Critical Thinking
Working out whether a claim is actually supported, especially a fluent one.
The behaviour, in one line: I check the evidence and reasoning behind claims — including confident AI output — before I accept them.
Given a rule to test, almost everyone goes looking for the case that would confirm it and skips the case that could break it. Critical thinking is largely the discipline of turning over the second card.
You are handed more claims in a day than you can check, and a growing share of them are produced by something that writes fluent, confident, well-formatted text whether or not it is true. Critical thinking is asking what the evidence is, who benefits if you believe it, and what would have to be true for it to hold. Fluency is not evidence. It was never even a good proxy for it.
| What it is | What it isn’t |
|---|---|
| Asking what would break it | Asking what supports it |
| The evidence behind the claim | The confidence in front of it |
| Checking what you agree with | Checking what you already doubt |
| Noticing what is not shown | Weighing what is shown |
What it looks like. A model hands you a well-structured answer with three plausible figures in it. You check the one you would most like to be true first, because that is the one you are least likely to come back to.
- The vendor demo. Everything works on stage and the number on the slide is enormous. Nobody has asked what the demo data was, or what happens when the API is slow.
- The confident wrong answer. Fluent, well formatted and quietly false. It reads better than what you would have written, which is exactly the problem.
- The number that supports your case. The hardest one to check, because checking it can only cost you.
- Ask what would change your mind. A claim with no answer to that question is not being evaluated, only agreed with.
- Separate the three things that arrive together: the claim, the evidence for it, and the incentive of whoever is making it.
- Look for what is not there — the missing comparison, the counterexample, the year the data stops.
- Check confident output the way you would check a confident stranger. The confidence is a property of the writing, not of the facts.
- Break it into parts
Four parts: asking how you know, separating claim from evidence from incentive, noticing what is missing, and checking confident output.
- The part that carries it
“What would change my mind?” A claim you cannot picture being wrong is not one you are evaluating.
- The order to learn it in
- Start with the source-and-incentive split, on claims you already doubt.
- Then apply “how do I know this?” to claims you agree with. Much harder, and where the value is.
- Finally, make checking confident output routine rather than occasional.
- Put something at stake
Name one belief you hold and the evidence that would overturn it, out loud, to someone — then go and find out whether that evidence exists.
- Compress it
How do I know this? Who benefits if I believe it? What is missing? What would change my mind?
- The smallest useful dose
One claim you agree with, checked properly, each week.
- Anchor it to something you know
A courtroom. Fluency is not testimony, confidence is not evidence, and the witness who was never called matters most.
“Extraordinary claims require extraordinary evidence.”
— Carl Sagan
The interactive version is in the course. The diagram, the simulation and the four-question check for Critical Thinking run in the free course, where your progress across all 24 skills is kept.
Open Critical Thinking in the course →
Not sure where you stand? Take the twelve-minute self-assessment, or see the 30-day course that turns these skills into a roadmap.