Any Good Criticism of Karl Popper's Epistemology?

I believe there are no correct, unaddressed criticisms of Karl Popper’s epistemology (Critical Rationalism – CR). If I'm mistaken, I'd like to be told. If others are mistaken, I'd like them to find out and take an interest in CR.

CR is important for the general reasons that epistemology is important (it tells you about how to think, how to learn, how to evaluate ideas and arguments, and how AGI could work). It also refutes induction, which is a popular belief here. If CR is correct, then people here have a lot to change their mind about.

I've found CR criticism falls into some broad categories, with some overlap:

  1. The people who heard Popper is wrong secondhand but didn’t read much Popper and have no idea what CR is actually about. They often try to rely on secondary sources to tell them what CR says, but most secondary sources on CR are bad.
  2. The pro-induction people who don’t engage with Popper’s ideas, just try to defend induction. They don’t understand Popper’s criticism of induction and focus on their own positive case for induction. They also commonly admit that some criticisms of induction are correct, but still won’t change their minds or start learning the solution to induction’s flaws (CR).
  3. The falsificationism straw man, which misinterprets Popper as advocating a simplistic, false view. (There are some other standard myths too, e.g. that Popper was a positivist.)
  4. Critics of The Logic of Scientific Discovery who ignore Popper’s later works and don’t engage with CR's best ideas.
  5. Critics with points which Popper answered while he was still alive. Most criticisms of Popper are already answered in his books, and if not there then in this collection of Popper criticism and Popper’s replies. (I linked volume two which has Popper’s replies, you will want volume 1 also.)

If you believe Popper is wrong, then: Do you believe you personally understand CR? And have you looked at Popper’s books and replies to his critics to see if your point is already answered? If so, have you written down why Popper is mistaken? If not, do you believe someone else has done all this? (They understand CR, are familiar with Popper’s books including his replies to his critics, and wrote down why Popper is mistaken.)

Whether it’s by you or someone else, you can reply with a reference to where this is publicly written down in English. I will answer it (or refer you to an answer or get a colleague to answer). Here is what I expect in return: if your reference is mistaken, you will study CR. You were wrong about CR’s falsity, so it’s time to learn it. If you would be unwilling to learn CR even if you agree that your referenced criticism of CR is false, then you shouldn’t have an opinion on CR. If you still wouldn’t want to learn CR even if all your objections were wrong, then you either aren’t participating in the field (epistemology) or shouldn’t be. (I have nothing against lay people as long as they are interested in learning and thinking. I do have something against people, whether lay or philosophy professors, who state their opinion that Popper is wrong but would not be willing to learn about Popper even if they found out their negative beliefs about Popper are false.)

If you believe one of the many criticisms of Popper is correct, but you don’t know which one and don’t want to pick one, then you are not treating the matter rationally. It’s unacceptable if your plan is, on having one criticism answered, to simply pick another one, and repeat indefinitely. You’re welcome to have one good reference which makes multiple important points, but you don’t get to just keep referencing different critical authors repetitively (as each one fails, you pick another) while not reconsidering your own beliefs. You need to stick your own neck out – as I do. If I can’t answer a challenge to CR I will reconsider my views.

If you want to bring up a couple criticisms at the start, which are written in different places, but you won't add any more later, then that could be reasonable – but provide a brief explanation of why it's needed. In this case where you want to bring up multiple points by different authors, I'd expect you to be referencing specific sections or short works, not multiple whole books. E.g. you could reasonably say you have 3 criticisms of Popper, chapter 3 of book X, chapter 7 of book Y, and paper Z.

Alternatively, if Popper is mistaken but no one has actually written correct criticism (including you), then how do you know he's mistaken? Maybe he's not!

Note: I'm interested in criticisms like "Popper's idea X is false b/c Y.", not like "I wasn't convinced by Popper's writing on topic X." (The second one is compatible with Popper being correct, and is too vague to answer.)


See also:

Yudkowsky on Popper

Criticism of Eliezer Yudkowsky on Karl Popper

CR Reading Recommendations


sorted by
magical algorithm
Highlighting new comments since Today at 3:20 AM
Select new highlight date
All comments loaded

Can you suggest a good statement by Popper of his final position? He seemed to be a) forever railing that people misunderstood him b) frequently changing his positions. So it is hard to know what to criticize.

And given he seems to be in low repute among professional philosophers (that I have spoken to) it would be good to hear the elevator pitch as to why understanding his ideas in detail is a good use of our time.

You could start with the criticisms in the Stanford Encyclopedia of Philosophy entry.


a) Refutation requires acceptance of the theory underlying the experiment that refuted the theory. E.g. Think of all the theory that needs to be accepted to agree that the Michelson-Morley experiment refuted absolute space-time. So if you cannot prove any theory refutation also fails.

b) All critical tests are actually repairable and are not fatal even if failed. Also all tests are probabilistic - you never have a 100% refutation. The way anomalies are actually dealt with is way beyond the scope of Popper's theory.

c) The importance of auxiliary hypotheses in any test. It is argued that Popper, in dealing with this issue, greatly reduced the scope of his theory, and made it far weaker and more subjective.

Public announcement: the author already posted quite a lot about the same topic on Less Wrong:

The posts from April 2011 are all in negative karma, except for the link to the David Deutsch video (which is a link to a lecture, without a description). The posts from October/November 2017 are currently at 0 or 1 karma points, because downvoting is disabled.

I think that reading the information above already tells a lot.

I am posting this here, because I have myself not noticed the April 2011 articles, so I was under a mistaken impression that this was a new user, completely unfamiliar with local common knowledge. I engaged in a debate... and it had a completely predictable trajectory, that could have been obvious from reading similar debates under the older articles.

That means, during the previous 6 years, this interaction provided zero benefit to both sides. I think we do not need another dozen articles of the same kind.

Some more context: The author is selling a $400 course about (his interpretation of) Popper.

> This doesn't solve the problem of figuring out which ideas are good, it just gives an arbitrary answer (shorter doesn't mean truer).

It taken literally this suggests you don't understand SI. Which says that shorter explanations are preferred all else being equal. A short hypothesis that strongly contradicts the data is not preferred to something consistent with the data in various ways.

I have ordered his response to critics and will read it. But I find the contrast bettwen his claimed disproof of induction and the no free lunch theorems reflects quite poorly on Popper.

What the NFL theorems say, roughly, is that to learn from inducation you need a prior on hypothesis. With that, you can learn. So example, a bias towards simplicity is a prior, as are spatial and temporal locality. No-one knows why these priors work but they do so far. Popper seems to say (like many philosphers he maintains plausible deniability) is that it us just am anazing fluke that it has worked so far.

What does Popper really offer? A room full of philosphers to work it out? Vague formulations like that we "prefer" certain hypothesis?

I don't see anything of value so far but I will read his response.

On the other hand I wish more people woudl listen to him on the open society and the value of free discourse.

CR offers a general pupose epistemology. Epistemology is the most important field (because thinking methods are used by every other field), and CR has the only known general purpose epistemology that isn't known to be wrong.

You asked for an elevator pitch, I provided one, and you then wrote "I don't see anything of value so far" while not engaging with it (you responded to some addenda by guessing I'm grossly ignorant for some reason which is unclear to me). And yes of course SI rejects empirically refuted ideas first, so what? There are still infinitely many ideas left over after that.

I have ordered his response to critics and will read it.

I hope you will also write which responses you consider mistaken, and why, clearly, with quotes and details. Someone should, out of the many people who disagree with Popper and claim to be thinkers, don't you think?

There are many people who are wrong and there's no reason to write low texts about how everybody of them is wrong.

Being a thinker doesn't mean that you have to argue about how everybody is wrong. If you think otherwise, how about writing a treatize about how Hegel was wrong (of course you actually have to read him first)?

There are of course pre-existing criticisms of Hegel, e.g. by Popper in OSE. People have written that.

Popper didn't change views significantly but LScD is harder to understand, and philosophy is more than hard enough to understand in general. Popper is in low repute because he disagreed with people, and advocated some things (e.g. that induction is imposisble) that they consider ridiculous (which isn't an answer to him). Plus most people go by secondary sources. My colleague surveyed over 100 textbooks and found none of them accurately represented Popper's views – they're broadly similar to the SEP.

(a) Simply ignores Popper's appraoch to fallibilism and conjectural knowledge. It's saying CR doesn't work given infallibilist premises that Popper disputes. (Note the demand for proof.) CR accepts fallibility (which has very compelling logical arguments) and then takes it seriously by e.g. developing a fallibilist theory of knowledge rather than demanding certainty of refutation (which is impossible).

(b) Yes you can repair criticism by modifying an idea or by criticizing the criticism (which is essentially modifying the idea by adding a footnote to address the criticism, which adds content that wasn't there previously). How is that a criticism? Also I don't know why you think fallibilism = probability. Uncertainty frequently isn't numeric. "We may be mistaken in some way we haven't thought of" isn't a probability, the future growth of knowledge is *unpredictable*.

(c) Again there's no understanding of Popper's views here. The thing you're complaining about is somethign Popper emphasized, explained, and addressed. And you don't say what about Popper's position on the matter is weak or subjective (neither of which are part of Popper's own account, and "weak" sounds suspiciously like "fallibile", while "subjective" sounds suspiciously contrary to Popper's theory of Objective Knowledge, which FYI is one of his book titles. I didn't find the weakness or subjectivism when I read Popper, and you haven't told me where to look with any specificity.)


Elevator pitch:

CR solves the fundamental problems of epistemology, like how knowledge can be created, which induction failed to solve. It's a very hard problem: the only solution ever devised is evolution (literally, not analogously – evolution is about replicators, not just genes). In terms of ideas, evolution takes the form of guesses and criticism. CR develops much better criticisms of induction than came before, which are decisive. CR challenges the conventional, infallibilist conception of knowledge – justified, true belief – and replaces it with a non-skeptical, non-authoritarian conception of knowledge: problem-solving information (information adapted to a purpose). Although we expect to learn better ideas in the future, that doesn't prevent our knoweldge from having value and solving problems in the current context. This epistemology is fully general purpose – it works with e.g. moral philosophy, aesthetics and explanations, not just science/observation/prediction. The underlying reason CR works to create knowledge is the same reason evolution works – it's a process of error correction. Rather than trying to positively justify ideas, we must accept they are tentative guesses and work to correct errors to improve them.

This position should not be judged by how nice or strong it sounds; it logically works OK unlike every rival. Decisive issues for why something can't work at all, like induction faces, have priority over how intuitive you find something or whether it does everything you'd like it to do (for example, CR is difficult to translate into computer code or math, which you may not like, but that doesn't matter if no rival epistemology works at all).

I expect someone to bring up Solomonoff Induction so I'll speak briefly to that. It attempts to answer the "infinite general patterns fit the data set" problem of induction (in other words, which idea should you induce from the many contradictory possibilities?) problem with a form of Occam's Razor: favor the ideas with shorter computer code in some language. This doesn't solve the problem of figuring out which ideas are good, it just gives an arbitrary answer (shorter doesn't mean truer). Shorter ideas are often worse because you can get shortness by omitting explanation, reasoning, background knowledge, answers to critics, generality that isn't necessary to the current issue, etc. This approach also, as with induction in general, ignores critical argument. And it's focused on prediction and doesn't address explanation. And, perhaps worst of all: how do you know Occam's Razor is any good? With epistemology we're trying to start at the beginning and address the foundations of thinking, so you can't just assume common sense intuitions in our culture. If we learn by induction, then we have to learn and argue for Occam's Razor itself by induction. But inductivists never argue with me by induction, they always write standard English explanatory arguments on philosophical topics like induction. So they need some prior epistemology to govern the use of the arguments for their epistemology, and then need to very carefully analyze what the prior epistemology is and how much of the work it's doing. (Perhaps the prior epistemology is CR and is doing 100% of the work? Or perhaps not, but that needs to be specified instead of ignored.) CR, by contrast, is an epistemology suitable for discussing epistemology, and doesn't need something else to get off the ground.

(If you'd like more detail, see the reading recommendations linked at the bottom of my post.)

You've done a whole lot of telling us how amazing this stuff is, but not much telling what it actually is. So I'm going to guess, in the hopes that you can tell me not just that I'm wrong, but specifically what a better version would be.

According to what you've said, it seems the process of people having valuable knowledge of future events (e.g., the sun will rise tomorrow), is that people generate guesses (by some unspecified process that's definitely not induction), and then over time, other people criticize guesses, and the guesses best able to stand up to criticism are what we should use to predict tomorrow's sunrise.

And the reason this works, according to you, is that it's like evolution. Just like how in evolution, mutation and selection leads to creatures that take advantage of patterns in their environment to get a fitness advantage, guessing and criticism leads to ideas that take advantage of some sort of pattern in the environment in order to be more valuable.

But, of course, taking advantage of patterns in the environment in order to make valuable predictions about the future totally isn't induction, and there's no way you could formalize it other than the precise way Popper chose to formalize it.

how do you know Occam's Razor is any good?

Imo chapter 28 of this book gives a good sense why Occam's Razor is good. I'll try to explain it here briefly as I understand it.

Suppose we have a class of simple models, with three free binary parameters, and a class of more complex models, with ten free binary parameters. We also have some data, and we want to know which model we should choose to explain the data. A priori, out of the parameter sets for the simple model each has a probability of 1/8 of being the best one, whereas for the complex model the probability is only 1/1024. As we observe the data, probability mass moves between the parameter sets. Given equally good fit between data and model, the best simple model will always have a higher probability than the best complex model. For one, because it started with a higher probability. For another, because there will be several complex models fitting the data about equally well. E.g. there may be 8 complex models, which all fit the data better than the second-best simple model. So the probability mass needs to be shared between all of those.

A complex model needs to fit the data better in order to gain enough probability mass to beat out the simpler model.

So even if we do not penalize complex models just for being more complex, we still favour simpler ones.

None of this is relevant to specifying the prior epistemology you are using to make this argument, plus you begin with "simple models" but don't address evaluating explanations/arguments/criticisms.

Given some data and multiple competing hypotheses that explain the data equally well, the laws of probability tell us that the simplest hypothesis is the likeliest. We call this principle of preferring simpler hypotheses Occam's Razor. Moreover, using this principle works well in practice. For example in machine learning, a simpler model will often generalize better. Therefore I know that Occam's Razor is "any good". Occam's Razor is a tool that can be used for problems as described by the italicized text above. It makes no claims regarding arguments or criticisms.

I don't really see why I would need a coherent/perfect/complete epistemology to make this kind of argument or come to that conclusion. It seems to me, like you are saying, that any claims that aren't attained via the One True Epistemology are useless/invalid/wrong. That you wouldn't even accept someone saying that the sky is blue, if that person didn't first show you that they are using the right epistemology.

I notice that I don't know what an argument that you would accept could even look like. You're a big fan of having discussions written down in public. Could you link to an example where you argued for one position and then changed your mind because of somebody else's argument(s)?

What is the relationship between CR and other processes that can create knowledge, such as induction and deduction? Are the latter a subset of the former? What does 'induction is impossible' mean, that it cannot be used as a starting point or something stronger? Can CR be not only a starting point but also the only process necessary?

Why is this so downvoted. This is good, we need more of this. It's exercises like this that push Epistemology forward. Upvoted.

I don't currently have the time to participate in this exercise, but I would follow this thread.

This is good, we need more of this.

Ignoring the value judgement part, we already have "more of this" on Less Wrong.

It's not the first thread by the person about the same topic. Here we have an outsider coming into the our community and advertising their own ideology and calling for us to engage that ideology on the terms that the outsider sets.

I have asked LW to specify terms (preferably pre-written not ad hoc) – an alternative to Paths Forward – and no one has.