Chapter 1 · Part I — The Mindset

Belief Isn’t the Enemy

Humans run on inherited belief — and that's how we took over the planet. The real skill isn't believing less. It's knowing which beliefs deserve your trust.

This book begins by questioning the idea that started it. The idea goes like this: the world runs on belief, not critical thinking. Beliefs passed down through generations, like Chinese whispers, have damaged people's ability to think. Thinking is what separates humans from animals, and we need to get back to it.

Part of that is exactly right, and part of it is almost exactly backwards. Sorting out which part is which is the best possible first lesson, because it's the whole skill in miniature: take a belief you care about, and look at it honestly.

1.1The explorers who starved in a pantry

Researchers now think the problem was the preparation. Raw nardoo contains an enzyme that destroys vitamin B1 in the body. The Yandruwandha ground the seeds with water and prepared them in ways that neutralised it. The explorers copied the what — collect, grind, bake — but skipped steps whose purpose they couldn't see. They ate full meals while their bodies ran out of an essential vitamin.

These were intelligent, educated men who reasoned their way to the dangerous version. What they lacked was not intelligence. It was inherited knowledge — the kind nobody could have worked out in a few weeks by thinking hard.

1.2A recipe smarter than the cook

The anthropologist Joseph Henrich, whose book The Secret of Our Success this chapter leans on, gives a sharper example: manioc (cassava), a staple root in the Amazon.

Bitter manioc contains compounds that release cyanide. Traditional processing is long and tedious — scraping, grating, washing, boiling, then leaving the pulp to sit for days before cooking. Now imagine a sensible, independent-minded cook who notices that boiling alone removes the bitter taste. Why waste two days? She simplifies the recipe. Nothing bad happens. Not this week, not this year.

The damage comes slowly: chronic cyanide poisoning builds up over years. By then, nobody can connect the illness to the shortcut. The link between cause and effect is simply too long and too hidden for one person to spot by thinking.

This really happened on a large scale. When the Portuguese carried manioc from South America to Africa, the plant travelled but the full processing tradition often didn't. Henrich notes that chronic cyanide poisoning from poorly processed cassava is still a health problem in parts of Africa, centuries later.

1.3Not the smartest animal — the best copier

Go back to your prediction. Toddlers and chimpanzees are roughly matched at solving problems about the physical world. Where human children win by a mile is in learning from others: watching, imitating, trusting, absorbing what the group already knows.

Henrich's argument is that this — not raw individual intelligence — is our species' real advantage. No single human could reinvent fire-making, boat-building, which plants are poisonous and how to make them safe, or how to hunt seals on Arctic ice. Those things were built by thousands of people over thousands of years, each one adding a little and passing it on. He calls this our collective brain.

That forces a correction to the thesis this book started with:

“The reason humans grew over other animals is the ability to think.”— The starting thesis

A better version: humans won because we are extraordinarily good at believing what others tell us and passing it on. Inherited belief isn't a bug in human thinking. It's the engine that built civilisation. You believe the Earth goes round the Sun, that germs cause disease, that your medicine contains what the label says — and almost none of that did you check yourself. You couldn't. Nobody can.

1.4So where does it go wrong?

The Chinese whispers part of the thesis is real. The same machinery that carries wisdom also carries noise, and it can't tell the difference by itself.

The milk miracle was mostly harmless. The next example wasn't.

Notice what these stories have in common, and how they differ from manioc. In both, the claim was checkable, cheaply, and the stakes were high — yet people acted on it at many hops away from any actual evidence. Nobody in the chain had seen a kidnapping. Most people who saw the milk miracle had only heard about it.

1.5Chesterton's fence

In 1929, the writer G. K. Chesterton described a reformer who comes across a fence or gate built across a road for no obvious reason. The modern reformer says, “I don't see the use of this; let us clear it away.” Chesterton's reply:

“If you don't see the use of it, I certainly won't let you clear it away. Go away and think. Then, when you can come back and tell me that you do see the use of it, I may allow you to destroy it.”— G. K. Chesterton, The Thing (1929)

This is not an argument for keeping every fence. It's an argument for finding out why before you remove it. Sometimes you'll find the reason is dead — a rule made for a world that no longer exists — and you can remove the fence with a clear conscience. Sometimes you'll find the fence was holding back a flood.

You meet fences at work all the time. An old procedure at a property that everyone calls pointless paperwork. A supplier nobody is allowed to switch. A check-in step that seems to waste guests' time. Before you tear one out, find the person who built it, or the incident that caused it. It might be a legal requirement, or the scar of a past disaster.

1.6The tool: Belief Triage

Here's the first tool in your kit. Hospitals use triage to decide who needs attention first. You can do the same with beliefs, because you can't examine them all.

1.7Where this leaves the thesis

The original thesis had a real insight and a wrong villain. The insight: much of what people believe arrives through long chains of retelling, and those chains distort. The wrong villain: belief itself.

A sharper diagnosis is this. We evolved to trust what our community passes down, in small groups where the chain was short and people could see who knew what. Today the chains are enormous, fast and anonymous. A WhatsApp forward can travel a thousand hops in a day, arriving with the same feeling of “someone I trust sent this” as your grandmother's recipe. Our trust instincts weren't built for that.

So the mission isn't to replace belief with thinking. It's to help people become better at deciding what to trust: which beliefs to hold firmly, which to hold lightly, which to check, and which fences to understand before tearing down.

That version is more accurate. It's also far easier to teach. You can't win many people over by telling them their family's beliefs are corrupted. You can by showing them how to sort beliefs into the right boxes — and letting them do the sorting themselves.

Sources and further reading

• Joseph Henrich, The Secret of Our Success (Princeton, 2015) — Burke and Wills, manioc processing, the collective brain.
• Herrmann, Call, Hernández-Lloreda, Hare & Tomasello, “Humans have evolved specialized skills of social cognition,” Science (2007).
• G. K. Chesterton, The Thing (1929), chapter “The Drift from Domesticity.”
• “Ganesha drinking milk miracle” — Wikipedia summary of 1995 reporting, including the Ministry of Science and Technology coloured-milk test.
• “Indian WhatsApp lynchings” — Wikipedia summary of 2017–2020 reporting, including the Rainpada (Dhule) killings.
• US FDA, Bad Bug Book — entry on phytohaemagglutinin in red kidney beans.

Chapter 2 · Part I — The Mindset

Scouts and Soldiers

Intelligent people don't usually believe wrong things because they can't reason. They believe them because part of their reasoning is quietly working for the other side.

Chapter 1 was about beliefs that reach you from outside. This chapter turns the lens inward. Even with perfect information, your own mind has a motive — and it isn't always accuracy.

This chapter follows Julia Galef's The Scout Mindset (2021). Her central image: when you reason, you are sometimes a soldier, defending a position against threats, and sometimes a scout, whose only job is to map the terrain as it really is.

2.1The officers who knew they were right

What's striking is not that the officers lied. Most of them believed their reasoning. Every piece of evidence — a matching handwriting, a non-matching handwriting, an empty house — got interpreted as more proof of guilt. From the inside, it felt exactly like thinking.

Psychologist Thomas Gilovich describes the switch this way. When we want to believe something, we ask, “Can I believe this?” and look for a single reason that permits it. When we don't want to believe something, we ask, “Must I believe this?” and look for a single reason to escape. Same mind, two different standards of evidence, and no sense of switching between them.

2.2Why the soldier is so attractive

If soldier mindset leads us wrong, why is it everywhere? Because it pays — immediately. Galef lists six benefits it gives us:

Every one of these is real. That's why lecturing people about their biases rarely works: you're asking them to give up comfort, pride and belonging in exchange for an abstract thing called accuracy.

Galef's counter-argument has two parts. First, the benefits of soldier mindset arrive now, but the costs — bad decisions, missed warnings, wasted years — arrive later, so we systematically underrate them. Second, you can usually get the same benefits honestly. You can find comfort in having a plan, rather than in pretending there's no problem. You can build morale by knowing a bet is worth taking even at 40% odds, rather than by believing the odds are 100%.

That second point matters for your teaching. You won't change minds by attacking the benefit someone gets from a belief. You change minds by offering another way to get it.

2.3Signs you're actually a scout

Almost everyone believes they're a scout. So Galef suggests looking at behaviour, not feelings. Be honest — nobody else will see this list.

2.4Five tests that catch the soldier

Motivated reasoning is invisible from the inside, so you need tricks that make it visible. Galef's five tests all do the same thing: they change one detail of the situation and check whether your judgement changes too. If it does, the detail was doing the reasoning, not the evidence.

1 · The double standard test

Are you judging this person, or group, by a standard you wouldn't apply to others? When your manager is late, he's “busy with important things”. When a junior staff member is late, they're “careless”. Swap the people and see whether your verdict survives.

2 · The outsider test

Imagine someone with no emotional stake looking at your situation. What would they do? In 1985, Intel was losing its memory-chip business to Japanese competitors. Memory chips had been Intel's identity. CEO Andy Grove asked his chairman, Gordon Moore: if we got kicked out and the board brought in a new CEO, what would he do? Moore answered at once: he'd get us out of memories. Grove's reply was, in effect: then why don't we walk out the door, come back in, and do it ourselves? They did, and Intel became the microprocessor company.

3 · The conformity test

Imagine the people whose opinion you respect suddenly announced the opposite view. Would you still hold yours? If your confidence would collapse, it may have been resting on them, not on the evidence.

4 · The selective sceptic test

Imagine the evidence pointed the other way. A study says your preferred diet works, and you share it. How closely would you have checked the same study if it had said the diet didn't work? Apply the scrutiny you'd use for unwelcome evidence to welcome evidence too.

5 · The status quo bias test

Imagine your current situation weren't the default. Would you actively choose it? If you weren't already running this service, living in this city, or keeping this supplier — would you start today? If the answer is no, “it's how things are” may be doing the work.

2.5How sure are you, really?

Scouts think in shades of grey. Instead of “this will work”, a scout says “I'd put this at about 70%”. But most of us have never practised putting numbers on beliefs, and our words — “definitely”, “probably”, “I'm sure” — hide a lot.

Galef suggests the equivalent bet test. Take a claim you feel sure about. Imagine two bets, each paying the same prize. Bet A wins if your claim turns out true. Bet B wins if you draw a red ball from a bag. Change the number of red balls until you can't decide between the two bets. That number is your real confidence.

2.6Hold your identity lightly

The beliefs that are hardest to examine are the ones that have become part of who you are: your politics, your religion, your profession's way of doing things, even “I'm a critical thinker.” Once a belief is part of your identity, evidence against it feels like an attack on you, and the soldier takes over.

The essayist Paul Graham suggested keeping your identity small. Galef's gentler version is to hold identities lightly: you can still be an entrepreneur, a Hindu, a liberal or a conservative, but treat these as facts about you rather than flags to defend. The test is simple: can you hear a good argument against your side and feel curious rather than threatened?

One practical tool is the Ideological Turing Test, a term coined by the economist Bryan Caplan. Can you explain a view you disagree with so well that people who hold it would say, “Yes, that's exactly why I believe it”? If you can only produce a caricature, you don't yet understand what you're rejecting.

Sources and further reading

• Julia Galef, The Scout Mindset: Why Some People See Things Clearly and Others Don't (Portfolio, 2021) — the Dreyfus story, the six benefits, the five tests, the equivalent bet.
• Thomas Gilovich, How We Know What Isn't So (1991) — “Can I believe it?” versus “Must I believe it?”
• Ola Svenson, “Are we all less risky and more skillful than our fellow drivers?” Acta Psychologica (1981).
• Andrew S. Grove, Only the Paranoid Survive (1996) — the Intel memory-chip decision.
• Paul Graham, “Keep Your Identity Small” (essay, 2009); Bryan Caplan, “The Ideological Turing Test” (2011).

Chapter 3 · Part I — The Mindset

Anatomy of a Claim

Before you can judge whether something is true, you have to know exactly what is being said, and why. Most bad reasoning is never refuted. It just never gets taken apart.

Chapters 1 and 2 were about attitude: when to trust, and noticing when you want something to be true. This chapter gives you the basic anatomy you'll use in every chapter that follows: claims, arguments, assumptions, and how to tell whether reasons actually support a conclusion.

3.1What kind of claim is it?

A claim is any statement that could be true or false, or better or worse supported. Claims come in different kinds, and each kind is tested differently. Mixing them up is one of the most common sources of pointless arguments.

Notice how a single business conversation can move through all five in two minutes. When two people seem to disagree about a policy, check whether they actually disagree about a fact, a cause, a prediction or a value. Often they agree on the facts and are arguing about values without realising it, or the other way round.

3.2Pin the claim down

Vague claims are almost impossible to refute, which is exactly why they spread. “This oil is good for your health.” Good how? For whom? Compared with which other oil? How much of it? A claim that can't be wrong can't tell you anything either.

Before you evaluate a claim, pin it down. You'll often find that the pinned version is either obviously checkable or quietly much weaker than it sounded.

3.3Arguments: reasons that point to a conclusion

An argument is a set of statements where some — the premises — are offered as reasons to accept another — the conclusion. Words like so, therefore, which means, that's why usually introduce a conclusion. Words like because, since, given that usually introduce a premise.

In real life, arguments come out messy and half-stated. The first job is to reconstruct them: write out the premises and conclusion as clearly as you can. Here's an example you might hear in any business.

Reconstructed, it looks like this:

1. P1. Bookings fell 20% in August.
2. P2. The price rise came just before the drop.
3. C1. So the price rise caused the drop.
4. C2. So we should reverse the price rise.

Written out like this, you can already see the gaps. P1 and P2 might both be true, and C1 still doesn't have to follow. And C2 needs more than C1.

3.4Hidden assumptions

Almost every real argument leans on premises nobody says out loud. These hidden assumptions are where most weak reasoning lives, because nobody examines what nobody states.

To get from P1 and P2 to C1, the argument quietly assumes:

• Nothing else changed in August — no monsoon, no new competitor, no bad review, no change in how the listing appeared on booking sites.
• A 20% drop is unusual — bigger than the normal month-to-month swing. (Was August last year also down?)

And to get from C1 to C2, it also assumes:

• The bookings lost cost more than the extra margin earned on the bookings kept. A 20% drop in bookings with a 30% higher price can still mean more profit.

The test for a hidden assumption is simple: if this were false, would the conclusion still stand? If not, the argument depends on it, so it deserves to be checked.

3.5Valid, sound, strong

Three words let you say precisely what's wrong with an argument.

• Valid: if the premises were true, the conclusion would have to be true. Validity is about structure only — the roses argument failed here.
• Sound: valid and the premises actually are true. This is what a deductive argument needs to prove its conclusion.
• Strong: most everyday arguments aren't meant to be airtight. They're inductive — the premises make the conclusion probable. We call these strong or weak rather than valid or invalid.

So there are two separate ways to attack any argument: challenge a premise (“that isn't true”) or challenge the link (“even if that's true, it doesn't follow”). Keeping them apart stops you from arguing past each other.

3.6Fallacies are questions, not labels

You've probably seen lists of logical fallacies. They're useful, but they're easy to misuse. Shouting “ad hominem!” wins nothing if you can't explain why the reasoning fails. And spotting a fallacy doesn't prove the conclusion is false — a bad argument can have a true conclusion, as the roses showed. Treating a fallacy as proof the other side is wrong is itself a fallacy, sometimes called the fallacy fallacy.

A better approach is to turn each fallacy into the question it's warning you to ask.

3.7Steelman before you judge

In Chapter 2 you met the Ideological Turing Test. Its everyday cousin is the steelman: before you criticise an argument, rebuild it in its strongest reasonable form. Fill in the most plausible hidden assumptions, not the silliest ones. Then judge that version.

There are two reasons to do this. First, if you can defeat the strongest version, you've actually learned something; defeating the weakest version teaches you nothing. Second, people notice when they're being understood. Nobody changes their mind for someone who has just misrepresented them.

3.8Putting it together: the Claim X-Ray

Try it on the kind of message that arrives in family groups every week. This one is a composite of several that circulate widely.

Sources and further reading

• Evans, Barston & Pollard, “On the conflict between logic and belief in syllogistic reasoning,” Memory & Cognition (1983) — belief bias.
• Anthony Weston, A Rulebook for Arguments — a short, practical guide to building and reconstructing arguments.
• IARC (WHO), press release on radio-frequency electromagnetic fields, Group 2B classification (2011).
• Karipidis et al., systematic review on mobile phone use and cancer, Environment International (2024), commissioned by the WHO.

Chapter 4 · Part II — Reasoning & Evidence

Why We Reason

If reasoning evolved to find the truth, it's strangely bad at the job. But if it evolved for arguing with other people, it's surprisingly good. That difference changes how you should learn — and how you should teach.

Part I was about attitude. Part II is about machinery: how arguments, evidence, numbers and causes actually work. It starts with the machine itself. What is human reasoning for?

This chapter follows The Enigma of Reason (2017) by the cognitive scientists Hugo Mercier and Dan Sperber. Their answer is surprising, and it's the single most useful idea in this book for anyone who wants to teach.

4.1The enigma

For centuries, reason was treated as our noblest faculty — the thing that lets us rise above instinct and see the truth. But experiments since the 1960s show that reasoning is full of flaws. We look for evidence that supports what we already think. We find it easy to invent justifications for our choices after the fact. We're persuaded more by conclusions we like than by logic, as you saw with the roses in Chapter 3.

The most consistent flaw has a name: myside bias. When we reason, we overwhelmingly produce arguments for our own side. Ask someone to think about a question and they'll mostly list reasons supporting their first instinct, and very few against.

If reasoning were designed to help lone individuals discover the truth, this would be a terrible design. A truth-seeking tool should look hardest for the reasons you're wrong. So either evolution did a remarkably bad job, or we've misunderstood what reasoning is for.

4.2Reasoning is for arguing

Mercier and Sperber's answer is the argumentative theory of reasoning. Reasoning didn't evolve to help us think alone. It evolved for social life: to produce reasons that justify ourselves and convince others, and to evaluate the reasons other people give us.

Seen that way, the flaws start to make sense:

• We're biased and lazy when producing arguments. If your job is to convince someone, looking for reasons you're wrong is wasted effort — the other person will supply those. So you produce reasons for your side, quickly and cheaply.
• We're sharp when evaluating other people's arguments. If you accepted every argument you heard, anyone could manipulate you. So when someone else argues, you check carefully.

The researchers called this the selective laziness of reasoning. We aren't bad at spotting weak arguments. We're bad at spotting weak arguments when they're ours.

4.3When groups get smarter

Go back to the card puzzle. Why did groups do so much better? Because when someone in the group says “E and 4”, someone else asks, “Why the 4? What would it prove?” The person who spotted the 7 can explain. Good arguments are easy to recognise once someone says them out loud, even if you couldn't have produced them yourself.

This is a division of cognitive labour. Each person does the cheap part — offering reasons for their view. Everyone does the valuable part — checking each other's. For problems where the right answer can be demonstrated, the truth tends to win.

But this only works under certain conditions:

• There has to be real disagreement. If everyone starts with the same view, nobody produces the counter-arguments.
• People have to share the goal of getting it right, rather than winning, saving face or pleasing the boss.
• It has to be safe to say “you're wrong”, including to someone senior.

4.4When groups get dumber

Remove those conditions and group reasoning goes wrong in predictable ways.

This is group polarisation. When like-minded people discuss, each person hears mostly arguments for the side they already lean towards, plus the social reward of agreeing. Nobody plays the evaluator. Think of a family or community WhatsApp group where everyone already agrees: every day of discussion pushes it further in the same direction.

Two other failures are common in workplaces:

• Groupthink — the psychologist Irving Janis's term for cohesive teams that suppress doubts to keep harmony. His classic case was the US government's 1961 Bay of Pigs invasion, a plan whose obvious flaws advisers kept to themselves.
• Deference to rank — when the most senior person speaks first, everyone else anchors on their view or stays quiet. The group ends up with one person's reasoning and the illusion of a discussion.

4.5Rules for reasoning together

If good reasoning happens between people, then the most practical skill isn't thinking harder on your own. It's designing conversations where the evaluator in everyone gets switched on.

4.6What this means for you

As a learner

You can't reliably be a scout alone. Your myside bias will quietly build the case for whatever you already believe, and it will feel like careful thought. What you need are sparring partners: people who will disagree with you in good faith and whose arguments you'll actually evaluate.

AI assistants can help — but with a warning. AI models tend to agree with the person they're talking to. If you ask “Isn't my plan great?”, you'll probably hear yes. Ask instead: “Give me the strongest case against this,” or “Argue the other side as well as its best defender would.” Then evaluate that case as hard as you'd evaluate anyone else's.

As a teacher

This is the part to underline. If people evaluate other people's arguments with suspicion, then lecturing someone about why they're wrong triggers the evaluator against you. Every point you make gets the “Must I believe this?” treatment.

But when people reach a conclusion through their own evaluation — because they worked through a puzzle, argued with peers, or answered a good question — they own it. That's why the card puzzle is so much more convincing when a group solves it than when a teacher explains it. It's also why the most effective critical thinking teaching, as you'll see in Chapter 14, is built on dialogue rather than lectures.

Sources and further reading

• Hugo Mercier & Dan Sperber, The Enigma of Reason (Harvard, 2017); and their “Why do humans reason?” Behavioral and Brain Sciences (2011).
• David Moshman & Molly Geil, “Collaborative reasoning: Evidence for collective rationality,” Thinking & Reasoning (1998).
• Trouche, Johansson, Hall & Mercier, “The selective laziness of reasoning,” Cognitive Science (2016).
• Schkade, Sunstein & Hastie, “What happened on Deliberation Day?” (2007) — the Colorado experiment.
• Irving Janis, Victims of Groupthink (1972).

Chapter 5 · Part II — Reasoning & Evidence

Evidence and Sources

Not all information deserves equal weight, even when it's technically true. This chapter is about judging evidence and sources quickly — the way professionals actually do it, which isn't the way most of us were taught.

In Chapter 1 you learned to count the hops between you and the original evidence. Now we look at the evidence itself, and at the people and pages that deliver it. You can't check everything — so the goal is to get good at checking fast, and at knowing what kind of evidence could actually settle a question.

5.1What kind of evidence is this?

Different kinds of evidence answer different questions. A lot of confusion comes from using one kind to answer a question only another kind can settle.

Anecdotes deserve a special word. They're the most persuasive kind of evidence for humans — vivid, personal, easy to remember — and the weakest for establishing how common something is. A single story can never tell you whether you're looking at the rule or the exception.

The discrimination question

The single most useful question to ask about any piece of evidence is this: would I expect to see this evidence if the claim were false?

A product has hundreds of five-star reviews. Would you expect that if the product were bad? Unfortunately, yes — reviews can be bought or filtered. So the reviews don't discriminate much between “good product” and “bad product with a marketing budget”. Now suppose an independent testing lab bought the product anonymously and it passed. You'd be much less likely to see that if the product were bad. That's strong evidence.

5.2Read sideways, not down

Most of us were taught to evaluate a source by studying it closely: Is it well written? Does it have references? Does it look professional? Is it a .org? What does its About page say?

The problem is that all of these are easy to fake. Anyone can buy a professional design, a .org address and an impressive name. A page can't be trusted to describe itself honestly, any more than a stranger's business card can.

Fact-checkers do something different, which Wineburg and McGrew called lateral reading. Within seconds of landing on an unfamiliar source, they leave it. They open new tabs and search for the organisation or author: Who funds them? What do independent sources — news reports, encyclopedias, other experts — say about them? They also showed what the researchers called click restraint: scanning the search results before clicking, instead of clicking the first link.

5.3SIFT: four moves in a minute

The digital literacy researcher Mike Caulfield turned the fact-checkers' habits into four quick moves, known as SIFT. They're designed to take about a minute, not an afternoon.

5.4Ten forwards, one source

Hearing a claim from many people feels like strong evidence. But evidence only adds up when the sources are independent — when each one checked the facts for itself. Ten people forwarding the same message is not ten sources. It's one source, copied ten times.

The same problem affects professional news. When one outlet reports something and others rewrite that report without checking, you see the claim in ten places. It looks confirmed. It's still one source. This is why “Trace to the original” matters: follow the chain back and see how many links actually touched the evidence.

5.5Who benefits?

Every source has incentives. That doesn't mean everyone with an interest is lying — the manufacturer of a product may know it best. But incentives tell you which errors to expect.

• Sponsored content and “advertorials” look like journalism but are paid for by the subject.
• “Top 10” and “best of” lists often earn a commission on every product linked, which rewards enthusiasm over accuracy.
• Press releases are frequently rewritten as news with little change.
• Reviews can be bought, filtered, or written by the seller's friends — something anyone in hospitality knows well.
• Viral content is rewarded for being shared, not for being true. Outrage and wonder travel furthest.

The question isn't “does this source have an interest?” — almost everyone does. It's “does this interest push them towards a particular answer, and is there independent evidence that doesn't share that push?”

5.6Photos, videos and AI

A photo feels like direct evidence: you're seeing it with your own eyes. But the most common trick isn't a faked image. It's a real image in the wrong context — a photo from a different year, a different country, or a different event, captioned to fit today's story. During floods, riots and elections, old images reliably resurface with new captions.

The fix is the T in SIFT: trace it. A reverse image search (Google Lens or similar) often shows where and when an image first appeared within seconds.

AI adds two new problems. First, fluent text and realistic images are now cheap. Polished writing used to be a weak signal of effort and expertise; now it's no signal at all. AI chatbots can state false things confidently, so check any factual claim from an AI that matters to a decision. Second, there's the liar's dividend: once people know fakes exist, anyone caught on real video can claim it's fake. Both make SIFT more important, not less.

5.7When “do your own research” backfires

This is why SIFT tells you to search the source and look for better coverage, rather than typing the claim itself into a search box. Searching the headline finds people who agree with the headline. Searching “who is [this outlet]” or “[claim] fact check” finds people who evaluated it.

Sources and further reading

• Sam Wineburg & Sarah McGrew, “Lateral Reading and the Nature of Expertise,” Teachers College Record (2019); Stanford Report, “Fact checkers outperform historians when evaluating online information” (2017).
• Mike Caulfield & Sam Wineburg, Verified: How to Think Straight, Get Duped Less, and Make Better Decisions about What to Believe Online (Chicago, 2023).
• Aslett et al., “Online searches to evaluate misinformation can increase its perceived veracity,” Nature (2024).
• Snopes and Factly fact-checks of the “UNESCO best national anthem” claim.

Chapter 6 · Part II — Reasoning & Evidence

Numbers That Lie

Numbers feel objective, which is exactly what makes them persuasive. Most misleading numbers aren't false. They're true numbers with the context removed.

This chapter draws on Calling Bullshit (2020) by Carl Bergstrom and Jevin West, who teach a university course by that name. Their key insight is that you don't need advanced statistics to catch most number tricks. You need a handful of questions, asked every time.

6.1Of what? Relative, absolute, and percentage points

Every percentage is a percentage of something. Change the something, and the same fact sounds bigger or smaller.

• Relative change compares the new number with the old one: from 8 to 9 is a 12.5% increase.
• Absolute change is the plain difference: from 8 in 100 to 9 in 100 is 1 extra case per 100 people.
• Percentage points are the absolute difference between two percentages. If a loan rate goes from 2% to 3%, it rose by one percentage point — which is also a 50% increase.

People who want a number to sound big use relative change: “risk doubles!” (from 1 in 10,000 to 2 in 10,000). People who want it to sound small use absolute change. Both are true. You need both to understand it.

6.2Out of how many? The denominator

A raw count without a denominator tells you almost nothing. Big places have more of everything — more crime, more hospitals, more accidents, more good news.

• “Most road accidents happen within a few kilometres of home.” Of course — most driving happens near home.
• “More people died on the roads in State A than State B.” How many people live there? How many vehicles? How many kilometres are driven?
• “Complaints doubled this month!” From 2 to 4? Out of 50 guests or 5,000?

Rates — per 1,000 people, per 100 guests, per crore rupees — make comparisons fair. But rates can hide scale, too. India's Ministry of Road Transport reported 1,68,491 road deaths in 2022: about 19 people every hour. A per-capita rate can sound small next to other countries while the absolute number is a national emergency. Good reasoning looks at both.

6.3What is the chart hiding?

Charts are arguments in picture form. The same numbers can tell opposite stories depending on the choices made in drawing them.

Common chart tricks:

• A truncated axis that doesn't start at zero, making small differences look huge (fine for line charts tracking change; misleading for bar charts comparing size).
• A cherry-picked time window — start the chart at the lowest point and any trend looks like a recovery.
• Cumulative totals, which can only go up, making a slowing trend look like steady growth.
• Two different y-axes on one chart, which can make any two lines appear to move together.
• 3D effects and pie charts that distort how big the slices look.

6.4Who got counted? Selection and survivors

A number can be perfectly accurate for the people it counted and completely misleading about everyone else. The question is always: how did these cases get into the data, and who's missing?

Survivorship bias is the most famous version. You see the survivors, not the failures, and draw lessons from a skewed sample. College dropouts who became billionaires are famous; the millions of dropouts who didn't are invisible. Investment ads show the funds that did well; funds that did badly were often closed or merged and quietly disappear from the record.

Selection effects are everywhere in business, too. Online reviews come mostly from people who were delighted or furious. A survey of your current guests leaves out everyone who looked at your listing and booked elsewhere — often the people you most need to hear from.

6.5When a number becomes a target

The economist Charles Goodhart gave his name to a principle usually stated like this: when a measure becomes a target, it ceases to be a good measure. People start optimising the number instead of the thing it was meant to track.

You'll see Goodhart's law wherever numbers are rewarded. Staff paid on review scores start pressuring guests for five stars. Schools judged on exam results teach to the test. Call centres measured on call length hang up early. When someone shows you an impressive metric, ask whether anyone was rewarded for moving it.

6.6Big data, confident garbage

Algorithms and AI can make weak data look authoritative. But you rarely need to understand the algorithm to spot the problem. You need to ask where the data came from.

Garbage in, garbage out: an algorithm trained on biased data produces biased results with extra confidence. The question “how was the data collected?” beats any amount of technical detail.

6.7The bullshit asymmetry

A programmer named Alberto Brandolini observed that the energy needed to refute bullshit is an order of magnitude bigger than the energy needed to produce it. A false statistic takes seconds to invent and hours to debunk. That has two lessons:

• You can't debunk everything. Use Belief Triage (Chapter 1) to decide what's worth your time.
• Prevention beats cure. Teaching people the questions in this chapter protects them against thousands of claims you'll never see. That's the logic behind “prebunking”, which you'll meet in Chapter 14.

And one rule of thumb from Bergstrom and West: if a claim seems too good or too bad to be true, it probably is — or at least it needs checking before you share it.

Sources and further reading

• Carl T. Bergstrom & Jevin D. West, Calling Bullshit: The Art of Skepticism in a Data-Driven World (Random House, 2020); callingbullshit.org case study, “Criminal machine learning.”
• IARC Monographs, red and processed meat (2015); UICC, “How to interpret IARC findings on red and processed meat as cancer risk factors.”
• Ministry of Road Transport and Highways, Road Accidents in India 2022.
• Michael G. Vann, “Of Rats, Rice, and Race: The Great Hanoi Rat Massacre” (2003).
• Darrell Huff, How to Lie with Statistics (1954).

Chapter 7 · Part II — Reasoning & Evidence

Cause and Correlation

Two things happening together is the most common evidence in the world, and the most commonly misread. This chapter is about telling “goes with” from “causes” — in the news, in health claims, and in your own business.

Almost every decision rests on a causal belief: if we do X, Y will happen. Raise prices and bookings fall. Run ads and occupancy rises. Take this supplement and feel better. Most of those beliefs come from noticing that two things occurred together. Sometimes that's because one caused the other. Often it isn't.

7.1Four explanations for any correlation

When A and B go together, there are always at least four possibilities:

1. A causes B. Maybe it really is the cause.
2. B causes A. The arrow points the other way — reverse causation.
3. Something else causes both. A third factor, called a confounder, drives both A and B.
4. Chance. Look at enough pairs of things and some will line up by coincidence.

The classic example: ice cream sales and drownings rise and fall together. Ice cream doesn't cause drowning. Hot weather drives both. Obvious when it's ice cream — much harder when it's a claim you want to believe.

7.2Confounders: the hidden third factor

Confounders are the most common reason correlations mislead, because the people who choose to do something are usually different from those who don't, in many ways at once.

• Children with more books at home do better at school. Is it the books, or the parents' education, income and attention that also bring the books?
• Students with private tuition score higher. Is it the tuition, or families that can afford tuition? (And here's a twist: weaker students are also more likely to be sent for tuition, which pushes the other way. Confounders can hide effects as well as invent them.)

7.3Reverse causation

Sometimes the arrow runs the other way. People who use walking sticks fall more often — but the sticks aren't causing falls. Properties that spend the most on advertising sometimes have lower occupancy — because low occupancy is why they're advertising.

A quick test: could B have come first? If you can tell a plausible story in which the “effect” produced the “cause”, you need evidence about timing before you can conclude anything.

7.4Regression to the mean

Whenever a result involves luck — sales in a month, a batter's score, a patient's symptoms, a landing — extreme results tend to be followed by less extreme ones. Not because anything changed, but because the luck that made them extreme doesn't repeat.

Try it. Below, thirty staff members each have a fixed level of skill. Each month, their result is their skill plus some random luck. Pick out the five best and five worst performers in Month 1, and see what happens to those same people in Month 2 — when nothing about them has changed.

Now imagine a manager who gives the bottom five a stern talk after Month 1. Next month, they improve. The talk “worked”. Or someone who introduces a new policy in the worst month in three years; the next month is better, and the policy gets the credit. Many treatments for things that come and go, like back pain or colds, seem to work partly because people seek treatment when they feel worst — the point from which they'd usually improve anyway.

7.5Why randomising works

The cleanest way to find out whether A causes B is to decide who gets A by chance. Flip a coin: heads, you get the treatment; tails, you don't. Because a coin doesn't care about your income, health, mood or motivation, the two groups end up similar in every way — including ways nobody thought to measure. If the groups then differ in outcome, the treatment is the most likely reason.

That's a randomised controlled trial. It's how medicines are tested, and increasingly how social programmes are, too. In 2019, Abhijit Banerjee, Esther Duflo and Michael Kremer won the Nobel Prize in economics for their experimental approach to fighting poverty, much of it tested through randomised trials in India and elsewhere.

Businesses can do the same thing at small scale. Show half your website visitors one version of a page and half another, chosen at random. Try a new check-in process at randomly chosen properties or on randomly chosen days. It's called an A/B test, and it beats arguing about what “obviously” works.

Randomisation has limits. You can't ethically randomise people to smoke. A trial in one population may not apply to another. Small trials can be fooled by chance. But when it's possible, it answers causal questions that no amount of observational data can.

7.6When you can't randomise

In 1965, Bradford Hill listed the kinds of evidence that make a causal link more convincing when you can't experiment. Paraphrased:

• Strength — a big association is harder to explain away than a tiny one.
• Consistency — it appears in different studies, places and populations.
• Timing — the cause comes before the effect.
• Dose-response — more of the cause, more of the effect.
• Plausibility — there's a believable mechanism.
• Experiment — removing the cause reduces the effect.

Researchers also look for natural experiments: situations where chance or an arbitrary rule assigned people to different conditions — a policy that applied only to people born after a certain date, or a lottery for school places.

Sources and further reading

• Daniel Kahneman, Thinking, Fast and Slow (2011), chapter 17 — the flight instructors.
• Writing Group for the Women's Health Initiative Investigators, JAMA (2002); later reappraisals of the WHI hormone trial.
• Doll & Hill, British doctors study (from 1951); Austin Bradford Hill, “The Environment and Disease: Association or Causation?” (1965).
• The Nobel Prize in Economic Sciences 2019 — Banerjee, Duflo and Kremer.
• Judea Pearl & Dana Mackenzie, The Book of Why (2018).

Chapter 8 · Part III — Uncertainty

Thinking in Probabilities

Most of life isn't true or false. It's more likely or less likely. This chapter replaces yes-or-no thinking with a few simple habits for reasoning about chance — no formulas required.

Part III is about uncertainty: how to think clearly when you can't be sure, which is almost always. You've already started. In Chapter 2 you put a number on your confidence. In Chapter 5 you asked whether evidence would be surprising if a claim were false. This chapter joins those ideas into one way of thinking.

8.1Say it in numbers

Words like “likely”, “possible”, “rare” and “almost certain” feel precise, but different people mean very different things by them.

Kent spent years arguing that analysts should attach numbers to their judgements. The same applies to you. “The new property will probably be profitable” hides whether you mean 55% or 90% — and those call for very different decisions. Saying a number feels uncomfortable because it can be wrong. That's the point: a number can be checked, so you can learn.

8.2Start with the base rate

The disease question fooled people because they jumped straight to the evidence (a positive test) and forgot to ask the first question: before any evidence, how common is this? That's the base rate. When something is rare, even good evidence produces lots of false alarms.

Base rates are everywhere once you look for them:

• A founder is sure their new restaurant will succeed. How many new restaurants in that area are still open after three years?
• A quiet, bookish person who loves poetry — more likely to be a librarian or to work in a bank? The description fits a librarian, but there are vastly more bank employees than librarians, so a bank job may well be more likely.
• A security system flags a guest as suspicious. How often are flagged guests actually a problem?

8.3Updating: Bayes without the maths

The Reverend Thomas Bayes, an 18th-century English minister, gave his name to the rule for how evidence should change your beliefs. You don't need the formula. You need three steps:

1. Start with a prior — your best estimate before the new evidence, usually anchored on a base rate.
2. Weigh the evidence — how much more likely is this evidence if the claim is true than if it's false? That's the discrimination question from Chapter 5, turned into a ratio.
3. Update in proportion — strong evidence moves you a lot, weak evidence a little. Neither should take you straight to 0% or 100%.

A tragic example: confusing the two questions

This error — confusing “how likely is this evidence if she's innocent?” with “how likely is she innocent, given this evidence?” — is called the prosecutor's fallacy. It's the same mistake as the disease test: a 5% false-alarm rate doesn't mean a 95% chance you're ill.

8.4Expected value — and the risk of ruin

Probabilities matter for decisions because outcomes have different sizes. The expected value of a choice is what you'd get on average if you could make it many times: each outcome's value multiplied by its probability, added up.

8.5Chance has no memory

In 1913, at the Monte Carlo Casino, the roulette ball reportedly landed on black 26 times in a row. Gamblers lost fortunes betting on red, convinced it was “due”. But a roulette wheel has no memory: each spin is independent. This is the gambler's fallacy.

The same mistake appears whenever people expect random things to “even out” in the short run: a toss that's been lost five times in a row, a month that “must” be better because the last three were bad. If the events are genuinely independent, the next one doesn't know about the others. (If they're not independent — if a bad run reflects a real change, like a new competitor — then the streak is information, and you should update on it.)

Sources and further reading

• Gerd Gigerenzer, Calculated Risks / Reckoning with Risk (2002); Hoffrage & Gigerenzer on natural frequencies in medical diagnosis (1998).
• Sherman Kent, “Words of Estimative Probability,” Studies in Intelligence (1964).
• Royal Statistical Society statement on the Sally Clark case (2001); R v Clark, Court of Appeal (2003).
• Tim Harford, The Data Detective (2021).

Chapter 9 · Part III — Uncertainty

Calibration and Forecasting

Predicting the future is hard, and most experts are worse at it than they sound. But a small group of ordinary people turned out to be remarkably good — and what they do can be learned.

Every plan is a forecast. Opening a property, hiring someone, launching a course: each assumes something about what will happen. This chapter follows the political scientist Philip Tetlock, who spent decades measuring who forecasts well and why, told in Superforecasting (2015, with Dan Gardner).

9.1Foxes and hedgehogs

Tetlock borrowed an ancient line made famous by the philosopher Isaiah Berlin: “The fox knows many things, but the hedgehog knows one big thing.”

• Hedgehogs organise their thinking around one big idea — a theory, an ideology, a favourite explanation — and fit everything into it. They speak with confidence, use words like “certainly” and “impossible”, and rarely change their minds.
• Foxes draw on many ideas, are comfortable with doubt and complexity, and adjust their views when facts change. They say “however” and “on the other hand” a lot.

The foxes forecast better. And Tetlock found an uncomfortable twist: the experts most in demand by the media tended to be hedgehogs. A confident, simple story makes better television than “it's about 60–40, and here's what would change my mind.” The qualities that make someone sound like an expert can be the opposite of the ones that make them right.

9.2The superforecasters

9.3What superforecasters do

Tetlock summarised the habits he saw. Paraphrased and condensed:

• Pin the question. Vague questions can't be scored. “Will the business do well?” becomes “Will monthly revenue exceed ₹15 lakh by March?”
• Break it into parts. A hard question becomes several easier ones.
• Outside view first. Start with the base rate for things like this, then adjust for what's special about this case.
• Update often, in small steps. React to new information, but don't lurch.
• Use fine-grained numbers. Superforecasters distinguish 60% from 65%, and that precision proved meaningful.
• Look for the forces pushing both ways. What makes it more likely? Less likely?
• Keep score, and study your misses — including the ones you got right for the wrong reasons.

9.4The outside view

The team's own estimate was the inside view: imagining the specific project and how it would go. Fox's answer was the outside view: what happened to similar projects. The inside view feels more relevant, because it's about your case. But it misses all the unexpected delays that reliably happen to projects in general, which is why projects so often run late and over budget. Psychologists call this the planning fallacy.

The fix is to start from the outside view, then adjust. How long do renovations like this usually take? What share of new properties in this area break even in year one? Your specific reasons can move you from the base rate — but they should move you from there, not replace it.

9.5Fermi estimation: guessing well

The physicist Enrico Fermi was famous for making surprisingly good estimates of things nobody could look up, by breaking them into parts he could roughly guess. The individual guesses are rough, but their errors tend to partly cancel out — and, more importantly, the method shows you exactly which assumptions matter.

9.6Keeping score

Forecasters get better only if their forecasts are scored. The standard measure is the Brier score: for each forecast, take the gap between the probability you gave and what happened (1 if it happened, 0 if not), square it, and average over all your forecasts.

• 0 is perfect: 100% on everything that happened, 0% on everything that didn't.
• 0.25 is what you'd get by saying 50% about everything.
• 1 is the worst: fully confident and always wrong.

The Brier score rewards two things at once: being calibrated (your 70% predictions come true about 70% of the time) and being decisive (saying 90% or 10% when the evidence justifies it, rather than hiding at 50%).

Your Thinking Journal already keeps score for you: every Predict question you lock in and every journal prediction you resolve feeds the calibration table. The more you use it, the more it tells you.

Sources and further reading

• Philip E. Tetlock, Expert Political Judgment (Princeton, 2005).
• Philip E. Tetlock & Dan Gardner, Superforecasting: The Art and Science of Prediction (2015).
• Mellers et al., “Psychological strategies for winning a geopolitical forecasting tournament,” Psychological Science (2014); Good Judgment Inc., reports on superforecaster performance.
• Daniel Kahneman, Thinking, Fast and Slow (2011), chapter 23 — the curriculum story and the outside view.

Chapter 10 · Part III — Uncertainty

The Mind's Shortcuts

Your brain uses shortcuts to make fast judgements. Usually they work. In predictable situations they misfire — and knowing their names, it turns out, barely protects you. Here's what does.

This is the chapter most critical thinking courses start with: the list of cognitive biases. It comes late in this book on purpose. On its own, a list of biases mostly teaches people to spot them in others. You now have the tools to use it better — and to be honest about which famous findings held up and which didn't.

10.1Shortcuts, not stupidity

Kahneman popularised a useful picture of two modes of thinking. System 1 is fast, automatic and effortless — recognising a face, reading a mood, getting a feel for a price. System 2 is slow, deliberate and effortful — working out 17 × 24, checking an argument. Treat these as a metaphor for two styles of thinking, not two parts of the brain.

Most of the time, System 1's shortcuts — heuristics — work brilliantly. The psychologist Gerd Gigerenzer has shown that simple rules of thumb often beat complex analysis in real, messy environments. You couldn't get through a day without them.

A bias is what happens when a shortcut is used in a situation it wasn't built for, and misfires in a consistent direction. The goal isn't to stop using shortcuts. It's to recognise the situations where they reliably fail.

10.2A field guide

Here are the shortcuts that matter most in daily life and business, with a countermeasure for each.

10.3Why knowing about biases barely helps

You might hope that learning this list makes you less biased. The evidence says: not much, on its own.

In 2002, Emily Pronin, Daniel Lin and Lee Ross described the bias blind spot: people readily see biases in others, but rate themselves as less biased than average. We judge others by their behaviour, but ourselves by our intentions — and our intentions always feel reasonable. A later study by Richard West, Russell Meserve and Keith Stanovich found that people with stronger reasoning skills didn't have a smaller blind spot; if anything, it was slightly larger.

This fits everything in this book so far. Biases work below awareness (Chapter 2). We're sharp at evaluating other people's reasoning and lazy about our own (Chapter 4). A list of biases, learned alone, mostly becomes a new weapon for criticising others.

10.4Which findings survived?

Honesty requires a detour. Some of the most famous findings in popular psychology books have not held up.

That last detail is a model of scout mindset from one of the field's most famous figures. And the broad picture is reassuring in a different way: anchoring, framing, availability and loss aversion have been widely replicated, though researchers still debate how large some effects are and when they apply. The lesson for you: treat any single striking study as a lead, not a law — especially if it appears in a bestseller. Chapter 11 shows you how to read studies with this in mind.

10.5What actually helps

If knowing isn't enough, what works? The evidence points to changing the situation and the process rather than relying on willpower in the moment.

• Checklists. In a 2009 study across eight hospitals in different countries, introducing a simple surgical safety checklist was followed by a fall in deaths and complications. Checklists don't make surgeons less biased; they make it harder for bias to matter.
• Write it down first. Predictions and estimates recorded before the outcome defeat hindsight bias and make calibration possible.
• Consider the opposite. A 1984 study by Charles Lord, Mark Lepper and Elizabeth Preston found that asking people to consider how they'd judge the evidence if it had pointed the other way reduced biased thinking. (It's the selective sceptic test from Chapter 2.)
• Practice with feedback. In 2015, Carey Morewedge and colleagues found that a single training session reduced several biases. An interactive game that gave personal feedback worked better than an instructional video — cutting bias by roughly a third immediately, with much of the benefit still there two to three months later.
• Other people. Structured disagreement (Chapter 4) catches what introspection can't.

Sources and further reading

• Tversky & Kahneman, “Judgment under Uncertainty: Heuristics and Biases,” Science (1974).
• McNeil, Pauker, Sox & Tversky, “On the elicitation of preferences for alternative therapies,” NEJM (1982).
• Pronin, Lin & Ross, “The Bias Blind Spot,” PSPB (2002); West, Meserve & Stanovich, “Cognitive sophistication does not attenuate the bias blind spot,” JPSP (2012).
• Open Science Collaboration, “Estimating the reproducibility of psychological science,” Science (2015); Hagger et al., ego-depletion registered replication (2016); Doyen et al., priming replication (2012); Kahneman's comments (2012, 2017).
• Haynes et al., surgical safety checklist, NEJM (2009); Lord, Lepper & Preston (1984); Morewedge et al., “Debiasing Decisions” (2015).
• Gerd Gigerenzer, Gut Feelings (2007); Stuart Ritchie, Science Fictions (2020).

Chapter 11 · Part IV — Judgment & Action

Science as a Way of Knowing

Science isn't a pile of facts or a group of people in lab coats. It's a set of habits for catching our own mistakes — slow, imperfect, and still the best error-correcting system humans have built.

Part IV turns from reasoning to action: how knowledge gets made, whom to trust, and how to decide. We start with science, because it's the method this whole book has been borrowing from — prediction, measurement, testing, updating — organised on a large scale.

11.1What makes science different

The physicist Richard Feynman put the heart of it in one line, in a 1974 speech to graduating students: “The first principle is that you must not fool yourself — and you are the easiest person to fool.”

Science is less a body of facts than a set of procedures designed to stop people fooling themselves — and each other. Individual scientists are as prone to soldier mindset as anyone (Chapter 2). What's different is the system: claims are made public, exposed to criticism, tested by people who'd love to prove them wrong, and revised.

Risky predictions

The philosopher Karl Popper argued that what makes a theory scientific is that it could be proven wrong — it forbids certain things from happening. A theory that fits every possible outcome tells you nothing.

His favourite example was Einstein. General relativity predicted that the Sun's gravity would bend starlight by a specific amount. In 1919, expeditions led by Arthur Eddington photographed stars during a solar eclipse to check. If the light hadn't bent as predicted, the theory would have been in serious trouble. It passed — and it was the risk of failing that made passing meaningful.

The point isn't to mock anyone. It's that a clear, fair test with a prediction that could have succeeded tells you far more than thousands of satisfied customers.

11.2The toolkit against fooling ourselves

Each standard part of scientific method exists to block a specific way of fooling yourself:

11.3Significant isn't the same as important

News stories often report that a study found a “significant” effect. In statistics, significant means roughly: this result would be surprising if there were no real effect at all. It does not mean the effect is large, important, or even certainly real.

• With a big enough sample, tiny effects become “significant”. A supplement that raises a 100-point memory score by half a point can be statistically significant and practically useless. Always ask how big the effect was.
• If researchers test many outcomes or try many analyses, some will look significant by chance. Statisticians call this the garden of forking paths, and deliberately fishing for significance is called p-hacking. Pre-registration exists to stop it.
• One significant study is a lead. Several independent ones pointing the same way are evidence.

11.4How to read a study without a PhD

You don't need to understand the statistics to ask good questions. Most misleading science news fails on the simplest ones.

11.5Consensus — and when it was wrong

When most experts in a field agree after reviewing the evidence, that consensus deserves heavy weight. You'll almost never be in a position to outthink an entire field on its own subject.

But consensus has been wrong, and it's worth knowing how it got corrected.

• In 1912, Alfred Wegener proposed that continents drift. Most geologists rejected it, partly because he had no convincing mechanism. By the 1960s, new evidence from the ocean floor led to plate tectonics, and the field changed its mind.
• In the early 1980s, Barry Marshall and Robin Warren argued that most stomach ulcers were caused by a bacterium, not stress or acid. Many doctors were sceptical. Marshall famously drank a culture of the bacterium and developed gastritis. The evidence accumulated, antibiotics became standard treatment, and in 2005 they won the Nobel Prize.

Notice how these corrections happened: through better evidence, presented within science's own channels, which eventually persuaded the field. Not through popularity, and not by the outsiders being loud.

That matters because of a common argument: “They laughed at Galileo, and they're laughing at me.” Being rejected doesn't make someone right. For every Wegener, there were many rejected ideas that were simply wrong. What made Wegener and Marshall different was evidence that others could check.

11.6Tradition as hypothesis, science as test

Chapter 1 argued that traditions can carry real knowledge. Science is how you find out which ones do.

This is the healthy relationship between tradition and science. Tradition supplied a hypothesis worth testing. Science supplied the test — and, just as importantly, sorted this remedy from the many traditional remedies that didn't work. Neither “it's ancient, so it's true” nor “it's traditional, so it's nonsense” would have found artemisinin.

Warning signs of pseudoscience

• Claims that can't fail — every outcome is explained after the fact.
• Testimonials instead of controlled tests.
• No change in the claims over decades, whatever new evidence appears.
• Science-sounding words used as decoration: “quantum”, “energy”, “detox”, “frequency”.
• Claims of suppression: “doctors don't want you to know”.
• Something for sale, with the seller as the main source of evidence.

Sources and further reading

• On Semmelweis: “The Semmelweis Effect,” Significance magazine; Semmelweis, The Etiology, Concept and Prophylaxis of Childbed Fever (1861).
• Richard Feynman, “Cargo Cult Science,” Caltech commencement address (1974); Karl Popper, Conjectures and Refutations (1963).
• Narlikar et al., “A statistical test of astrology,” Current Science (2009); “An Indian Test of Indian Astrology,” Skeptical Inquirer (2013).
• The Nobel Prize in Physiology or Medicine 2005 (Marshall & Warren) and 2015 (Tu Youyou).
• Carl Sagan, The Demon-Haunted World (1995); Ben Goldacre, Bad Science (2008).

Chapter 12 · Part IV — Judgment & Action

Whom to Trust

Chapter 1 ended with a puzzle: you can't check everything, so almost everything you know comes from trusting someone. This chapter is the answer — how to choose whom to trust, on what, and how much.

Critical thinking is often presented as doubting everything. That's impossible, and Chapter 1 showed why it would be a disaster: humans thrive by learning from others. The real skill is calibrated trust — trusting the right people about the right things, to the right degree.

12.1We're vigilant, not gullible

If people were simply gullible, the fix for misinformation would be to make them more sceptical. Mercier argues that the bigger problem is often the reverse: people are sceptical of the wrong things, and credulous about claims that fit what they already believe or want.

Remember the lynching rumours in Chapter 1. People didn't believe them because they'd believe anything. They believed them because the messages arrived from trusted people, matched existing fears about strangers, and asked them to protect children — something they already cared about. The trust was misplaced, but it wasn't random.

Mercier describes several checks we naturally use when someone tells us something:

• Plausibility — does it fit what I already know?
• Arguments — are the reasons they give good? (Chapter 4: we're good at judging others' arguments.)
• Competence — are they in a position to know?
• Benevolence — are their interests aligned with mine, or might they want to mislead me?
• Accountability — would they pay a price, in reputation, if they turned out to be wrong?

These are good instincts. They go wrong in predictable ways — especially when we mix up two of them.

12.2Kind isn't the same as knowledgeable

The most common trust mistake is treating benevolence as if it were competence. Your uncle loves you and would never try to mislead you. That makes him trustworthy about his intentions. It doesn't make him knowledgeable about vaccines, share prices or constitutional law.

Family WhatsApp groups run on benevolence: messages are trusted because of who sent them, not because the sender knows the subject. A forward from a cousin feels safer than the same text from a stranger — even though the cousin is usually just passing along something from a stranger.

12.3When to trust expert intuition

Not all expertise is equal. Some experts have gut feelings you should trust; others have confidence you shouldn't.

In 2009, Daniel Kahneman — a sceptic of expert intuition — and Gary Klein — who had spent his career studying how firefighters and nurses make fast, accurate calls — wrote a joint paper about where they agreed. Their conclusion: intuitive expertise is real when two conditions hold.

1. The environment is regular enough to be predictable — patterns really do repeat.
2. The expert has had lots of practice with quick, clear feedback about whether they were right.

Firefighters, chess players, experienced nurses, and a front-desk manager who has watched thousands of guests check in: their intuitions are trained by constant feedback. Long-range political pundits and people picking individual stocks work in environments that are either too unpredictable or give feedback too slowly and noisily. They may feel just as confident — confidence is not the signal (Chapter 9).

12.4Follow the incentives

The novelist Upton Sinclair put it memorably in 1935: “It is difficult to get a man to understand something, when his salary depends upon his not understanding it.”

This doesn't mean anyone with an interest is lying. It means asking: does this source's incentive push towards a particular answer — and do independent experts, without that incentive, agree? A drug company's trial is more credible when independent researchers get the same result. An industry-funded study that contradicts every independent study deserves extra scrutiny.

12.5When experts disagree

Experts disagree all the time, and that confuses people into thinking nobody knows anything. Some questions help:

• Is the disagreement about the core or the details? Scientists argue about how much and how fast, while agreeing on whether. News coverage often blurs the two.
• Is it one maverick against a field, or a field genuinely split? A lone dissenter is sometimes right (Chapter 11), but the base rate favours the field.
• Are the experts independent? Twenty experts all funded by the same interest are closer to one voice than twenty.
• What do bodies that review all the evidence say? Systematic reviews and national academies exist precisely to weigh disagreement.

12.6“Do your own research” — carefully

In Chapter 5 you met the 2024 finding that searching online to check false news made people more likely to believe it. Doing “your own research” on a technical question usually means searching for the claim, finding pages that repeat it, and feeling confirmed.

A better version is: research the sources, not the subject. You probably can't evaluate the chemistry of a vaccine or the economics of a policy. You can evaluate who's making the claim, what their track record and incentives are, and whether independent experts agree. That's the research that's actually within reach.

Sources and further reading

• Joshua Kalla & David Broockman, “The Minimal Persuasive Effects of Campaign Contact in General Elections: Evidence from 49 Field Experiments,” American Political Science Review (2018).
• Hugo Mercier, Not Born Yesterday: The Science of Who We Trust and What We Believe (Princeton, 2020).
• Daniel Kahneman & Gary Klein, “Conditions for intuitive expertise: A failure to disagree,” American Psychologist (2009).
• Upton Sinclair, I, Candidate for Governor: And How I Got Licked (1935); Naomi Oreskes & Erik Conway, Merchants of Doubt (2010).

Chapter 13 · Part IV — Judgment & Action

Deciding Under Uncertainty

Everything in this book ends up here: a choice, made without knowing how it will turn out. Good decisions don't guarantee good outcomes — but good decision habits, repeated, tilt the odds.

You've learned to pin claims, weigh evidence, think in probabilities and catch your own biases. A decision combines all of these with one more ingredient: values — what you actually want. This chapter gives you a practical routine for decisions that matter.

13.1What a decision is made of

Every decision has three parts, and confusion usually comes from mixing them:

• Options — what you could do.
• Beliefs — what you think will happen if you do each (predictions, with probabilities).
• Values — how much you care about each possible outcome.

Chapter 3 showed how two partners can argue for an hour because one is making a prediction and the other stating a value. Separating the three lets you argue about each in the right way: evidence for beliefs, honest conversation for values, and creativity for options.

13.2Don't judge decisions by results alone

The former professional poker player Annie Duke calls it resulting: judging the quality of a decision by how it turned out. It's natural, and it's one of the main reasons people learn the wrong lessons from experience.

Whether the decision was good depends on what Dhoni knew and what his alternatives were at that moment — not on where the ball landed. Outcomes mix decision quality with luck. Duke's four boxes make this explicit:

13.3Widen your options

Many bad decisions are made between too few options — often just one: “Should we do this, yes or no?” The management researcher Paul Nutt studied hundreds of decisions in organisations and found that about half failed. Failures were linked to a narrow search for alternatives: decision-makers seized on one idea early and spent their energy defending it.

Chip and Dan Heath's book Decisive suggests two quick fixes:

• The vanishing options test. Imagine none of your current options were possible. What would you do instead? This forces new ideas onto the table.
• “And”, not “or”. Could you do more than one, at smaller scale? Test two marketing channels at once instead of betting on one.

13.4One-way and two-way doors

In his 2015 letter to shareholders, Amazon's founder Jeff Bezos distinguished two kinds of decision. Some are one-way doors: hard or impossible to reverse, so they deserve slow, careful thought. Most are two-way doors: if you don't like what's on the other side, you can walk back through. Those should be made quickly, by the people closest to them, and adjusted as you learn.

Signing a ten-year lease, taking on large debt, or firing a senior person are one-way doors. Trying a new menu, a new price for a month, or a new booking channel are usually two-way doors. A common mistake is treating two-way doors with one-way caution — and endlessly debating things that a small experiment (Chapter 7) would settle faster.

13.5Putting options side by side

For decisions with several options and several things you care about, a simple decision matrix helps. List options as columns and criteria as rows, give each criterion a weight for how much it matters, and score each option.

A warning: the matrix is a tool for thinking and discussing, not a machine that produces the answer. The weights are your values, stated honestly. The scores are predictions, which could be wrong. If the result surprises you, don't simply obey it — ask which weight or score you disagree with. That's where the real decision is.

13.6The pre-mortem

A post-mortem asks why something died. A pre-mortem, Gary Klein's technique, asks the same question before you start, while you can still change course.

13.7Keep a decision journal

Because of hindsight bias (Chapter 10), you won't remember what you knew or expected when you made a decision. So write it down at the time: the options you considered, what you expected to happen and how confident you were, the key reasons, and how you felt. Months later, compare with what actually happened.

The point isn't to feel good or bad about outcomes. It's to learn which parts of your process are reliable. Your Thinking Journal has a “Decision” entry type for exactly this.

Sources and further reading

• Mitchell, Russo & Pennington, “Back to the future: Temporal perspective in the explanation of events,” Journal of Behavioral Decision Making (1989); Gary Klein, “Performing a Project Premortem,” Harvard Business Review (2007).
• Annie Duke, Thinking in Bets (2018) and How to Decide (2020).
• Paul C. Nutt, “Surprising but true: Half the decisions in organizations fail,” Academy of Management Executive (1999); Chip & Dan Heath, Decisive (2013).
• Jeff Bezos, Letter to Amazon shareholders (2015).

Chapter 14 · Part V — Teaching

Teaching Critical Thinking

If you want to pass these ideas on, start here. The research is clear on one thing: telling people to think critically almost never works. Here's what does, and how to design your first session.

Everything in this book so far has been preparation for this chapter. You now know how reasoning works (Chapter 4), why people resist (Chapter 2), whom they trust (Chapter 12), and why knowing about biases isn't enough (Chapter 10). Each of those findings turns out to be a teaching principle.

14.1What the research says works

In 2015, Philip Abrami and colleagues published a large review of studies on teaching critical thinking: 341 effect sizes from many classrooms. The average effect was positive but modest (an effect size of about 0.3). Three instructional ingredients stood out:

• Dialogue — discussion, questioning and debate, not just listening.
• Authentic problems — real, engaging problems relevant to the learners' lives (the researchers call this “anchored instruction”).
• Mentoring — one-to-one guidance and feedback from someone more experienced.

They worked best in combination. That's also a fair description of a good workshop: people wrestling with real problems, arguing about them, with someone guiding and giving feedback.

Why transfer is hard

The psychologist Daniel Willingham argued in a widely read 2007 article that critical thinking isn't a general skill like riding a bicycle, which, once learned, works everywhere. People who reason brilliantly about cricket statistics can reason badly about health claims. Thinking well depends on knowing the domain — and on recognising that a new problem has the same deep structure as one you've seen before. That recognition is what usually fails.

So teaching for transfer means:

• Name the tool explicitly — “this is a denominator problem”, “this is regression to the mean”. A named pattern is easier to spot again.
• Use many examples of the same idea in different settings — health, business, cricket, family, news.
• Put two examples side by side and ask what they share. Research by Dedre Gentner and colleagues found that comparing two cases helps people extract the underlying principle better than studying the cases one at a time.
• Practise over time. One session rarely sticks; four spaced sessions did in Bihar.

14.2Don't make it a fight

You might fear that correcting people makes them dig in — the so-called backfire effect. The good news: in large studies, genuine backfire is rare. Thomas Wood and Ethan Porter tested corrections on 52 issues with over 10,000 participants and found that, on average, people moved towards the facts, even on politically charged topics — though often modestly.

The bad news: people move far less when a belief is tied to their identity, and a one-off correction rarely changes deeper attitudes. What does?

The pattern matches everything in this book. People change their minds through their own reasoning (Chapter 4), when they don't feel attacked (Chapter 2), in conversation with someone who has shown they understand (the Ideological Turing Test, Chapter 2).

14.3Prebunking: vaccinate, don't just cure

Remember Brandolini's law (Chapter 6): refuting nonsense takes far more effort than producing it. You can't debunk every false claim. But you can teach people to recognise the techniques behind them.

The psychologists Jon Roozenbeek and Sander van der Linden call this inoculation or prebunking: exposing people to a weakened dose of a manipulation technique, and showing how it works, so they resist it later. Their online game Bad News puts players in the role of a fake-news producer, using techniques like impersonation, emotional language, polarisation and conspiracy. Playing it improved people's ability to spot those techniques. In a 2022 study, short prebunking videos about manipulation techniques, shown as YouTube ads in a large field test, also improved people's recognition of those techniques.

For your teaching, the lesson is to teach patterns rather than individual claims: “Here's how an emotional hook works. Here's what a false dilemma sounds like. Here's the ‘doctors don't want you to know’ move.” Then let people find the patterns in real examples themselves.

14.4Teach with questions

The oldest method of teaching critical thinking is also one of the best: Socratic questioning. Instead of telling people what to think, you ask questions that help them examine what they already think. It works because it uses the evaluator in each person on their own ideas — the one thing they usually skip.

14.5Designing your first session

Here's a blueprint that combines everything above. It's built for about 90 minutes with 8–20 people, but it scales down to a conversation with one friend.

Measure whether it worked

Don't rely on people saying they enjoyed it. At the start and end of a course, give participants a short set of the same kinds of tasks — spot the flaw, judge a source, answer a base-rate question — with confidence ratings. Compare accuracy and calibration. Better still, check again a month later. That's how you'll know whether you're changing thinking, and it's the honest way to describe your results to anyone who pays for your teaching.

14.6The ethics of teaching thinking

• Teach methods, not conclusions. The moment a critical thinking course becomes a way to bring people round to the teacher's views, it has become the thing it warns against. Your learners should end up able to disagree with you well.
• Show your own mistakes. Share predictions you got wrong, beliefs you changed. It's the most convincing demonstration of scout mindset, and it makes it safe for others.
• No contempt. Words like “sheep” or “idiots” for people who believe false things are soldier mindset in scout's clothing. Chapter 1 showed why believing what trusted people tell you is normal and often wise.
• Tools are not weapons. Teach people to use these tools on their own beliefs first, and on others' only with care.

14.7Where this leaves you

In Chapter 1, you rewrote the book's opening thesis about the world running on belief. Here's your version, if you saved it:

You've since learned that humans run on trust, not doubt; that reasoning works best between people who disagree; that numbers, causes and studies can mislead in predictable ways; that confidence should be a number, and that the number should be scored; that knowing about biases isn't enough, but processes are; and that people change their minds when they reason their own way there.

That last point is the one to carry into your teaching. You can't think for anyone else. But you can give people good questions, safe disagreement, and a few tools — and then let them do the most human thing there is: learn from each other.

Sources and further reading

• Sumitra Badrinathan, “Educative Interventions to Combat Misinformation: Evidence from a Field Experiment in India,” American Political Science Review (2021); Badrinathan et al., “Countering Misinformation Early: Evidence from a Classroom-Based Field Experiment in India,” APSR (2025).
• Abrami et al., “Strategies for Teaching Students to Think Critically: A Meta-Analysis,” Review of Educational Research (2015).
• Daniel Willingham, “Critical Thinking: Why Is It So Hard to Teach?” American Educator (2007); Gentner, Loewenstein & Thompson, “Learning and transfer: A general role for analogical encoding” (2003).
• Wood & Porter, “The Elusive Backfire Effect,” Political Behavior (2019); Broockman & Kalla, “Durably reducing transphobia,” Science (2016).
• Roozenbeek & van der Linden, “Fake news game confers psychological resistance against online misinformation,” Palgrave Communications (2019); Roozenbeek et al., “Psychological inoculation improves resilience against misinformation on social media,” Science Advances (2022).
• Guess et al., “A digital media literacy intervention increases discernment between mainstream and false news in the United States and India,” PNAS (2020).