2.5 — Two Speeds of Thinking

A bat and a ball cost ₹110 together. The bat costs ₹100 more than the ball. How much does the ball cost?

An answer arrived immediately, and for most people it was ₹10. Check it: if the ball is ₹10 and the bat is ₹100 more, the bat is ₹110, and the total is ₹120. Wrong. The ball is ₹5, the bat is ₹105, and the difference is ₹100.

The interesting thing is not that people get it wrong. It is that the wrong answer arrives with no sense of effort and no sense of doubt. It does not feel like a guess. It feels like reading the answer off the page. Among students at some of the most selective universities in the world, a majority give ₹10.

That gap — between an answer being produced instantly and being correct — is the subject of this page.

The two systems

Psychologists describe this as dual-process theory, and Daniel Kahneman's naming is the one that stuck.

System 1 is fast, automatic, effortless and always running. It reads words you look at, recognises faces, detects hostility in a voice, completes "bread and…", drives a familiar route, and produces answers to questions before you have decided to think about them. You cannot turn it off. It has no sense of its own uncertainty.

System 2 is slow, deliberate, effortful and lazy. It multiplies 17 by 24, compares two flats on five criteria, checks an argument for a flaw, and holds back a reply you want to send. It is capacity-limited by working memory (2.3), it costs glucose and attention, and it will accept System 1's answer without checking whenever it plausibly can.

A worked example of the division. You see a face and instantly know it is angry — System 1, no effort, no choice. You then work out whether the anger is at you or at something that happened before you arrived — System 2, slow, and easily skipped if you are tired, which is precisely why tired people take things personally (5.8).

Two honest caveats before going further. First, these are not two anatomical structures with a wall between them; they are a description of two modes of processing that overlap. Second, some of the specific results in the popular literature on this — particularly the ones about priming — did not survive the replication crisis of 1.3. The core division is solid; some of the famous demonstrations are not.

Why the fast system is right nearly all the time

It is easy to read a list of biases and conclude that human intuition is broken. That would be badly wrong.

System 1 is a set of heuristics — rules that give a good answer cheaply, most of the time, in the conditions they were built for. Judging distance by haziness, judging a stranger's intent from posture and face, judging whether a floor will hold you. These run in milliseconds and are right often enough that the alternative — deliberating about everything — would leave you paralysed. Nobody could cross a road on System 2 alone.

A bias is what you get when a rule that is usually right meets a case it was not built for. The rule is not the error. The mismatch is. This is exactly the Müller-Lyer situation from 2.2, and it is the correct frame for the whole of 11.5.

Two conditions determine whether your intuition can be trusted, and they are worth memorising because they answer the practical question directly.

Is the environment regular enough to contain learnable patterns? Chess positions, yes. Medical presentations in a familiar specialty, yes. Long-run stock prices, no. Predicting which new employee will succeed, barely.

Have you had many repetitions with fast, clear feedback? A firefighter who has entered four hundred burning buildings and found out immediately each time whether their read was right has genuine expert intuition. A manager who makes twelve hiring decisions a year, never learns how the rejected candidates would have done, and gets ambiguous signals about the ones hired, has confidence without calibration.

Expert intuition is real where those two conditions hold, and is indistinguishable from the inside where they do not. That last clause is why it matters. The feeling of knowing is produced identically in both cases.

The four substitutions that produce most errors

System 1's characteristic move is to answer an easier question than the one asked, without noticing the swap.

Asked "how likely is this?", it answers "how easily can I think of an example?" The availability heuristic. Vivid, recent and emotionally charged events come to mind easily, so their frequency is overestimated. Plane crashes over car crashes, violent crime over disease, the dramatic risk over the boring one. This is also why the last three interactions with a person dominate your sense of them.

Asked "how likely is this to be an X?", it answers "how much does it resemble a typical X?" The representativeness heuristic. It ignores how common X is in the first place, which is the base-rate error from 1.4. It also produces the conjunction error: told about a woman who studied philosophy and cared about social justice, most people rate "she is a bank teller and active in a feminist movement" as more likely than "she is a bank teller" — which cannot be true, since every member of the first group is in the second.

Asked for a number, it starts from whatever number is nearby. Anchoring, and it works with numbers that are obviously irrelevant. Asked whether Gandhi died before or after age 140 and then to estimate his actual age at death, people give higher estimates than a group anchored on 30. The practical version is that the first figure named in a negotiation moves the outcome (17.12), and you should form your own number before you hear theirs.

Asked "is this true?", it answers "does this feel easy and familiar?" The fluency effect from 1.5. Clear typefaces, rhyming phrases, repeated statements and confident delivery all increase believed truth.

The affect heuristic, which runs your day

The most consequential of them all, and the one least often named. Asked "what do I think about this?", System 1 answers "how do I feel about this?" — and the feeling then generates reasons.

The evidence for the ordering is uncomfortable. People frequently reach a judgement immediately and produce their justification afterwards, and when the justification is dismantled, the judgement usually stays. In studies of moral reasoning, people confronted with a scenario they find repugnant but cannot argue against will run out of arguments and hold the position anyway — Jonathan Haidt called this moral dumbfounding.

This is not a reason to distrust every judgement. It is a reason to notice a specific pattern in yourself: when you find yourself generating a third reason for a conclusion you reached instantly, the reasons are probably not why you believe it. That is the tell, and it is checkable in real time.

When to override, and how

Deliberate thinking is expensive and mostly unnecessary. The skill is knowing the few situations where the fast answer is likely to be confidently wrong.

Switch to slow when:

The stakes are high and reversal is hard. Job, house, marriage, surgery, anything with a signature.

The situation is unusual for you. No repetitions means no calibration, however strong the feeling.

Numbers, probabilities or percentages are involved. System 1 is bad at these in specific, documented ways, and the errors are not random — they lean predictably.

Someone is creating urgency. As in 1.5. Pressure exists to keep System 2 out of the room, and that is a reliable signal rather than a coincidence.

You feel strongly. Not a reason to distrust the conclusion, but a reason to check whether the feeling arrived first.

And the mechanics of overriding, which is the part usually left out. You cannot override by trying harder — System 1 has already answered and its answer feels like perception. What works is a procedure that forces a second pass:

Slow the input. Read the question again, aloud. In the bat and ball problem, most people who get it right report that something felt off and they re-read. Studies find the error rate drops when the question is printed in a hard-to-read font, because the difficulty itself recruits System 2.

Write it down. Externalising forces sequence and exposes gaps that a fluent internal answer conceals.

Ask the disconfirming question. From 1.8: what would make this wrong? System 1 does not generate this; it must be asked deliberately.

Use a checklist for anything you do repeatedly and cannot afford to get wrong. This is why surgical checklists reduced deaths substantially in large trials, despite surgeons initially finding them insulting. A checklist is not a substitute for expertise. It is a mechanical guarantee that the expert's slow system was engaged on the items that matter.

Sleep on it, where you can. Not mysticism — the mood that produced the fast answer will have changed by morning, and if the conclusion survives a different mood it is more likely to be about the world.

What to do with this

Learn your own two lists. Write down the domains where you have had hundreds of repetitions with clear feedback: those are where your gut is a genuine instrument, and second-guessing it will make you worse. Then write down the domains where you have had a handful of attempts and never really learned the outcome — hiring, investing, predicting how a relationship will go, judging a stranger's character. In the second list your confidence is manufactured, and a slow process beats you every time.

Install one rule with a number in it. Any decision above a threshold you set — an amount of money, an amount of time, anything involving other people's lives — gets written down and left overnight. The rule works precisely because it is mechanical and does not require you to correctly judge, in the moment, whether this is one of the times you are being fooled.

And when you catch the fast answer being wrong, keep the example. Not for self-criticism. The bat-and-ball feeling — the answer that arrived clean and was not true — is the only reliable internal signal you will ever have, and people who can recognise it in themselves make substantially fewer expensive mistakes than people who cannot.

Next: 2.6 asks what the fast system is actually doing when it produces those instant answers, and arrives at a single principle — the brain is not reacting to the world, it is predicting it, and everything you experience is the difference between the prediction and what arrived.