I understand that having beliefs that are falsifiable in principle and make predictions about experience is incredibly important. But I have always wondered if my belief in falsifiability was itself falsifiable. In any possible universe I can imagine it seems that holding the principle of falsifiability for our beliefs would be a good idea. I can’t imagine a universe or an experience that would make me give this up.
How can I believe in the principle of falsifiability that is itself unfalsifiable?! I feel as though something has gone wrong in my thinking but I can’t tell what. Please help!
Excellent, because it illustrates the problem with “believing in” the principle of falsifiability, as opposed to using it and understanding how it relates to the rest of my thinking.
Forget that the principle of falsifiability is itself incredibly important. What sorts of beliefs does the principle of falsifiability tell me to increase my confidence in? To decrease my confidence in?
What would the world have to be like for the former beliefs to be in general less likely than the latter?
Thanks for the reply Dave. Are you saying I should not look at falsifiability as a belief, but rather a tool of some sort? That distinction sounds interesting but is not 100% clear to me. Perhaps someone should do a larger post about why the principle should not be applied to itself.
I have also thought of putting the problem this way: Eliezer states that the only ideas worth having are the ones we would be willing to give up. Is he willing to give up that idea? I don’t think so..., and I would be really interested to know why he doesn’t believe this to be a contradiction.
What I’m saying is that the important thing is what I can do with my beliefs. If the “principle of falsifiability” does some valuable thing X, then in worlds where the PoF doesn’t do X, I should be willing to discard it. If the PoF doesn’t do any valuable thing X, then I should be willing to discard it in this world.
It seems we have empirical and non-empirical beliefs that can both be rational, but what we mean by “rational” has a different sense in each case. We call empirical beliefs “rational” when we have good evidence for them, we call non-empirical beliefs like the PoF “rational” when we find that they have a high utility value, meaning there is a lot we can do with the principle (it excludes maps that can’t conform to any territory).
To answer my original question, it seems a consequence of this is that the PoF doesn’t apply to itself, as it is a principle that is meant for empirical beliefs only. Because the PoF is a different kind of belief from an empirical belief, it need not be falsifiable, only more useful than our current alternatives.
What do you think about that?
If it can be restated as “I will on average be more effective at achieving my goals if I only adopting falsifiable beliefs,” for example, then it is equivalent to an empirical belief (and is, incidentally, falsifiable).
If it can be restated as “I should only adopt falsifiable beliefs, whether doing so gets me anything I want or not” then there exists no empirical belief to which it is equivalent (and is, incidentally, worth discarding).
How can I believe in the principle of falsifiability that is itself unfalsifiable?! I feel as though something has gone wrong in my thinking but I can’t tell what.
You have just refuted the contention that all warranted beliefs must be falsifiable in principle. Karl Popper, who introduced the falsifiability criterion and pushed it as far if not further than it can go, never advocated that all beliefs should be falsifiable. Rather, he used falsifiability as the criterion of demarcation between science and non-science, while denying that all beliefs should be scientific. His contention that falsifiability demarcates science does imply, as he recognized, that the criterion of falsifiability is not itself a scientific hypothesis.
Rational beliefs are not necessarily scientific beliefs. Mathematics is rational without being falsifiable. The same is true of philosophical beliefs, such as the belief that scientific beliefs are falsifiable. But rational beliefs that are not scientific must be refutable, and falsifiable beliefs are a proper subset of refutable beliefs. Falsifiable beliefs are refutable in one particular way: they are refutable by observation statements, which I think are equivalent to EY’s anticipations. Science is special because it is 1) empirical (unlike mathematics) and 2) has an unusual capacity to grow human knowledge systematically (unlike philosophy). But that does not imply that we can make do with scientific beliefs exclusively, one reason being the one that you mention about criteria for the acceptance of scientific theories.
The broader criterion of refutability doesn’t necessarily involve refutation by observation statements. How would you refute the falsifiability criterion? It would be false if science it were the case that scientists secured the advance of science by using some other criteria (such as verification).
It’s a mistake to conflate the questions of whether a theory is scientific and whether it’s corroborated (by attempted falsifications). Or to conflate whether it’s scientific or it’s rationally believable. Theories aren’t bad because they aren’t science. They’re bad because they’re set up so they resist any form of refutation. Rational thought involves making your thinking vulnerable to potential refutation, rather than protecting it from any refutation.In science, the mode of refutation is observation, direct connection to sensory data. But it won’t do (as you’ve realized by trying to apply falsifiability to itself) to limit one’s thinking entirely to that which is falsifiable.
You later ask (in effect) whether the refutability criterion is itself even refutable. Would EY be willing, ever, to give it up? He should be, were someone to show that sheer dogmatism conduces to the growth of knowledge. That I can’t conceive of a plausible argument to that end doesn’t obviate the refutability of the contention
I think that resolves your confusion, but I don’t want to imply that Popper uttered the last word—there are problems with neglecting verification in favor of strict falsificationism.
What are the criteria that we use for accepting or refuting rational non-empirical beliefs? You mention that falsifiability would be refuted if some other criteria “secured the advance of science.” You also mention that we should give up the refutability criterion if “sheer dogmatism conduces to the growth of knowledge.” It sounds like our criteria for the refutability of non-empirical beliefs are mostly practical; we accept the epistemic assumptions that make things “work best.” Is there more to it than this?
To be pedantic and Popperian, I’d have to correct your use of “empirical beliefs.” The philosophical positions at issue aren’t scientific but they are empirical. “Empirical”—to be the basis for scientific observation statements—must be expressible in low-level observation sentences that all competent scientists agree on.
The belief in question is that science’s crucial distinguishing feature allowing it to advance is the subjection of science’s claims to empirical testing, allowing strict falsification. We can’t run an experiment or otherwise record observation statements, so we resort to philosophical debate aimed at refutation. Refutation is obtained by plausible argument. For instance, in the discussion about demarcation, an example of a potentially plausible argument goes if we relied on falsification exclusively, we would never have evidence that a claim is true, only that it isn’t false. But we rely on scientific theories and consider them close to the truth (or at least as probably so). Therefore, falsifiability can’t explain the distinctiveness of science.
This involves highly plausible claims, based on observation, about how we in fact use scientific theories. But although the result of observation, it can’t be reduced to something everyone agrees on that is closely tied to direct perception, as with an observation statement.
For me the principle of falsifiability is best understood as a way of distinguishing scientific theories about the world from other theories about the world. In other words, falsifiability is one way of defining what science is and is not. A theory that does not constrain experience (“God works in mysterious ways”) is not a scientific theory because it can explain any occurrence and is therefore not falsifiable.
Because falsifiability is a definition, not a theory about the world, there’s no reason to think it can be falsified. The definition could be wrong by failing to accurately or usefully define scientific theory, but that’s conceptually different.
For me the principle of falsifiability is best understood as a way of distinguishing scientific theories about the world from other theories about the world. In other words, falsifiability is one way of defining what science is and is not. A theory that does not constrain experience (“God works in mysterious ways”) is not a scientific theory because it can explain any occurrence and is therefore not falsifiable.
Because falsifiability is a definition, not a theory about the world, there’s no reason to think it can be falsified. The definition could be wrong by failing to accurately or usefully define scientific theory, but that’s conceptually different.
Falsifiability is a very bad way to define science (or scientific theories). If falsifiability was all it took for a theory to be scientific, then all theories known to be false would be scientific (after all, if something is known to be false, it must be falsifiable). Do we really want a definition of science that says astrology is science because it’s false?
I don’t think the current line of enquiry is particularly useful.
“Astrology works” is a scientific theory to the degree that it is, in fact, acceptable science to do an experiment to see whether or not astrology has predictive power. It’s rhetorically inaccurate to say that means “astrology is science” though, because of course the practice of astrology is not. But sure, it’s probably a good idea to include other conditions. Excessively unlikely (or non-reductionist?) hypotheses could be classified as non-scientific, for the simple reason that even considering them in the first place would be a case of privileging the hypothesis.
None of this contradicts falsifiability being “a way of distinguishing scientific theories about the world from other theories about the world”, if we have other ways of distinguishing scientific from non-scientific, such as “reductionism”.
Astrology does seem to consist of scientific hypotheses.
I chose astrology because it has a reverse halo effect around here (and so would serve me rhetorically). Feel free to replace it with any other known to be false set of propositions.
I agree that falsifiability is not a complete definition. My point was only that falsifiability is not applicable to the principle of falsifiability, any more than it applies to mathematics.
That said, Newton’s physics and geocentric theories are false. Are they not science simply for that reason?
I agree that falsifiability is not a complete definition. My point was only that falsifiability is not applicable to the principle of falsifiability, any more than it applies to mathematics.
Yes. Falsifiability is a poor definition of science and is self-undermining in the sense that it can’t pass its own test.
That said, Newton’s physics and geocentric theories are false. Are they not science simply for that reason?
Of course not. I’m not claiming a scientific theory must be true. I’m claiming that known falseness (which implies falsifiability) is not a sufficient condition for being scientific.
A theory that does not constrain experience (“God works in mysterious ways”) is not a scientific theory because it can explain any occurrence and is therefore not falsifiable.
That statement does not itself constrain experience. That’s not a useful critique of the statement.
I’m claiming that known falseness (which implies falsifiability) is not a sufficient condition for being scientific.
Know falseness is not really same thing as falsifiability. Known falseness is useless in deciding whether a theory is scientific. Both the Greek pantheon and geocentric theories are known to be false.
Falsifiability is simply the requirement that a scientific theory to list things that can’t happen under that theory. Falsifiability says scientific theory don’t look for evidence in support, they look for evidence to test the theory.
The fact that no false statements appear doesn’t mean that the scientific theory isn’t falsifiable. The fact that every statement of a theory has been true does not mean that the theory is falsifiable.
That statement does not itself constrain experience. That’s not a useful critique of the statement.
That doesn’t seem true. The statement seems to perfectly constrain experience: you will not experience situations where theories which do not constrain experience will still be falsified.
And indeed, watching the world go by over the years, I see theories like ‘Christianity’ or ‘psychoanalysis’ which do not constrain experience at all have yet to be falsified—exactly as predicted.
Fine, you want to be contrary. What experience would falsify the partial definition of scientific theory that I have labelled “the principle of falsifiability”? If no such experience exists, does this call into doubt the usefulness of the principle?
What experience would falsify the partial definition of scientific theory that I have labelled “the principle of falsifiability”?
Are you even trying here? Here’s what would falsify falsifiability: observing superior predictions being made by unfalsifiable theories, theories which have no reason to work but which do. Imagine a Christianity which came with texts loaded with prophetic symbolism which could be interpreted any way and is unfalsifiable, but which nevertheless keep turning out literally true (writes my hypothetical self, as he is tormented by Satanic wasps with the faces of humans prior to the sea turning into blood or something like that). In such a universe, falsifiability would be pretty useless.
Isn’t that essentially the best case for things like Nostradamus? Even assuming that the prophecies are accurate, they aren’t useful because they are so vague. The moment that the predictions are specific enough to be useful, they could be falsified.
What use is it to call that science? How could it possibly produce superior predictions in a world in which science works at all?
I don’t understand where we disagree, so let me clarify my position: A prophecy that is so vague that it can’t be disproved is so vague that it doesn’t tell you what will happen ahead of time. Calling that a prediction abuses the term to the point of incoherency.
Yes, that’s almost entirely a definitional point. Definitions aren’t necessarily empirical statements. They are either useful or not useful in thinking carefully. Thus, the fact that they cannot be falsified is not a relevant thing to say, in the same way that it isn’t useful to object that the Pythagorean theory can’t be falsified.
Know falseness is not really same thing as falsifiability. Known falseness is useless in deciding whether a theory is scientific. Both the Greek pantheon and geocentric theories are known to be false.
Falsifiability is simply the requirement that a scientific theory to list things that can’t happen under that theory. Falsifiability says scientific theory don’t look for evidence in support, they look for evidence to test the theory.
The fact that no false statements appear doesn’t mean that the scientific theory isn’t falsifiable. The fact that every statement of a theory has been true does not mean that the theory is falsifiable.
Nothing in this reply contradicts anything I have asserted. I was merely claiming that if falsifiability is a sufficient condition for a hypothesis to be “scientific”, then all theories known to be false are scientific (because if we know they are false, then they must be falsifiable). I’m not being contrarian; I’m pointing out a deductive consequence of the very definition of falsifiability that you linked to. Hopefully this closes the inferential distance:
If a hypothesis is falsifiable, then it is scientific.
If a hypothesis is known to be false, then it is falsifiable.
Therefore, if a hypothesis is known to be false then it is scientific.
I am merely denying the first premise via reductio ad absurdum, because the conclusion is obviously false (and the second premise isn’t). If you took my claim to be something other than this, then you have simply misread me.
That’s much clearer. I didn’t intend to assert that falsifiability was a sufficient condition for a theory being scientific, only that it is a necessary condition. That’s what I mean by saying it was a partial definition.
Thus, I don’t intend to assert the first sentence of your syllogism. Instead, I would say, “If a hypothesis is not falsifiable, then it is not scientific.” Adding the second statement yields: “If a hypothesis is know to be false, then it might be scientific.” That’s a true statement, but I don’t claim it is very insightful.
I understand that having beliefs that are falsifiable in principle and make predictions about experience is incredibly important. But I have always wondered if my belief in falsifiability was itself falsifiable. In any possible universe I can imagine it seems that holding the principle of falsifiability for our beliefs would be a good idea. I can’t imagine a universe or an experience that would make me give this up.
How can I believe in the principle of falsifiability that is itself unfalsifiable?! I feel as though something has gone wrong in my thinking but I can’t tell what. Please help!
Excellent question!
Excellent, because it illustrates the problem with “believing in” the principle of falsifiability, as opposed to using it and understanding how it relates to the rest of my thinking.
Forget that the principle of falsifiability is itself incredibly important. What sorts of beliefs does the principle of falsifiability tell me to increase my confidence in? To decrease my confidence in?
What would the world have to be like for the former beliefs to be in general less likely than the latter?
Thanks for the reply Dave. Are you saying I should not look at falsifiability as a belief, but rather a tool of some sort? That distinction sounds interesting but is not 100% clear to me. Perhaps someone should do a larger post about why the principle should not be applied to itself.
I have also thought of putting the problem this way: Eliezer states that the only ideas worth having are the ones we would be willing to give up. Is he willing to give up that idea? I don’t think so..., and I would be really interested to know why he doesn’t believe this to be a contradiction.
What I’m saying is that the important thing is what I can do with my beliefs. If the “principle of falsifiability” does some valuable thing X, then in worlds where the PoF doesn’t do X, I should be willing to discard it. If the PoF doesn’t do any valuable thing X, then I should be willing to discard it in this world.
It seems we have empirical and non-empirical beliefs that can both be rational, but what we mean by “rational” has a different sense in each case. We call empirical beliefs “rational” when we have good evidence for them, we call non-empirical beliefs like the PoF “rational” when we find that they have a high utility value, meaning there is a lot we can do with the principle (it excludes maps that can’t conform to any territory).
To answer my original question, it seems a consequence of this is that the PoF doesn’t apply to itself, as it is a principle that is meant for empirical beliefs only. Because the PoF is a different kind of belief from an empirical belief, it need not be falsifiable, only more useful than our current alternatives. What do you think about that?
I think it depends on what the PoF actually is.
If it can be restated as “I will on average be more effective at achieving my goals if I only adopting falsifiable beliefs,” for example, then it is equivalent to an empirical belief (and is, incidentally, falsifiable).
If it can be restated as “I should only adopt falsifiable beliefs, whether doing so gets me anything I want or not” then there exists no empirical belief to which it is equivalent (and is, incidentally, worth discarding).
You have just refuted the contention that all warranted beliefs must be falsifiable in principle. Karl Popper, who introduced the falsifiability criterion and pushed it as far if not further than it can go, never advocated that all beliefs should be falsifiable. Rather, he used falsifiability as the criterion of demarcation between science and non-science, while denying that all beliefs should be scientific. His contention that falsifiability demarcates science does imply, as he recognized, that the criterion of falsifiability is not itself a scientific hypothesis.
Rational beliefs are not necessarily scientific beliefs. Mathematics is rational without being falsifiable. The same is true of philosophical beliefs, such as the belief that scientific beliefs are falsifiable. But rational beliefs that are not scientific must be refutable, and falsifiable beliefs are a proper subset of refutable beliefs. Falsifiable beliefs are refutable in one particular way: they are refutable by observation statements, which I think are equivalent to EY’s anticipations. Science is special because it is 1) empirical (unlike mathematics) and 2) has an unusual capacity to grow human knowledge systematically (unlike philosophy). But that does not imply that we can make do with scientific beliefs exclusively, one reason being the one that you mention about criteria for the acceptance of scientific theories.
The broader criterion of refutability doesn’t necessarily involve refutation by observation statements. How would you refute the falsifiability criterion? It would be false if science it were the case that scientists secured the advance of science by using some other criteria (such as verification).
It’s a mistake to conflate the questions of whether a theory is scientific and whether it’s corroborated (by attempted falsifications). Or to conflate whether it’s scientific or it’s rationally believable. Theories aren’t bad because they aren’t science. They’re bad because they’re set up so they resist any form of refutation. Rational thought involves making your thinking vulnerable to potential refutation, rather than protecting it from any refutation.In science, the mode of refutation is observation, direct connection to sensory data. But it won’t do (as you’ve realized by trying to apply falsifiability to itself) to limit one’s thinking entirely to that which is falsifiable.
You later ask (in effect) whether the refutability criterion is itself even refutable. Would EY be willing, ever, to give it up? He should be, were someone to show that sheer dogmatism conduces to the growth of knowledge. That I can’t conceive of a plausible argument to that end doesn’t obviate the refutability of the contention
I think that resolves your confusion, but I don’t want to imply that Popper uttered the last word—there are problems with neglecting verification in favor of strict falsificationism.
Thank you for your thoughts.
What are the criteria that we use for accepting or refuting rational non-empirical beliefs? You mention that falsifiability would be refuted if some other criteria “secured the advance of science.” You also mention that we should give up the refutability criterion if “sheer dogmatism conduces to the growth of knowledge.” It sounds like our criteria for the refutability of non-empirical beliefs are mostly practical; we accept the epistemic assumptions that make things “work best.” Is there more to it than this?
To be pedantic and Popperian, I’d have to correct your use of “empirical beliefs.” The philosophical positions at issue aren’t scientific but they are empirical. “Empirical”—to be the basis for scientific observation statements—must be expressible in low-level observation sentences that all competent scientists agree on.
The belief in question is that science’s crucial distinguishing feature allowing it to advance is the subjection of science’s claims to empirical testing, allowing strict falsification. We can’t run an experiment or otherwise record observation statements, so we resort to philosophical debate aimed at refutation. Refutation is obtained by plausible argument. For instance, in the discussion about demarcation, an example of a potentially plausible argument goes if we relied on falsification exclusively, we would never have evidence that a claim is true, only that it isn’t false. But we rely on scientific theories and consider them close to the truth (or at least as probably so). Therefore, falsifiability can’t explain the distinctiveness of science.
This involves highly plausible claims, based on observation, about how we in fact use scientific theories. But although the result of observation, it can’t be reduced to something everyone agrees on that is closely tied to direct perception, as with an observation statement.
For me the principle of falsifiability is best understood as a way of distinguishing scientific theories about the world from other theories about the world. In other words, falsifiability is one way of defining what science is and is not. A theory that does not constrain experience (“God works in mysterious ways”) is not a scientific theory because it can explain any occurrence and is therefore not falsifiable.
Because falsifiability is a definition, not a theory about the world, there’s no reason to think it can be falsified. The definition could be wrong by failing to accurately or usefully define scientific theory, but that’s conceptually different.
Falsifiability is a very bad way to define science (or scientific theories). If falsifiability was all it took for a theory to be scientific, then all theories known to be false would be scientific (after all, if something is known to be false, it must be falsifiable). Do we really want a definition of science that says astrology is science because it’s false?
*shrug*
I don’t think the current line of enquiry is particularly useful.
“Astrology works” is a scientific theory to the degree that it is, in fact, acceptable science to do an experiment to see whether or not astrology has predictive power. It’s rhetorically inaccurate to say that means “astrology is science” though, because of course the practice of astrology is not. But sure, it’s probably a good idea to include other conditions. Excessively unlikely (or non-reductionist?) hypotheses could be classified as non-scientific, for the simple reason that even considering them in the first place would be a case of privileging the hypothesis.
None of this contradicts falsifiability being “a way of distinguishing scientific theories about the world from other theories about the world”, if we have other ways of distinguishing scientific from non-scientific, such as “reductionism”.
Astrology does seem to consist of scientific hypotheses.
I chose astrology because it has a reverse halo effect around here (and so would serve me rhetorically). Feel free to replace it with any other known to be false set of propositions.
I agree that falsifiability is not a complete definition. My point was only that falsifiability is not applicable to the principle of falsifiability, any more than it applies to mathematics.
That said, Newton’s physics and geocentric theories are false. Are they not science simply for that reason?
Yes. Falsifiability is a poor definition of science and is self-undermining in the sense that it can’t pass its own test.
Of course not. I’m not claiming a scientific theory must be true. I’m claiming that known falseness (which implies falsifiability) is not a sufficient condition for being scientific.
That statement does not itself constrain experience. That’s not a useful critique of the statement.
Know falseness is not really same thing as falsifiability. Known falseness is useless in deciding whether a theory is scientific. Both the Greek pantheon and geocentric theories are known to be false.
Falsifiability is simply the requirement that a scientific theory to list things that can’t happen under that theory. Falsifiability says scientific theory don’t look for evidence in support, they look for evidence to test the theory.
The fact that no false statements appear doesn’t mean that the scientific theory isn’t falsifiable. The fact that every statement of a theory has been true does not mean that the theory is falsifiable.
That doesn’t seem true. The statement seems to perfectly constrain experience: you will not experience situations where theories which do not constrain experience will still be falsified.
And indeed, watching the world go by over the years, I see theories like ‘Christianity’ or ‘psychoanalysis’ which do not constrain experience at all have yet to be falsified—exactly as predicted.
Fine, you want to be contrary. What experience would falsify the partial definition of scientific theory that I have labelled “the principle of falsifiability”? If no such experience exists, does this call into doubt the usefulness of the principle?
Are you even trying here? Here’s what would falsify falsifiability: observing superior predictions being made by unfalsifiable theories, theories which have no reason to work but which do. Imagine a Christianity which came with texts loaded with prophetic symbolism which could be interpreted any way and is unfalsifiable, but which nevertheless keep turning out literally true (writes my hypothetical self, as he is tormented by Satanic wasps with the faces of humans prior to the sea turning into blood or something like that). In such a universe, falsifiability would be pretty useless.
Isn’t that essentially the best case for things like Nostradamus? Even assuming that the prophecies are accurate, they aren’t useful because they are so vague. The moment that the predictions are specific enough to be useful, they could be falsified.
What use is it to call that science? How could it possibly produce superior predictions in a world in which science works at all?
Yes, that is rather the question you should be answering if you want to criticize the desirability of falsifiability as being unfalsifiable itself...
I don’t understand where we disagree, so let me clarify my position: A prophecy that is so vague that it can’t be disproved is so vague that it doesn’t tell you what will happen ahead of time. Calling that a prediction abuses the term to the point of incoherency.
Yes, that’s almost entirely a definitional point. Definitions aren’t necessarily empirical statements. They are either useful or not useful in thinking carefully. Thus, the fact that they cannot be falsified is not a relevant thing to say, in the same way that it isn’t useful to object that the Pythagorean theory can’t be falsified.
If you intend to invoke some other critique of Popper and his use of falsifiability to distinguish science from non-science, please by more explicit, because I don’t understand your argument.
Nothing in this reply contradicts anything I have asserted. I was merely claiming that if falsifiability is a sufficient condition for a hypothesis to be “scientific”, then all theories known to be false are scientific (because if we know they are false, then they must be falsifiable). I’m not being contrarian; I’m pointing out a deductive consequence of the very definition of falsifiability that you linked to. Hopefully this closes the inferential distance:
If a hypothesis is falsifiable, then it is scientific.
If a hypothesis is known to be false, then it is falsifiable.
Therefore, if a hypothesis is known to be false then it is scientific.
I am merely denying the first premise via reductio ad absurdum, because the conclusion is obviously false (and the second premise isn’t). If you took my claim to be something other than this, then you have simply misread me.
That’s much clearer. I didn’t intend to assert that falsifiability was a sufficient condition for a theory being scientific, only that it is a necessary condition. That’s what I mean by saying it was a partial definition.
Thus, I don’t intend to assert the first sentence of your syllogism. Instead, I would say, “If a hypothesis is not falsifiable, then it is not scientific.” Adding the second statement yields: “If a hypothesis is know to be false, then it might be scientific.” That’s a true statement, but I don’t claim it is very insightful.