Twenty-two biases,
four problems the brain solves.

The engine asks“If you had never heard that first figure, what would you estimate from scratch?”

The mind latches onto an initial value as a reference point and adjusts away from it insufficiently — even when the anchor is arbitrary or irrelevant. It pulls your estimate whether or not you believe it.

Example A jacket marked “$400, now $250” feels like a steal, even if it is only worth $150.

Counter Form your own estimate before seeing any number; or test an opposite anchor and see where you land.

Tversky & Kahneman (1974), Judgment under Uncertainty: Heuristics and Biases, Science.

The engine asks“Is this actually likely, or just recent and vivid?”

We estimate how likely or frequent something is by how easily examples come to mind. Vivid, recent, or emotional events feel more probable than they are; common-but-dull risks feel rarer.

Example After plane-crash headlines, flying feels dangerous — though the drive to the airport was the riskier leg.

Counter Ask for base rates and real frequencies, not what you can recall.

Tversky & Kahneman (1973), Availability: A Heuristic for Judging Frequency and Probability, Cognitive Psychology.

The engine asks“What evidence would change your mind — and have you gone looking for it?”

We seek, interpret, and remember information that confirms what we already believe, and discount what contradicts it. The search itself is lopsided, so the evidence we gather is too.

Example Convinced a new hire is a star, you read their every move generously and explain away the misses.

Counter Actively hunt disconfirming evidence — search for what would prove you wrong as hard as for what proves you right.

Nickerson (1998), Confirmation Bias: A Ubiquitous Phenomenon in Many Guises, Review of General Psychology.

The engine asks“Who or what is not in your data because they did not survive to be counted?”

We draw conclusions from the cases that “survived” some selection, while the failures — silently filtered out — never enter the data. The graveyard is invisible, so the pattern looks stronger than it is.

Example “Successful founders dropped out of college” ignores the far larger crowd of dropouts who failed and were never studied.

Counter Ask where the missing cases are — who did not survive to be counted — and whether they would overturn the pattern.

Abraham Wald (1943; repr. 1980), A Method of Estimating Plane Vulnerability Based on Damage of Survivors, Statistical Research Group / Center for Naval Analyses.

The engine asks“Does this still hold if you restate it the opposite way?”

Our choices change with how options are described — gains vs. losses, 90% survival vs. 10% mortality — even when the underlying facts are identical. Presentation, not substance, moves the decision.

Example “90% fat-free” sells; “10% fat” does not — same yogurt.

Counter Restate the option the opposite way (gain ↔ loss) and check whether your preference survives the rewording.

Tversky & Kahneman (1981), The Framing of Decisions and the Psychology of Choice, Science.

The engine asks“What is the base rate for cases like this, regardless of how compelling this one feels?”

We judge probability by how closely something matches a mental stereotype, ignoring how common it actually is (the base rate). A vivid match feels like evidence, but it is not.

Example Told someone is “quiet and loves books,” people guess “librarian” over “salesperson” — though salespeople vastly outnumber librarians.

Counter Start from the base rate of the reference class, then adjust for specifics — not the other way around.

Kahneman & Tversky (1973), On the Psychology of Prediction, Psychological Review.

The engine asks“What is missing that, if you had it, could flip this?”

The mind builds a coherent story from the information at hand and treats it as the whole picture, ignoring what is unknown or missing. Confidence comes from the story’s coherence, not its completeness.

Example A glowing one-page résumé makes the candidate feel like a sure thing — it is one page about a whole career.

Counter Ask what you would need to know to be wrong, and what is conspicuously absent from the information you have.

Kahneman (2011), Thinking, Fast and Slow.

The engine asks“Are you rating the whole thing highly because of one standout feature?”

A single positive (or negative) quality — charisma, attractiveness, a prestigious logo — spills over into our rating of unrelated qualities. We infer competence from confidence.

Example A polished speaker is assumed to be a deep thinker; we mistake fluency for substance.

Counter Rate each quality separately, on its own evidence; ask which judgments are riding on one standout trait.

Thorndike (1920), A Constant Error in Psychological Ratings, Journal of Applied Psychology.

The engine asks“Is this a real mechanism, or a story drawn around randomness?”

We turn sequences of events into tidy cause-and-effect stories because stories are easier to hold than randomness. The coherence of the tale gets mistaken for evidence it is true.

Example After a stock jumps, pundits supply a confident reason — the same move would have been “explained” just as fluently had it dropped.

Counter Ask whether there is a real mechanism or just a satisfying story; notice how easily the opposite outcome could have been narrated.

Taleb (2007), The Black Swan.

The engine asks“Would this argument survive if a junior person made it?”

We give outsized weight to authority figures — titles, seniority, credentials — independent of the actual argument. The messenger’s status substitutes for scrutiny of the message.

Example A team adopts a flawed plan because the most senior person proposed it, and no one pressure-tests it.

Counter Judge the argument on its merits; ask whether it would survive if a junior or anonymous person had made it.

Milgram (1963), Behavioral Study of Obedience, Journal of Abnormal and Social Psychology (see also Cialdini, Influence, 1984).

The engine asks“Are you doing this because it is right, or because everyone else is?”

We treat “many others are doing it” as evidence it is correct, especially under uncertainty. The more people seem to agree, the more we assume they cannot all be wrong — though they often are, for the same reason.

Example A packed restaurant draws a longer queue; everyone is in line partly because everyone else is.

Counter Separate popularity from validity; ask whether you would choose this if no one else were.

Cialdini (1984), Influence: How and Why People Agree to Things (see also Asch, 1951).

The engine asks“Honestly, what is the chance you are wrong — and would you bet on it?”

We systematically overestimate the accuracy of our judgments and the precision of our knowledge. Felt confidence is calibrated to how good the story sounds, not to how likely we are to be right.

Example People who say they are “99% sure” are wrong far more than 1% of the time.

Counter State a confidence number and a range you would bet on; ask the base rate of being wrong for calls like this.

Moore & Healy (2008), The Trouble with Overconfidence, Psychological Review.

The engine asks“Are you avoiding this mainly to dodge a loss, not to capture a better outcome?”

Losses loom larger than equivalent gains — roughly twice as painful — so we take irrational risks to avoid losses and pass up good bets that carry any possible loss. The asymmetry warps every risk decision.

Example People refuse a coin-flip to win $150 or lose $100, though the bet is clearly favourable.

Counter Reframe in terms of final outcomes and expected value, not gains vs. losses from today’s reference point.

Kahneman & Tversky (1979), Prospect Theory: An Analysis of Decision under Risk, Econometrica.

The engine asks“Starting fresh today with nothing invested, would you still choose this?”

We continue a course of action to justify past, unrecoverable investments — money, time, effort — instead of weighing only future costs and benefits. The more we have poured in, the harder it is to stop.

Example Sitting through a bad film because you paid for the ticket. The money is gone either way.

Counter Ignore what is spent. Ask only “what is the best use of the next dollar or hour from here?” — as if arriving fresh.

Arkes & Blumer (1985), The Psychology of Sunk Cost, Organizational Behavior and Human Decision Processes.

The engine asks“What did similar efforts actually take — not what you hope this one will?”

We predict our own projects will run as smoothly as the best case, underestimating time, cost, and risk — even knowing similar projects usually overran. Optimism about the specific case overrides the track record.

Example Renovations and software launches routinely take about twice the estimate; the next one is always “different.”

Counter Build the estimate from the actual outcomes of similar past projects (the reference class), not an imagined smooth run.

Buehler, Griffin & Ross (1994); term coined by Kahneman & Tversky (1979).

The engine asks“Are you choosing this because it is best, or because it is the default?”

We prefer things to stay as they are, treating the current state or default as a baseline and any change as a risk. Inaction feels safer than action, even when change is the better bet.

Example People rarely switch the default insurance or retirement-plan option, whatever it is — the default quietly decides for them.

Counter Compare every option (including the status quo) as if choosing fresh today; ask whether you are picking it or just defaulting to it.

Samuelson & Zeckhauser (1988), Status Quo Bias in Decision Making, Journal of Risk and Uncertainty.

The engine asks“What is the realistic downside, and have you actually priced it in?”

We expect our own future to be sunnier than the odds warrant — overestimating good outcomes and underestimating risks, for ourselves more than for others. Hope quietly rewrites the probabilities.

Example Most people think they are less likely than average to get divorced, get sick, or crash a car.

Counter Price the realistic downside explicitly; ask what the outcome distribution looks like for people in your situation.

Weinstein (1980), Unrealistic Optimism About Future Life Events, Journal of Personality and Social Psychology.

The engine asks“Before the outcome, did you truly predict it — or does it just feel obvious now?”

Once we know how something turned out, we believe we predicted it — or could have. The outcome reshapes our memory of what we thought beforehand, making events seem more foreseeable than they were.

Example After a crash, everyone “saw it coming” — though almost no one acted on it at the time.

Counter Recover what you actually predicted beforehand (write predictions down); ask whether the signs were truly obvious before the result was known.

Fischhoff (1975), Hindsight ≠ Foresight, Journal of Experimental Psychology: Human Perception and Performance.

The engine asks“Bad decision, or a reasonable call with a bad outcome — would you make it again knowing only what you knew then?”

We rate the quality of a decision by its result rather than the reasoning available at the time. Because luck shapes outcomes, good decisions sometimes fail and bad ones succeed — so judging by result alone trains the wrong lessons.

Example A surgeon makes the textbook-correct call; a rare complication kills the patient and it is branded a blunder.

Counter Separate decision-quality from outcome-quality. Ask “given only what was known then, was this reasonable?”

Baron & Hershey (1988), Outcome Bias in Decision Evaluation, Journal of Personality and Social Psychology.

The engine asks“If this had gone the other way, would you be assigning the cause the same way?”

We attribute successes to our own ability and effort, and failures to bad luck, circumstances, or other people. The asymmetry protects self-image but corrupts the lessons we draw.

Example “I aced it because I am sharp; I bombed it because the questions were unfair.”

Counter Attribute symmetrically — apply the same standard to wins and losses; ask how you would assign cause if the result had flipped.

Miller & Ross (1975), Self-Serving Biases in the Attribution of Causality: Fact or Fiction?, Psychological Bulletin.

The engine asks“Are recent events drowning out the longer track record?”

We give disproportionate weight to the most recent information, letting it overshadow a longer track record. What just happened feels like the trend.

Example One great (or terrible) recent quarter reshapes a strategy built on years of contrary data.

Counter Zoom out to the full history; ask whether recent events are a real shift or just the latest data point.

Murdock (1962), The Serial Position Effect of Free Recall, Journal of Experimental Psychology.

The engine asks“Are you judging the whole option by one high point or how it ended?”

We remember and evaluate an experience largely by its most intense moment and how it ended — not by the sum or average of the whole. Duration barely registers.

Example A two-week trip that ended on a sour final day is remembered as worse than a shorter one that ended well.

Counter Reconstruct the whole experience, not just its peak and finish; ask whether one moment or the ending is standing in for the entirety.

Kahneman, Fredrickson, Schreiber & Redelmeier (1993); Redelmeier & Kahneman (1996).