Introduction to Critical Rationalism

Summary: I explain Critical Rationalism’s fallibilist, evolutionary epistemology, including criticism of induction and justificationism. These are ideas that Critical Fallibilism (CF) agrees with and builds on.

If you'd prefer to watch a summary video before reading this article, click here.

Introduction

Critical Rationalism (CR) was developed by Karl Popper (1902-1994). It’s a fallibilist philosophy which says our knowledge is always tentative (open to reconsideration and revision). Fallibility says we can’t have a guarantee against error; no matter what we do, any idea could contain an error that we missed. Perfection is out of reach. And, actually, mistakes are common. What we can have instead of perfection is progress. We can keep making improvements and never hit the end of the road (perfection). There’s no limit to how much better our knowledge can get.

CR refutes induction and justificationism (which is the use of positive arguments to justify or support ideas). Philosophers have long known about difficulties with those approaches, but have tried (unsuccessfully) to fix the problems. CR says those approaches can’t work, gives up on fixing them, and proposes alternatives. Previously, a surprisingly low amount of effort went into seeking alternatives.

CR’s approach is an evolutionary epistemology that says we learn by error correction – brainstorming and criticism, conjectures and refutations, trial and error, critical discussion. CR offers insight into the scientific method, the history of science, the role of observation, and rationality.

Popper worked on other ideas too, including political ideas that I have some disagreements with. My favorite Popper work, besides his epistemology, is his writing about Presocratic philosophers in The World of Parmenides.

Evolution

Evolution is the only known process that creates new knowledge. It works by replication with variation and selection. Genes and ideas (“memes”) are both replicators. Intelligence works by literally evolving ideas in your subconscious mind. Intelligence works on the same principle – evolutionary replication with variation and selection – as the development of the animal and plant species. This isn't an analogy or metaphor.

Biological evolution adapts genes to a purpose. Evolution requires selection criteria. The criterion in nature is, roughly, having grandchildren. Evolution selects for genes which result in survival, having offspring, and those offspring surviving to have their own offspring. This is often summarized as “survival of the fittest”. The genes that build the most effective animals will, on average, make more copies of themselves and play a larger role in the gene pool of the next generation.

Intellectual evolution adapts ideas to a purpose. But humans can choose their own purposes. We aren’t born with a single, fixed purpose. On one day I can adapt ideas to one purpose, and on another day I can change my mind and adapt them in a different way. I can use different selection criteria at different times or for different topics.

Knowledge is purposeful information. It’s information that is adapted to a purpose (a goal, an objective). It’s information that seems designed instead of random or arbitrary. Design means or implies designed for some purpose. A designer tries to achieve something.

When we see the appearance of design, that indicates knowledge is present. Design comes from evolution – either evolution in nature or from a designer that uses an evolutionary thought process.

Richard Dawkins’ book,The Selfish Gene, can help with understanding evolution. It contains more modern ideas about evolution than Popper's work. The connections between modern ideas about evolution and CR were also expanded on by David Deutsch. For more details, see my articles Knowledge, Knowledge Creation, Evolution and Knowledge and Evolution.

Critical Discussion

CR says to look for flaws in ideas. We can discuss what we think is wrong with ideas and try to improve our ideas. But we shouldn’t give arguments to (positively) support our ideas. Our intellectual tool is (negative) criticism, not justification. There’s no way to prove an idea is not wrong. No matter how much you investigate and ponder, there could be an error that you didn't find.

Criticism works from a logical perspective because you can make an argument that (logically) contradicts something. When you do that, it means that if your criticism is correct (including some background knowledge and underlying premises), then the idea you’re criticizing must be incorrect.

There’s a logical asymmetry between supporting arguments and criticism. A single counter-example can refute an idea, while many compatible pieces of evidence cannot prove an idea true. If you claim “All ravens are black”, a single orange raven is a decisive criticism, but a thousand black ravens aren't decisive support. A thousand black ravens aren't even a good hint; seeing them does not mean it’s probably true that all ravens are black.

This approach to critical arguments doesn’t work with supporting arguments. In general, it’s impossible to make supporting arguments such that if your argument is correct, then the idea you’re supporting must be true. You can't completely cover all possibilities and get a guarantee against failure, error or fallibility. You could always have missed something. There could be some problem you didn’t consider and didn’t address. Positive arguments that rule out 500 potential errors still can’t completely prove an idea is true, even if you’re correct about ruling all of those 500 issues out.

CR says the point of discussion is to find and fix errors. In other words, we should use negative arguments not positive arguments.

CR says we learn by conjectures and refutations. This means we create ideas and look for errors to eliminate. You can think of it as learning by brainstorming plus critical thinking. This process follows evolution. Brainstorming replicates and varies ideas while criticism does selection.

CR says to judge ideas, and choose between them, by how well they survive criticism. We can look at how vigorously they were criticized and put to the test, and how well they handled those attempts at refutation. This is called forming critical preferences: preferences about ideas based on criticizing them.

Critical preferences is one of the main areas where Critical Fallibilism differs from CR. CF says to judge ideas by whether or not they're decisively refuted. Decisive criticisms indicate ideas will fail at goals. When pursuing a goal, we shouldn't use decisively refuted ideas – ideas that as best we know definitely won't work; instead we should use non-refuted ideas that might work. Considering whether criticisms are decisive for our goals or not, rather than looking at the degree to which ideas survive criticism, is one of my attempts to improve on CR.

CR doesn’t provide a lot of specifics about how to have a rational, critical discussion or how to do that kind of thinking alone (it’s similar with or without other people). CR talks about abstract principles, which is great, but there isn’t enough guidance about what exactly to do. That's something I’ve aimed to improve with CF by creating more specific methodology for discussion, debate and analysis.

Justificationism

Justificationism is the (false) idea that positive arguments can be used to say how good an idea is. It says we can build up ideas with supporting arguments, supporting evidence, proofs, the pros on a pro/con list, etc. Even if no individual argument is a full proof, many positive arguments together can (allegedly) add up to a strong case. In short, we should judge how much goodness our ideas have. Then we should believe or act on the ideas with the highest quantities of goodness. Some justificationists use math and numbers for this, and others don’t, but the underlying concept of justification is the same whether or not its numerical.

A common (false) view among philosophers is that knowledge is “justified, true belief” (JTB). CR explains that all three words in JTB are wrong. I've already criticized the "justified" part of JTB. The “true” part is infallibilist and perfectionist: it says that anything that turns out to contain any error was never knowledge at all. But then how and why are our ideas useful and effective when basically none of them are actual knowledge just because they aren't absolutely perfect? The “belief” part is problematic because it denies that books or genes contain knowledge (since books and genes aren't conscious or intelligent agents and don’t have beliefs).

JTB advocates currently object, saying that they already know JTB is imperfect and have been discussing its problems for many years, so it’s unfair to criticize JTB. But what do they think is wrong with it? They think JTB isn’t restrictive enough and needs a fourth letter to exclude more bad ideas from being knowledge. They think all three letters are good points that are merely incomplete. So they haven't already accepted and understood CR’s criticisms.

While JTB involves an infallibilist attitude, other justificationists are fallibilists and don’t demand truth as a required component of knowledge. They are satisfied with strong arguments and evidence, which they believe make ideas probable or high credence, or give us high (justified) confidence in ideas.

A typical phrase like “arguments and evidence for that” is problematic. CR explains that evidence doesn’t speak for itself. All evidence must be interpreted according to our theories and background knowledge. Raw data doesn’t have a single, inherent meaning. So we only learn from evidence via arguments and interpretations. We use evidence in arguments, and connect it with ideas, but we can’t use raw evidence alone with no ideas.

Many justificationists accept critical arguments in addition to positive arguments. They may say to first look at decisive critical arguments, like “that idea is refuted by the evidence” or “that idea has an internal logical contradiction”. That rules out some particularly bad ideas. Then, for the remaining ideas, they'd say to look at both the positive and negative arguments (pros and cons), and try to judge how good ideas are. CR, in short, says to only look at cons, not pros. (CF says that many pros can be rephrased as cons of rival ideas, so try doing that before rejecting them.)

The problem with positive or justifying arguments is that establishing ideas as good or true is impossible. No matter how many good traits an idea has, it could be wrong. And there are infinitely many other ideas, in the set of all logically possible ideas, that have all of those good traits but reach wildly different conclusions.

No number of confirming data points or supporting arguments can ever logically imply (or even hint) that an idea is correct, even if we accept those data points and arguments are correct. However, a single criticism that we accept can refute an idea. Why? Because a criticism contradicts an idea, and contradiction is a meaningful, useful, important relationship. The relationship of positive arguments to an idea is merely failure to contradict. No amount of ideas that don’t contradict you, don’t refute you or don’t criticize you ever constitute a good case that you’re right. The idea that positive arguments have some kind of closer relationship to the ideas they favor than non-contradiction is one of philosophy’s largest myths. Non-contradiction is all the justificationists have but non-contradiction is not positive support.

Naming, criticizing and rejecting justification was one of Popper's major achievements (along with, even more importantly, providing his evolutionary epistemology as a usable alternative). Although the way I explained this doesn't follow any particular Popper essay, and some of my wordings are different than his, I think this follows his concepts and intention well.

Fallibilism

Fallibilism is the idea that we're capable of making mistakes and any of our ideas may be mistaken. We can't get 100% certain proof of anything no matter how confident we feel or how much we double check our work. We could have missed something. Being fallible is the permanent human condition, but that's OK. We can still learn and make progress by looking for and correcting errors.

There are logical arguments regarding fallibilism, but they aren't the main point. CR views errors as common, everyday occurrences, not just technically possible.

Justificationism can be viewed as an attempt to fight against fallibility. The goal of justification is to prove ideas are true or reduce the possibility of ideas being false – in other words to make ideas infallible or less fallible. Popperians accept our fallibility and focus on error correction.

Although we can never be done thinking and know there are no more errors (nor that errors are very unlikely), even in a limited topic area, we can improve our knowledge and make ongoing progress. We can solve problems, learn new things, invent cars, smartphones and antibiotics, all without any guarantees against error. We just need to do our best using rational methods. Sometimes mistakes will be made, even big ones, but we can recover and keep going. That's what human history has been like as a whole, and it's what individuals lives are like too.

To best deal with our fallibility, we should be open to criticism, including from sources that don't have official credentials or authority. Ignoring critics and refusing to debate is one of the ways of preventing error correction. We should also strive to think critically and unbiasedly about our own ideas. And we should question ideas, even very popular ideas, to see if we can find any errors before accepting them.

Induction

Induction is an attempt to use evidence without arguments or intelligent thoughts. It has to be, because it’s supposed to be a theory that explains how humans are intelligent and learn. So it can’t presuppose or use intelligence or common sense, or it’d fail at its goal. It's fine for induction to use arithmetic and other more basic tools.

CR’s evolutionary epistemology attempts to explain how we can learn without presupposing any intelligence or common sense. Replication, variation and selection of ideas are prior to intelligence. Evolution explains how the knowledge in the animal species could have been created without an intelligent designer. Evolution is capable of explaining intelligence because it doesn’t require intelligence as a prerequisite. For induction to be a rival to CR, it'd have to offer an explanation of intelligence too, not use intelligence as a premise.

The basic concept of induction is to find patterns in data and learn from them. At a high level, this has major flaws. It isn’t trying to explain and understand the world conceptually. And it isn’t about error correction. It’s trying to support conclusions with data instead of trying to find and fix mistakes. Also, how do you find patterns and learn except by using your preexisting intelligence?

Induction has some more specific problems too. Logically, there are infinitely many patterns compatible with any finite data set. Which patterns should we induce? Induction doesn’t offer usable guidance for choosing between huge numbers of options. It assumes you only brainstorm a few candidate theories (using your intelligence, common sense and intuition to make them pretty good) and then induction pays selective attention to only those ideas.

Similarly, induction makes claims like “the future will probably resemble the past” or “patterns tend to hold over time”. This again shows selective attention to only some future events or patterns. No matter what happens, infinitely many patterns hold and infinitely many patterns are broken. No matter what happens, the future resembles the past in some ways but not others. xkcd provides some examples. So the key issue, which induction doesn’t address, is which patterns will hold over time and which won’t? In which ways should we expect the future to resemble (or not resemble) the past?

Induction basically means to find a pattern – a correlation – and then assume there is (or probably is) a causation. This is bad enough because correlation doesn’t imply causation. But it’s much worse than that because which pattern(s) should you find out of the infinitely many logical possibilities? For any conclusion you reach, there are many different patterns, which also fits all the data, which reach contradictory conclusions.

There’s also been inadequate attention given to precisely defining what counts as a “pattern” or what counts as “similar”. How exactly do you judge whether a future event is similar enough to “resemble” a past event?

Also, induction says evidence supports ideas. But any piece of evidence is logically compatible with (doesn’t contradict) infinitely many ideas and is incompatible with infinitely many ideas. The belief that a piece of evidence can support one idea more than it supports infinitely many others is a myth. All evidence can do is contradict or fail to contradict an idea. And then you’re left with infinitely many ideas which are consistent with all your evidence, and no meaningful guidance from induction about how to choose between them.

One of the main attempts to defend justification or induction involves probability. If one idea assigns an event a 90% probability, and another assigns it a 60% probability, and the event happens, it's said that the first idea is more supported and more likely to be true. Popper criticized some problems with that approach. Probability in epistemology is a big topic which I won't go into more detail on here.

Criticizing and rejecting induction, along with providing an alternative, was one of Popper's major accomplishments. I've thought a lot about induction and explained this in my own way which is different than Popper's presentation. I hope my presentation is helpful to modern audiences. I think I stayed pretty true to Popper's concepts and intentions regarding induction.

Interpreting Observation

Popper said that we have to interpret our sense data. Our interpretations are fallible. We can’t just look at the world, see raw data, and trust it. We may be confused about what we’re seeing or our sensory organs may make errors.

Because our observations are based on our interpreting ideas and background knowledge, we couldn't assume they’re true even if we could fully trust our sensory organs. They’re just as fallible as our other ideas.

There's a story that Popper started a lecture by saying "Observe." and then waited silently until the audience asked what to observe. This made his point that we need to have ideas first to guide us, rather than passively letting observations flow into us. We need a goal or purpose to help us know what to observe. We use our senses to search for relevant or useful data (relevance or usefulness is determined by our ideas), rather than just taking in everything as raw data (there's way too much data to pay attention to all of it). Noticing when our observations don't match our expectations is another good thing to look for.

Decisive Criticism

CR and CF take seriously that we only have contradiction and non-contradiction, not positive support or justification. Non-contradiction is also called consistency or compatibility. CR concludes that we must use negative, critical arguments (that make use of logical contradiction) and focus on error correction.

CF goes further and concludes that all ideas should be evaluated as either refuted or non-refuted. A refuted idea means we accept an argument that contradicts it. A non-refuted idea means we don’t know of any error, criticism, refutation or contradiction. CF says all correct criticism is decisive (contradiction). There are no weaker criticisms. There are no partial arguments. Arguments are binary, not a matter of degree. Strong and weak arguments are myths. (Popper believed that we have strong and weak criticisms, and that we could evaluate how good ideas were by how well – to what degree – they survived criticism. I believe that was a mistake in CR: Popper made major breakthroughs but didn’t perfectly figure everything out.)

Logical Concerns

If we believe two ideas contradict, could they both be true? Yes, because we could have made a mistake when we determined that they contradict, or we could have a misconception about logic. In general, our background knowledge or context could have an error which affects a conclusion. How does CR deal with this? By concerning itself primarily with actual errors that anyone has found, not potential errors. CR isn't about guarantees or perfection. If you point out an error in someone's conception of logic, or in their analysis, or in their relevant background knowledge, then that's an issue to be addressed. But if no one sees an error, then there's no known error there, and that's the best we can ever do with anything (we use ideas that we don't know errors in, not ideas that have a guarantee against error, which isn't possible to get).

Can any refuted idea be rescued by modifying it with ad hoc auxiliary hypotheses so that it no longer contradicts any data or no longer has the specific error that a criticism pointed out? Yes. How can refutation be an effective tool when people can keep making up non-refuted ideas? First, don't do that; try to operate in good faith and with integrity. Only propose ideas you think may be correct or may be useful to consider. But that's only a partial solution.

In general, adding auxiliary hypotheses ruins the explanations, concepts and principles involved in ideas. It turns them into arbitrary claims. Bad explanations and arbitrary claims can be criticized using epistemological criticism. If you only ever used topical criticism, then ad hoc claims would be a major problem. E.g. if the topic is dinosaurs, and you only ever made arguments about topical issues like the fossil record and what sort of teeth could penetrate what sort of hide, then people could (in bad faith) endlessly come up with arbitrary changes to their claims to avoid your refutations. But if you also use philosophical criticisms about bad methodology, bad explanations, arbitrariness, etc., then you can deal with it. Suppose someone said "That fossil appears to contradict my claims, but it doesn't really because it doesn't actually come from dead dinosaurs: God put it there to trick us." You can't refute that by making arguments about dinosaurs or archaeology, but you can give philosophical criticisms of it.

Popper covered the issue of auxiliary hypotheses in a reasonable way but I don't think he provided a full, complete solution. I think there's room for improvement. Deutsch helped contribute some ideas about modifications ruining explanations, and I think I've contributed something in this area too.

CF claims that when you start criticizing general categories of bad thinking, it becomes harder and harder to come up with arbitrary new ideas that don't fit a pattern that has already been criticized. And if someone does come up with a dumb new idea that isn't refuted by existing arguments, that's actually useful and helpful. Either it's actually a reasonable idea (if it doesn't fit the types of ideas we already have criticisms of, then we shouldn't assume it's bad) or else it's an example that will help us identify and criticize a new pattern of bad ideas.

Reading Recommendations

I wrote a previous article, Critical Rationalism Overview, and an Introduction to Critical Fallibilism.

Popper’s books are fairly difficult to read and understand, and I don't agree with everything, but they contain a lot of good ideas. Two of his best and most relevant books are Conjectures and Refutations and Objective Knowledge. I also put together a recommended list of selections from many Popper books. Following it is a way to read his most important epistemology ideas in fewer pages. My 22 selections total 541 pages, which is less than one of Popper’s big books.

Popper's books contain some great ideas that are also helpful for understanding Critical Fallibilism.

This article also has a companion video: