performed, prescriptions of hydroxychloroquine surged. Notwithstanding the fact that multiple subsequent clinical trials showed no benefit, and possibly increased harm, associated with hydroxychloroquine, it is unsettling that early conjectures were rapidly adopted across the medical community before a theory could emerge and a fuller understanding of its risks and benefits be appreciated. When evaluating a new idea, Popper thus encourages clinicians to ask the following questions: does the available body of knowledge describe a conjecture or a theory? Have the data been evaluated by a high-quality test? And if not, could it theoretically be corroborated or falsified by such a test in the future, provided that there remains equipoise about the intervention [ 5 ]? These questions can help clinicians decide if an idea is in a rudimentary or more advanced phase of development—and accordingly, whether it deserves further testing or is ready for application.

Differences between conjectures and theories

Second, although falsifiability is a binary concept (an idea is either falsifiable or it isn’t), theories are more complex: they might be completely true under some  conditions, completely untrue under others, or partially true depending on which aspects are considered. When interpreting new studies, it is therefore important to appreciate these nuances and resist the tendency to oversimplify. Consider the example of dexamethasone for the treatment of patients with COVID-19. In June 2020, preliminary results from the RECOVERY trial were released, which showed that in mechanically ventilated patients with COVID-19, treatment with dexamethasone resulted in a 11.7% absolute risk reduction in 28-day mortality compared to usual care [ 6 ]. Among non-ventilated patients receiving oxygen, dexamethasone resulted in a less pronounced, but significant, absolute risk reduction in 28-day mortality of 3.5%. However, in the group of patients not receiving oxygen or mechanical ventilation, no mortality benefit with dexamethasone was observed. These landmark findings were rapidly communicated throughout the lay press, often with the simple bottom-line message that dexamethasone saved lives, without further elaboration into the groups most likely to benefit or not benefit at all. In reality, the role of corticosteroids in COVID-19 is far more nuanced, with a differential response depending on disease severity. To ignore such complexities—as has been done by politicians and policymakers on various matters throughout this pandemic—misrepresents the truth and propagates misunderstandings. It is therefore always worth asking what the study corroborated or falsified before making a judgment about the theory on the whole.

Third, theories which have accumulated a wealth of evidence, generally over a long period of time and by many examiners, have withstood Popper’s falsifiability test, perhaps many times over: they are the best approximators of real truth. In Bayesian terms, ideas with a long history of consistent messaging offer a reliable set of priors to understand and evaluate new evidence. This concept could be remembered, for example, when one is confronted by ideas that do not fit with established priors. Consider the example regarding the existence of “H” and “L” phenotypes of COVID-19-related acute respiratory distress syndrome (ARDS) [ 7 ]. Investigators hypothesized that there are two distinct COVID-19 ARDS phenotypes (with a spectrum in between) that mandate different approaches to mechanical ventilation. They suggested that not identifying the correct phenotype might lead to selection of the wrong ventilation approach and patient harm. These hypothetical phenotypes were not previously demonstrated to exist, but more importantly, the ventilation strategy proposed for “L”-type patients (i.e., use of high tidal volumes) contradicts decades of ARDS research, which has shown multiple benefits in favor of lower tidal volume ventilation [ 8 ]. This is not to say that the “H” vs “L” phenotype conjecture should not be subject to further testing—but rather, it is difficult to justify abandoning a robust set of priors when a single new concept challenges them. As Popper might argue, the preponderance of existing evidence on an idea should guide clinicians in deciding where to place their trust while awaiting the results of additional investigations.

Popper applied the notion of falsifiability to distinguish between non-science and science. Clinicians might apply the same notion to understand and evaluate new ideas. This process entails three key considerations. First, conjectures should be seen as invitations to design further studies to evaluate them. While many conceptually interesting new conjectures move their fields in a novel direction, they often still require confirmation by high-quality studies, and their application to patient care might be premature (or even harmful) before such validation occurs. Second, for theories that have been apparently corroborated or falsified by high-quality tests, it is still worth asking: what did the tests actually show? Did they prove or disprove the entire theory, or only some aspect of it? This interpretation of data, formalized in the GRADE system of assessing certainty of evidence [ 2 ], should guide the application of research findings into the real world. Finally, when evaluating an idea for which there is existing knowledge, it is worth placing the idea in its available context. Ample time also provides ample opportunities for falsification. Ideas that have withstood Popper’s test are probably robust—and more likely to be “true” as compared to the “new truths” with which we are all nowadays regularly confronted.

Change history

17 june 2021.

A Correction to this paper has been published: https://doi.org/10.1007/s00134-021-06457-4

Popper K (1968) The Logic of Scientific Discovery, 7th edn. Harper & Row, New York

Google Scholar  

Guyatt GH, Oxman AD, Vist GE et al (2008) GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ 336:924–926

Article   Google Scholar  

Yao X, Ye F, Zhang M et al (2020) In vitro antiviral activity and projection of optimized dosing design of hydroxychloroquine for the treatment of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Clin Infect Dis 71:732–739

Article   CAS   Google Scholar  

Gautret P, Lagier JC, Parola P et al (2020) Hydroxychloroquine and azithromycin as a treatment of COVID-19: results of an open-label non-randomized clinical trial. Int J Antimicrob Agents 56:105949

Yeh RW, Valsdottir LR, Yeh MW et al (2018) Parachute use to prevent death and major trauma when jumping from aircraft: randomized controlled trial. BMJ 363:5094

Horby P, Lim WS, Emberson J, et al. Effect of Dexamethasone in Hospitalized Patients with COVID-19 – Preliminary Report. medRxiv 2020:2020.06.22.20137273

Gattinoni L, Chiumello D, Caironi P et al (2020) COVID-19 pneumonia: different respiratory treatments for different phenotypes? Intensive Care Med 46:1099–1102

Fan E, Del Sorbo L, Goligher EC et al (2017) An official american thoracic society/european society of intensive care medicine/society of critical care medicine clinical practice guideline: mechanical ventilation in adult patients with acute respiratory distress syndrome. Am J Respir Crit Care Med 195:1253–1263

Download references

Author information

Authors and affiliations.

Interdepartmental Division of Critical Care Medicine, Li Ka Shing Knowledge Institute, University of Toronto, 204 Victoria Street, 4th floor room 411, Toronto, ON, M5B 1T8, Canada

Shaurya Taran

Sunnybrook Health Sciences Centre, Toronto, ON, Canada

Neill K. J. Adhikari

Institute for Health Policy, Management, and Evaluation, University of Toronto, Toronto, ON, Canada

Neill K. J. Adhikari & Eddy Fan

Toronto General Hospital, University Health Network, Toronto, ON, Canada

You can also search for this author in PubMed   Google Scholar

Contributions

All authors were involved in article conception, manuscript preparation, and critical revisions.

Corresponding author

Correspondence to Shaurya Taran .

Ethics declarations

Conflicts of interest.

All authors declare that they have no competing conflicts of interest.

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

The original online version of this article was revised: Due to an error in an author name.

Rights and permissions

Reprints and permissions

About this article

Taran, S., Adhikari, N.K.J. & Fan, E. Falsifiability in medicine: what clinicians can learn from Karl Popper. Intensive Care Med 47 , 1054–1056 (2021). https://doi.org/10.1007/s00134-021-06432-z

Download citation

Received : 03 April 2021

Accepted : 07 May 2021

Published : 22 May 2021

Issue Date : September 2021

DOI : https://doi.org/10.1007/s00134-021-06432-z

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Find a journal
• Publish with us
• Track your research

• Technical support

falsifiability

• Robert Sheldon
• Ivy Wigmore

What is falsifiability?

Falsifiability is the capacity for some proposition, statement, theory or hypothesis to be proven wrong. The concept of falsifiability was introduced in 1935 by Austrian philosopher and scientist Karl Popper (1902-1994). Since then, the scientific community has come to consider falsifiability to be one of the fundamental tenets of the scientific method , along with attributes such as replicability and testability.

A scientific hypothesis, according to the doctrine of falsifiability, is credible only if it is inherently falsifiable. This means that the hypothesis must be capable of being tested and proven wrong. It does not automatically mean that the hypothesis is invalid or incorrect, only that the potential exists for the hypothesis to be refuted at some possible time or place.

For example, one could hypothesize that a divine being with green scales, mauve hair, ochre-colored teeth and a propensity for humming show tunes rules over the physical universe from a different dimension. Even if millions of people were to swear their allegiance to such a being, there is no practical way to disprove this hypothesis, which means that it is not falsifiable. As a result, it cannot be considered a scientific assertion, according to the rules of falsifiability.

On the other hand, Einstein's theory of relativity is considered credible science according to these rules because it could be proven incorrect at some point in time through scientific experimentation and advanced testing techniques, especially as the methods continue to expand our body of knowledge. In fact, it's already widely accepted that Einstein's theory is at odds with the fundamentals of quantum mechanics, not unlike the way Newton's theory of gravity could not fully account for Mercury's orbit.

Another implication of falsifiability is that conclusions should not be drawn from simple observations of a particular phenomenon . The white swan hypothesis illustrates this problem. For many centuries, Europeans saw only white swans in their surroundings, so they assumed that all swans were white. However, this theory is clearly falsifiable because it takes the discovery of only one non-white swan to disprove its hypothesis, which is exactly what occurred when Dutch explorers found black swans in Australia in the late 17th century.

Falsifiability is often closely linked with the idea of the null hypothesis in hypothesis testing. The null hypothesis states the contrary of an alternative hypothesis. It provides the basis of falsifiability, describing what the outcome would demonstrate if the prediction of the alternative hypothesis is not supported. The alternative hypothesis might predict, for example, that fewer work hours correlates to lower employee productivity. A null hypothesis might propose that fewer work hours correlates with higher productivity or that there is no change in productivity when employees spend less time at work.

Popper makes the case for falsifiability

Karl Popper introduced the concept of falsifiability in his book The Logic of Scientific Discovery (first published in German in 1935 under the title Logik der Forschung ). The book centered on the demarcation problem, which explored the difficulty of separating science from pseudoscience . Popper claimed that only if a theory is falsifiable can it be considered scientific. In contrast, areas of study such as astrology, Marxism or even psychoanalysis were merely pseudosciences.

Popper's theories on falsifiability and pseudoscience have had a significant impact on what is now considered to be true science. Even so, there is no universal agreement about the role of falsifiability in science because of the limitations inherent in testing any hypothesis. Part of this comes from the fact that testing a hypothesis often brings its own set of assumptions, as well as an inability to account for all the factors that could potentially impact the outcome of a test, putting the test in question as much as the original hypothesis.

In addition, the tests we have at hand might be approaching their practical limitations when up against hypotheses such as string theory or multiple universes. It might not be possible to ever fully test such hypotheses to the degree envisioned by Popper. The question also arises whether falsifiability has anything to do with actual scientific discovery or whether the theory of falsification is itself falsifiable.

No doubt many researchers would argue that their brand of social or psychological science meets a set of criteria that is equally viable as those laid out by Popper. Even so, the important role that falsifiability has played in the scientific model cannot be denied, but Popper's black-and-white demarcation between science and pseudoscience might need to give way to a more comprehensive perspective of what we understand as being scientific.

See also:  empirical analysis ,  validated learning ,  OODA loop , black swan event,  deep learning .

Continue Reading About falsifiability

• 15 common data science techniques to know and use
• The data science process: 6 key steps on analytics applications
• How evolutionary architecture simplified hypothesis driven development
• 8 types of bias in data analysis and how to avoid them
• 14 most in-demand data science skills you need to succeed

Related Terms

In general, asynchronous -- from Greek asyn- ('not with/together') and chronos ('time') -- describes objects or events not ...

A URL (Uniform Resource Locator) is a unique identifier used to locate a resource on the internet.

File Transfer Protocol (FTP) is a network protocol for transmitting files between computers over TCP/IP connections.

Network detection and response (NDR) technology continuously scrutinizes network traffic to identify suspicious activity and ...

Identity threat detection and response (ITDR) is a collection of tools and best practices aimed at defending against cyberattacks...

Managed extended detection and response (MXDR) is an outsourced service that collects and analyzes threat data from across an ...

A software license is a document that provides legally binding guidelines for the use and distribution of software.

Data storytelling is the process of translating complex data analyses into understandable terms to inform a business decision or ...

Demand shaping is an operational supply chain management (SCM) strategy where a company uses tactics such as price incentives, ...

Employee self-service (ESS) is a widely used human resources technology that enables employees to perform many job-related ...

Diversity, equity and inclusion is a term used to describe policies and programs that promote the representation and ...

Payroll software automates the process of paying salaried, hourly and contingent employees.

Salesforce Developer Experience (Salesforce DX) is a set of software development tools that lets developers build, test and ship ...

Salesforce Commerce Cloud is a cloud-based suite of products that enable e-commerce businesses to set up e-commerce sites, drive ...

Multichannel marketing refers to the practice of companies interacting with customers via multiple direct and indirect channels ...