Appearance
1.4 — How to Check a Thought Before You Believe It
Here is a puzzle. Four cards are on a table. Each has a letter on one side and a number on the other. You can see:
A K 4 7
The rule to test is: if a card has a vowel on one side, it has an even number on the other. Which cards must you turn over to find out whether the rule is broken?
Take a moment before reading on.
Most people say A, or A and 4. Both are wrong. The correct answer is A and 7, and fewer than one person in ten gets it. Turning A is obviously right — an odd number behind it breaks the rule. Turning 4 is useless: the rule says nothing about what must be behind an even number, so whatever you find there cannot break it. And 7 is the one everybody misses — if there is a vowel behind that 7, the rule is dead.
This is the Wason selection task, from 1966, and it exposes the deepest bug in ordinary thinking. We instinctively look for the case that would confirm a rule, and we do not think to look for the case that would break it. That single habit, called confirmation bias, is behind a large share of bad decisions, stuck arguments, wrong beliefs about other people, and everything Part 11.5 catalogues.
Now the twist that makes this page useful. Reframe the identical logic problem: four people are drinking in a bar, and the rule is if you are drinking alcohol you must be over 18. The cards show: beer, cola, 25 years old, 16 years old. Which do you check?
Almost everybody instantly says beer and the 16-year-old. That is exactly the same logic as A and 7. The problem is not that your reasoning is broken. It is that abstract rules do not engage the machinery, and concrete social situations do. Which is the first practical lesson of this chapter: when you cannot think clearly about something, make it concrete and put people in it.
What critical thinking is not
It is not disagreeing. It is not scepticism as an attitude, and the person who reflexively doubts everything mainstream is running exactly the same automatic process as the person who believes everything mainstream, just with the sign flipped. It is not intelligence either — high-scoring people are, if anything, slightly better at constructing elaborate defences of things they already believed.
Critical thinking is the willingness to do a specific piece of work before accepting a conclusion, especially a conclusion you like. It is a procedure, not a personality. The rest of this page is the procedure.
Step one: separate the claim from the story
Before evaluating anything, state what is actually being asserted in a sentence that could be true or false. This sounds trivial and it is where most confusion dies.
Take: "Since they started working from home, standards have completely collapsed." Buried in there are at least four separate claims. That output fell. That quality fell. That home working caused it. And that "completely" is a fair description. Each has different evidence, and three of them are probably not what the speaker would defend if asked.
Three particular things to pull apart every time.
Fact from interpretation. "He left the meeting early" is a fact. "He stormed out" is an interpretation wearing a fact's clothing. Nearly all conflict between people runs on interpretations delivered as facts, which is why 12.7 makes this separation the first move in handling an argument.
Description from value. "This policy will reduce spending by 12 per cent" and "this policy is good" are different kinds of statement and cannot be settled by the same evidence. Evidence can settle the first. Only your values can settle the second, and pretending otherwise is how people talk past each other for hours.
The claim from the person making it. These are independent. A dislikeable person can be right and a kind one wrong.
Step two: ask what would change your mind
This is the single highest-value question in this volume, and it takes four seconds.
"What specific observation would make me abandon this belief?"
If you can answer it — "if the numbers came back flat for three consecutive quarters, I would be wrong" — you hold a real belief about the world, and you now know exactly what to look for. If you cannot answer it, you are not holding a belief. You are holding an attachment, and no amount of evidence will move it because you have not left a door for evidence to come through.
Run this on yourself first and on other people second, because used as a weapon it is merely a debating trick, and used inward it is the most reliable way to find out which of your convictions are actually about the world.
The related move is to make the belief pay rent in predictions. A belief that explains everything after the fact predicts nothing in advance. "He is a selfish person" explains his generosity as strategic and his stinginess as typical; it cannot lose, which means it is not telling you anything. A belief that earns its place says: if that is true, then next week when there is nothing in it for him, he will not help. Then you watch.
Step three: look for the missing comparison
Almost every misleading claim you meet is missing a comparison group, exactly as in 1.3, and once you start looking the pattern is everywhere.
"Most successful founders dropped out of university" — compared with what? Most unsuccessful founders also dropped out; there are simply far more of them and nobody writes about them. This is survivorship bias, and it is the reason advice from winners is worth so much less than it appears.
"Crime has gone up" — compared with when, and is that reported crime or committed crime, and did the reporting rules change?
"Nine out of ten dentists recommend it" — out of how many asked, and recommend it over what alternative?
The habit is a single question: compared to what? Say it out loud when a number arrives without one.
Step four: consider the base rate
This is the hardest of the moves and the one that fixes the most expensive errors.
A test for a disease is 99 per cent accurate. The disease affects one person in ten thousand. You test positive. What is the chance you have it?
The intuitive answer is 99 per cent. The real answer is about one per cent, and the arithmetic is simple enough to do in your head. Imagine a million people. About 100 of them have the disease, and the test catches 99 of those. The other 999,900 do not have it, and a one per cent error rate means about 9,999 of them test positive anyway. So roughly 10,098 people have a positive result and only 99 of them are ill. Ninety-nine out of ten thousand and ninety-eight is about one per cent.
The rare thing stays rare even after a good test points at it. This is why 14.1 warns you so strongly about recognising yourself in disorder descriptions, why a single striking behaviour is weak evidence that someone is a manipulator (13.2), and why unusual explanations need much more support than ordinary ones. When you hear hoofbeats, the base rate says horses.
Step five: name the move being used on you
Not every bad argument is an honest mistake. These are the patterns worth being able to name on sight, because naming one immediately defuses it.
Attacking the person instead of the claim. Sometimes disguised as concern about motives. Someone's funding is a reason to check their work harder, never a reason to conclude it is wrong.
Rebuilding the argument into a weaker one and beating that. "So you're saying we should have no rules at all." The counter is to restate what you actually said and ask them to address it.
The false choice. Two options presented as the whole field when there are five.
Loading the question. "Why are you always so defensive?" contains its conclusion. Answer the embedded claim, not the question.
Appeal to nature, tradition, or novelty. That something is natural, old, or new says nothing about whether it is good. Arsenic is natural.
Moving the goalposts. You meet their stated condition and a new one appears. Notice it aloud the second time: "last week the test was X, and now it is Y."
The unfalsifiable defence. "The fact that you disagree proves you are conditioned." Any claim built so that objection confirms it has left the realm of things that can be discussed.
Step six: know which of your own faults is doing the talking
Everybody's reasoning fails in a personal, characteristic way, and knowing yours is worth more than knowing the whole list in 11.5. The four that do the most damage:
Wanting it to be true. Motivated reasoning is not a failure of logic — the logic is often excellent. It is the direction of effort. You examine unwelcome evidence hard and welcome evidence hardly at all. The tell is asymmetry: notice when you have found three flaws in the study you disliked and none in the one you liked.
Already having decided. Once you have said a thing out loud, especially to people whose opinion you value, changing your mind costs something, and your reasoning quietly starts protecting the investment.
Vividness beating frequency. One horrifying news story outweighs a thousand safe journeys. Your sense of how likely something is comes mostly from how easily an example comes to mind, which is a measure of how memorable the thing is, not how common. This is the availability heuristic.
Judging by how it was said. Fluent, confident, well-produced statements feel truer. This effect is real and large, and it means the accuracy of a claim and its persuasiveness are almost independent variables — which is precisely the fact the next page is about.
What to do with this
Keep four questions where you can reach them. They take under a minute together and they are the whole of this page in usable form.
- What is actually being claimed, in one sentence that could be false?
- What would change my mind?
- Compared to what?
- How common is this kind of thing anyway?
Then add the two habits that do the rest.
Write the prediction down before you find out. Before the meeting, before the results, before the argument resolves: one line, dated, in a note on your phone. Then look back at it. This is the only reliable defence against hindsight, which will otherwise convince you every single time that you knew it all along. A month of this is genuinely humbling and permanently useful.
Argue the other side until you could pass for a believer. Not a caricature — the strongest honest version, the one an intelligent person who holds it would recognise as their own. If you cannot produce it, you do not understand the disagreement yet, and you are arguing with a copy of the position that you built. That practice has a name and a page of its own: 1.8.
Next: 1.5 takes the same machinery and points it at the hardest target — the beliefs you never chose, which arrived so smoothly that they feel like your own conclusions.