The below is what I wrote this morning. It’s an unedited first draft. It’s not rewriting a prior draft. I did no outlining or planning ahead. I didn’t come up with a good title. I tend not to post stuff like this for a few reasons. I don’t generally get useful enough feedback, I don’t think sharing this kind of writing addresses important bottlenecks or breakpoints like the ones in the way of getting 1,000 paying fans, and I think posting stuff that isn’t curated and edited enough can confuse people. But I like to occasionally share other types of things for variety and this one seems relatively harmless. Unpolished/unedited/fast work shares information about one’s intuitions and knowledge that could enable additional useful criticism for me and learning opportunities for others.
Critical Fallibilism (CF) is primarily about problem solving, learning, critical thinking, error correction, knowledge creation or decision making, which it says are all largely the same thing. In each case, you brainstorm ideas and analyze them critically. Criticism helps find and correct errors. The combination of brainstorming and error correction is evolution: it’s the same general principle behind biological evolution: replication + variation + selection = knowledge. Brainstorming covers replication and variation while criticism covers selection. The idea of evolution is that trying random changes from the best knowledge so far, then doing error correction, and repeating that process a bunch, can make progress. Besides explaining how to make progress, this process also explains the origins of knowledge. It can work with an arbitrary starting point: a replicator that was generated by random accident, not a well-designed replicator.
Problem solving is done by creating knowledge of the solution to a problem. Decision making is done by creating knowledge of what decision to make. And decisions are solutions to problems. That’s abstract but it can be viewed more concretely too.
CF says, to solve a problem, we brainstorm a list of candidate solutions. And we come up with criticism and criteria for evaluating the solutions. The criteria are binary (pass/fail). Non-binary issues are converted to binary with breakpoints. Then we pick a solution that is non-refuted based on the criteria we care about. We want something which doesn’t fail at our goal (solving the problem); something that will actually work. If we can’t figure that out, we may change our goal; we might realize we were too ambitious and that something less would be satisfactory.
For decision making, we brainstorm a list of candidate decisions (plans, options), then come up with binary criticisms/criteria to evaluate them with, then pick a non-refuted decision. The method is the same.
For doing scientific research, we brainstorm ideas about a scientific issue, then evaluate them with binary criticisms/criteria, then conclude something that passes rather than fails the critical criteria we care about.
For learning, we brainstorm ideas, evaluate with binary critical criteria, and only learn non-refuted ideas, never refuted ideas.
Details vary, like for learning existing knowledge we may read a textbook and base our brainstorming heavily on the ideas it suggests. For making new discoveries in a field, we’ll have less help and guidance for our brainstorming. For decision making, we often will want to reach agreement with other people (when a decision affects other people at the same business, or others doing the same group activity, or others in your family). But despite a few differences, it’s still the same method.
If there are multiple non-refuted ideas, what should you do? This is another generic part of the method: you can do more critical thinking and come up with more critical criteria that differentiate between them or you can be satisfied and use any of them. If multiple solutions will solve your problem, you can be happy and use one of them, or you can be more ambitious and set a harder goal that asks for a bit more.
A key part of the method is that we evaluate multiple ideas for multiple criteria. The same idea gets more than one evaluation. Instead of trying to evaluate how good an idea is, as a single complex answer, we use many simple (pass/fail) evaluations. Complexity comes from combining multiple evaluations (most often with “and”: we want to pass not fail at a list of things).
People often argue about ambiguous ideas. Because of their premise that ideas can only have one evaluation, they argue about which is the real meaning of the idea that its evaluation should be based on. CF says to write down each distinct meaning that the ambiguous idea could mean that anyone is interested in, then evaluate each of those separately.
CF says that you can take many versions of a complex idea like capitalism, socialism or induction and brainstorm them all as separate ideas. Instead of trying to figure out how good capitalism is, you can try to analyze what variations of capitalism exist and then for each well-defined version you’ll have a much easier time evaluating if it passes or fails each critical criterion. The critical criteria should also be specific enough to be fairly easy to evaluate. “Is it awesome? pass/fail” is the wrong way to do it. You need less ambiguous criteria. There are many definitions of awesome. Pick some and make them separate criteria.
Clarity is one of the big goals here. The simplicity of pass/fail enables dealing with much larger numbers of ideas, criteria and evaluations. And dealing with far more things lets us have many similar things, which enables a lot of disambiguation. This system organizes complexity differently than the usual approaches and that has advantages.
Another reason for pass/fail is that degree of goodness based systems fundamentally, logically don’t work. My biggest motivation with CF was to figure out a system that works at all. Most don’t. But it’s unsurprising that making a working system would bring a variety of advantages that were missing from broken systems.
How are people able to think at all if they’re using broken systems? Introspection is hard. They don’t know in detail what they really do. How can I know that? By logical analysis of what can and can’t be done. If someone says they guessed what their friend was thinking using telepathy, I can conclude that they’re wrong without having to understand what’s going on in their head better than they do or know anything about them. I just have to know that telepathy is scientifically impossible, so they must be doing something else and be confused about what they’re doing. They probably just have good intuitions about their friend based on a mix of past experience and some skills like cold reading or warm reading that many untrained people are pretty good at.
If a method says “Step 4: identify the one pattern.” but there are actually 500 patterns, you cannot use that method. Also if it doesn’t say how to identify patterns at all, it isn’t usable, but assuming it provides a pattern-identifying method, and that method commonly yields 500 results, and the method relies on it yielding exactly one result, then the method doesn’t work. People familiar with my other writing may know I’m talking about induction here: many modern versions of induction rely on pattern finding but fail to provide any answer to which pattern(s) to focus on out of the infinitely many that logically exist for the data. They also often give inadequate guidance about how to find any patterns at all, but that’s a much lesser problem, since finding patterns is certainly possible.
One way to interpret induction is just assume whatever pattern you find first is the right one. Make sure you have a positive not cynical mindset so your intuition guides you towards a good choice of pattern and you don’t self-sabotage by finding a silly pattern first. Your intuition has a pretty good idea of which patterns may be important, so just rely on that. But how does your intuition know that? There must be some method that your intuition uses and induction fails to explain how that could work. My answer to how it works is it uses evolution: it guesses and criticizes patterns in order to find patterns useful for your goals. So that’s not induction. You just figure stuff out with CF or CR thinking, then use those answers to make induction look like it’s working. Induce the knowledge that you created using non-induction will get you decent knowledge, but the induction step that’s tacked on at the end isn’t helping anything and is actually detrimental.