Karl Popper’s philosophy, Critical Rationalism (CR), rejects positive arguments (arguments in favor of ideas). It rejects trying to figure out how much support, evidence or justification an idea has. And it rejects induction, which is a type of positive argument.
CR says we learn by conjectures and refutations (which is literally evolution). In other words, we brainstorm unjustified guesses and we improve them with negative, critical arguments (a negative argument is an argument against an idea – a criticism). We learn by error correction. We have to look for and fix errors, but we can’t prove we’re right or probably right. We can’t make our ideas be better and better established (as true, probably true, good or preferable); instead we can test and criticize them and prefer ideas that survive criticism. We can tentatively establish that ideas are wrong, but there’s no way to establish that ideas are right, not even tentatively, fallibly, partially or imperfectly.
CR explains an asymmetry between positive and negative arguments: A universal scientific theory can be refuted by one counter-example, but cannot be proven true by a million data points that agree with the theory. So refuting evidence is highly effective while confirming evidence isn’t. CR also gives other arguments about why negative arguments work (fallibly, imperfectly) but positive arguments don’t work (at all).
I agree so far.
According to CR, how do we choose between competing ideas? We brainstorm criticisms, including testing ideas with scientific experiments. Some ideas will be refuted, and some won’t.
And then how do we choose between the remaining ideas, which we don’t have a clear refutation of, which are incompatible rivals? (Some ideas talk about separate issues, so there’s no need to choose between them. Rival ideas give conflicting answers to the same issue.)
CR says: Form a critical preference for the idea that survives criticism and seems best when taking into account our critical arguments. In other words, consider negative arguments and refute what you can (e.g. ideas that are contradicted by evidence or which contain internal logical contradictions). Then, taking into account the (negative) arguments, make a judgment about which non-refuted idea is best. How much do we like an idea, and think it’s good, compared to other ideas, when considering the (critical) arguments?
CR’s view is actually similar to the standard view (SV) that it rejects. SV says: Come up with both positive and negative arguments. Use negative arguments to decisively refute what you can (e.g. ideas that are contradicted by evidence or which contain internal logical contradictions). Then, taking into account the (both positive and negative) arguments, make a judgment about which non-refuted idea is best. How much do we like an idea, and think it’s good, compared to other ideas, when considering the (supporting and critical) arguments?
CR and SV both accept partially-effective arguments (which I call “degree arguments”, because their effectiveness is a matter of degree, rather than all-or-nothing). They view arguments as having degrees of strength. They look at a weighted sum of the number and strength of arguments for or against an idea (or in the CR view, only against it).
Note: Degree arguments may be numerical or non-numerical. They may be seen as adding or subtracting points from an idea’s score, or they may be discussed in non-numerical terminology like a “strong argument” or “weak evidence”. I don’t think it matters much whether you use numbers for your degree arguments or not, and my criticisms apply either way.
The shared idea between CR and SV is to judge ideas by their amount or degree of goodness. Look at how strong or weak various arguments are, and how many arguments apply to an idea, and estimate how good the idea is overall.
I reject strong and weak arguments. Instead of evaluating how good an idea is (the degree or quantity of goodness, plausibility, probability, support, confidence, evidence, justification, authority or many other things people call it), we should evaluate ideas as non-refuted or refuted. All criticisms (negative arguments) are either decisive or don’t do anything. An idea can’t partially survive criticism.
CR rejects positive arguments. Critical Fallibilism (CF) rejects both positive arguments and degree arguments. CF accepts only decisive, negative arguments. A decisive criticism says why an idea doesn’t work (for a purpose, in a context). In other words, it points out that an idea is an error and therefore will result in failure (unless the criticism is mistaken) at a particular goal or purpose. A decisive criticism is incompatible with the idea it criticizes: you can’t (rationally or reasonably) accept the criticism and also accept the idea. You should never act on or believe an idea while accepting a decisive criticism of it. By contrast, degree arguments point out good or bad things about ideas. It’s possible to have five negative degree arguments, telling you five bad things about an idea, but still regard it as the best available idea to accept, believe or act on.
Some people take this the wrong way and hear that e.g. they should quit their job immediately if it has any flaw that they can criticize. That’s not what I’m saying. When you take an action or accept an idea, there’s a purpose to it. Ideas are meant to work for something, succeed at something. A criticism says that an idea will fail at a purpose. You should never use an idea, for a purpose, which you accept a refutation of. If (due to a criticism) you believe idea X will fail to accomplish goal/purpose Y, then it never makes sense to use X to try to accomplish Y. Don’t act on known errors because (as best you know) that won’t work. You could still use idea X for some other goal which none of your criticisms apply to. Even if your job has flaws, it succeeds at some purposes like getting you some money. It’s not refuted for all purposes. That makes whether to quit more complicated. Don’t just find one flaw and give up on something for all purposes. One decisive flaw for an idea/purpose pair means you should reject that specific pairing.
CF also says that many positive arguments can be converted into negative arguments while keeping the same main point (same knowledge). This explains why many people think positive arguments work. Many positive arguments are imprecise or approximate, but are roughly equivalent to and convertible to negative arguments. Using positive arguments is actually OK when no one objects (it’s less precise, but we should increase precision when there is a problem, not just maximize precision at all times). Positive arguments should be converted to negative arguments, to make them more rigorous, if people have trouble reaching agreement or think the positive arguments are causing a problem in this case.
If a positive argument can’t be expressed in a negative way, then it must be incorrect. That’s because all correct positive arguments are convertible. However, only some incorrect positive arguments are convertible to negative arguments (being incoherent or illogical can screw up conversion). So if you can’t find a conversion, you should suspect the argument is incorrect.
The conversion from positive to negative arguments is generally simple. A positive argument says “Idea X is good because it has trait Y, and Y is good because…”. To convert to a negative argument, change it to “Ideas that lack Y are bad because…”. If something is good, then lacking it is bad, and ideas which lack it can be criticized for that.
Similarly, many degree arguments are imprecise statements which can be converted to decisive arguments while keeping the same main point (same knowledge). To convert a degree argument to a decisive argument, first convert it to a negative argument if it’s positive. Then, explain why the criticized flaw isn’t merely undesirable but will actually result in failure. To do this often requires clarifying the goal or purpose by specifying which outcomes count as success or as failure. That often involves determining a “good enough” breakpoint. Converting degree arguments to decisive arguments is harder than converting positive arguments to negative arguments, and often requires more creative thought.
Why convert or reject all arguments except decisive criticisms? There are a bunch of complicated criticisms of both positive and degree arguments. And the resulting system, built around decisive negative arguments, is elegant and useful, in addition to not having any known criticisms of it that it can’t answer.
In short, positive arguments don’t work because it only takes one error to fail. If fifty things go right, and one important thing goes wrong, the result is failure. So saying all the things you think will go right never tells you whether your idea will actually work. You need to look for reasons an idea won’t work. If you can’t find any errors, then it might work. And logically there’s no way to ever get from “this idea has these good traits” to “it doesn’t have errors”.
And, in short, degree arguments don’t work because they don’t focus on what’s important. What are the requirements for success? We should look for errors that will cause failure. Any “error” that is compatible with success is not really an error. If we avoid anything that will cause failure, then we’ll succeed. That’s what we should be aiming for. Many degree arguments analyze how well something does at a local optimum that doesn’t matter to the constraint, bottleneck or limiting factors for the global optimum, purpose or goal. (Most local optima have excess capacity, so optimizing them is unimportant, which also means that they can’t be converted into decisive arguments since they aren’t about something important).
Also, error correction is inherently digital, which is incompatible with matters of degree (which are analog). I won’t discuss the (complicated) reasons here except to explain what “digital” means. Digital means discrete, separate things instead of a continuum or spectrum. The integers are digital while the real numbers are analog. Categories or types are digital, while amounts, quantities and degrees are analog. Digital means things are in separate “buckets” instead of having different amounts or degrees of the same thing. Understanding which things are digital or analog – when to use each approach – is important and useful for life (I’m advocating digital as superior for critical evaluations of ideas, but analog is good for other things). For error correction to work well, it’s also important to clearly specify what is and isn’t success, rather than trying to have amounts or degrees of success (“degree goals”).
Related is CF’s idea of considering many different goals (digital choices) instead of just one (which is commonly done as an analog spectrum with degrees of success). Instead of talking about achieving “the” goal, people should define more specific goals and evaluate whether an idea succeeds or fails at each of the goals. For example, “get money” is a vague degree goal (you can basically get any amount of money, so there’s a spectrum of different outcomes). “Get at least $200” and “Get at least $300” are examples of more specific goals. An idea that gets you $250 would succeed at one of those goals while failing at the other. It’s better to understand its decisive success at one goal, and decisive failure at another goal, than to vaguely say that it has a medium amount of goodness for getting money.
Does this approach lead to millions of goals – e.g. one for each amount of money you could get? That’d be too much to think about, which is one reason people favor degree goals. The solution is to only think about important goals. An important goal is one that corresponds to a breakpoint, so the number of goals to consider equals the number of breakpoints. Breakpoints are specific amounts that matter for a purpose. E.g. if you have zero dollars and want to buy a steak, then the breakpoint is the price of the steak: if you get that much or more you can buy the steak, but if you get less money then you can’t.
The most common number of breakpoints is one, and each higher number is significantly rarer than the previous (two breakpoints is rarer than one; three is rarer than two; etc). So having more than five breakpoints is rare. People may also be mistaken about breakpoints, which can lead to extra breakpoints being proposed and considered, but the number of breakpoints (and therefore goals) considered should remain low and manageable. (In the rare cases where there are too many breakpoints, there are techniques to reduce the number. It also can be worth the effort to analyze many breakpoints, especially for projects with a large budget and hundreds of people.)
Besides their vagueness and incompatibility with error correction, degree goals are also bad because they ignore breakpoints.
For completeness, there are binary goals, degree goals, and also digital but non-binary goals. Binary means “two”, and binary goals are evaluated with only two outcomes: success or failure. This is similar to binary logic, which is the familiar logic where propositions are evaluated with only two options: true or false. A digital but non-binary goal might have three or six outcomes. The upside of that is it could potentially be compatible with error correction, unlike analog approaches. The problem is there are no known, reasonable, useful alternatives to success and failure that aren’t a matter of degree. It’s like trying to come up with alternatives to true and false in order to invent a new, better form of logic.
Breakpoints, constraints, bottlenecks, limiting factors, local optima, global optima, and excess capacity are all important topics that merit more explanation. I’ll write about them elsewhere and you can also learn about them from Eli Goldratt’s Theory of Constraints (TOC), which he wrote about in books like The Goal. I rejected degree arguments before studying TOC, based on thinking about CR, but then I found that TOC concepts independently converge on the same conclusions.
I’ve written a lot about CF’s rejection of degree arguments, which I sometimes call “Yes or No Philosophy” or “Binary Epistemology”:
In Rejecting Gradations of Certainty, I criticize using inconclusive reasoning to score some non-refuted ideas above others.
One Criticism Is Decisive addresses people’s belief that we need a better idea before we reject an idea. Actually, we should always reject an idea with only one refutation, and we can always easily come up with a better idea. At worst, “I don’t know” is better than an error, but there are often better meta ideas easily available.
In part 5 of my Aubrey de Grey discussion, I explained epistemology in terms of judging ideas as refuted or non-refuted rather than by amounts of justification or goodness. The whole discussion with de Grey is good, too.
In Epistemology Without Weights and the Mistake Objectivism and Critical Rationalism Both Made, I rejected weighing inconclusive evidence or arguments in order to favor one idea over another.
Critical Rationalism Epistemology Explanations helps explain the “hard to vary” criterion and how refutation works.
Corroboration is a short post about what’s good about an idea surviving criticism. My answer is that we build up an archive of known criticisms. That makes it harder to come up with new rival ideas which are not already refuted by existing criticism.
Regress Problems discusses how Yes or No Philosophy avoids infinite regresses that other epistemologies struggle with.
In Criticism is Contextual, I give an example of how the “same idea” can work in one context but not another. In other words, a solution can solve some problems but not others, or work in some situations but not others. Criticism says why an idea fails at a goal in a context, not that an idea is universally, inherently or essentially bad.
I sell organized, polished material teaching about these issues. Yes or No Philosophy ($400) is about my best original philosophy idea: judging ideas in a binary way as refuted or non-refuted, rather than evaluating degrees of goodness. I spent a month making those videos and articles.
I taught my Critical Fallibilism Course ($880) to students over Zoom. You can buy videos of the eight sessions for a quarter of the live course’s price. It focuses on teaching tools for thinking more effectively. It explains a lot of Yes or No Philosophy, including how it connects with TOC ideas like breakpoints, constraints, excess capacity, and local and global optima.