AI Study Notebook AI-generated
Rationality: What It Is, Why It Seems Scarce, Why It Matters
Steven Pinker
On this page
Rationality: What It Is, Why It Seems Scarce, Why It Matters — Chapter-by-Chapter Outline
Author: Steven Pinker First published: 2021 (Viking / Penguin Random House, US; Allen Lane / Penguin, UK) Edition covered: First edition, 2021 (432 pp.; US ISBN 978-0-525-56199-6 hardcover, 978-0-525-56201-6 paperback; UK ISBN 978-0-241-38027-7). No revised edition with added or removed chapters has been published as of this writing.
Central thesis
Rationality is not a luxury or an academic abstraction — it is the most potent tool humans possess for improving their own lives and the condition of the world. Pinker argues that humans are neither the reliably rational agents classical economics assumed nor the hopelessly biased creatures that a misreading of behavioral science implies. The truth is more interesting: people reason very well in the ecological niches they have inhabited for millennia (tracking game, detecting cheaters, reading social intent) and very poorly in the formal, decontextualized settings that modern life and modern institutions require (assessing base rates, updating beliefs on statistical evidence, cooperating with strangers on abstract problems).
The book's project is therefore two-part. First, it serves as a practical introduction to the canonical tools of rational thought — logic, probability, Bayesian inference, expected-utility theory, signal detection theory, game theory, causal inference — showing readers what each tool is, what errors it prevents, and how it applies to everyday decisions. Second, it confronts the puzzle of why these tools seem so rarely used in public life: the persistence of conspiracy theories, pseudoscience, motivated reasoning, and tribal epistemology. The answer, Pinker argues, is not stupidity but a mismatch between our evolved social minds and the demands of a large, anonymous, science-reliant civilization.
Why are we capable of such feats of reason and yet so often fail to reason well about the things that matter most?
Chapter 1 — How Rational an Animal?
Central question
Are humans fundamentally rational or fundamentally irrational — and what does the answer depend on?
Main argument
The rationalist and irrationalist stereotypes are both wrong. Pinker opens by staging a tension. On one side, rationality defines humanity: Homo sapiens is the species that reasons, plans, builds models of the world, and bends nature to its purposes. On the other side, two decades of behavioral science have catalogued hundreds of biases and illusions that seem to show humans as systematically irrational — the stock market panics, the climate denialism, the flat-earth YouTube channels.
The San hunters as a baseline. Pinker grounds the debate with Louis Liebenberg's fieldwork among the San people of the Kalahari. The San track game through arid desert by reading footprints, droppings, sweat marks, and disturbed vegetation, constructing a running probabilistic model of an animal's speed, gait, health, and likely direction. They use something close to Bayesian reasoning intuitively and accurately. This is not naïve cognition — it is sophisticated inference applied to the problems the environment has trained humans to solve over hundreds of millennia. The example establishes a key theme: humans are not generically rational or generically irrational; they are rational in some domains and not in others.
What rationality is — and is not. Pinker defines rationality as the use of knowledge to attain goals: specifically, forming accurate beliefs and making effective decisions given those beliefs. This is a means-end definition: rationality is always rationality relative to some goals and some evidence. It is distinct from intelligence, wisdom, common sense, and virtue. A highly intelligent person can reason irrationally; a person with modest analytic skills can behave quite rationally in their domain.
The Monty Hall problem as a provocation. Pinker introduces the famous Monty Hall problem — switching doors doubles your probability of winning, yet nearly everyone's intuition (including many statisticians') says otherwise — not to indict human reason but to show that formal probabilistic reasoning sometimes conflicts sharply with intuition. These conflicts are the puzzles the rest of the book's tool kit resolves.
Preview of the two puzzles the book will solve. Chapter 1 closes by naming the two questions that structure the book: (1) What are the tools of rational thought and how do they work? (2) If these tools exist and are teachable, why do individuals and societies so often fail to apply them? The answer to the second, Pinker previews, has to do with the social functions that beliefs serve — functions that are often orthogonal to truth.
Key ideas
- Rationality is not a single faculty but a collection of domain-specific skills that evolved, or can be learned, to solve specific types of problems.
- The San hunters demonstrate that intuitive probabilistic inference can be highly accurate when calibrated against a familiar environment.
- The Monty Hall problem illustrates that formal reasoning tools sometimes deliver conclusions that feel deeply wrong — signaling a mismatch, not a human defect.
- The definition of rationality as means-end belief/action coherence separates it from morality, intelligence, and common sense.
- The book's central tension: humans built science, mathematics, and liberal democracy — tools requiring extreme rationality — yet individually fall prey to obvious fallacies.
Key takeaway
Humans are context-dependent reasoners: brilliantly rational in the ecological niches evolution prepared them for, error-prone when the environment demands formal inference detached from social and physical cues.
Chapter 2 — Rationality and Irrationality
Central question
What is the best current framework for understanding why humans are rational in some situations but not others — and does the evidence from behavioral science really show that we are fundamentally irrational?
Main argument
The heuristics-and-biases program and its discontents. Pinker surveys the Kahneman-Tversky research tradition, which demonstrated that people use cognitive shortcuts — heuristics — that produce systematic errors: the availability heuristic (judging frequency by how easily examples come to mind), the representativeness heuristic (judging probability by resemblance to a category), anchoring, and dozens more. This program, culminating in Kahneman's Thinking, Fast and Slow, became the dominant narrative: System 1 (fast, automatic) leads us astray; System 2 (slow, deliberate) corrects it.
System 1 / System 2 and its limitations. Pinker explains the dual-process framework while noting that it can be overstated. Many System 1 processes are extremely accurate (face recognition, social cue reading, motor skills). Many System 2 processes are fallible too. The framework is useful but does not by itself explain when each system leads us right or wrong.
Ecological rationality: Gigerenzer's corrective. Pinker gives substantial attention to Gerd Gigerenzer's competing research program, which argues that heuristics that look like biases in artificial laboratory tasks are actually well-adapted to real-world environments with particular statistical structures. Gigerenzer's "fast-and-frugal" heuristics perform well in environments with the right features — they are not bugs, but evolved tools calibrated to nature's statistics. Pinker treats this not as a refutation of the biases program but as a complement: the question is always whether a given heuristic fits the environment it is being applied to.
Evolutionary and ecological rationality. Humans are, in this view, rational for the environment of evolutionary adaptedness (EEA) — small-group social life, immediate physical risks, recurrent foraging problems. The modern world creates environments radically different from the EEA: large anonymous institutions, statistical data, low-probability/high-stakes risks, long-run policy tradeoffs. Much apparent irrationality is a mismatch: a heuristic perfectly calibrated for ancestral life being applied in a context it was never designed for.
Rational irrationality: the social function of beliefs. Pinker introduces a crucial distinction: some irrational-seeming beliefs are, at the individual level, strategically rational. Holding a belief that one's tribe endorses signals loyalty and confers social benefits; that benefit can outweigh the epistemic cost of holding a false belief. This concept — rational irrationality, developed by economist Bryan Caplan — will become central in Chapter 10's analysis of myside bias.
The replication crisis as a rationality failure. Pinker ties the replication crisis in social psychology to failures of rational inference: researchers using underpowered studies, flexible analyses ("p-hacking"), and insufficient attention to prior probabilities inflated the literature with false positives. This is a systematic rationality failure at the institutional level, not merely an individual one.
Key ideas
- The availability heuristic causes people to overestimate the frequency of vivid, memorable events (plane crashes) and underestimate mundane but more frequent ones (car accidents).
- The representativeness heuristic causes the Linda problem error: "bank teller and feminist activist" seems more probable than "bank teller" because Linda matches the feminist stereotype — violating the conjunction rule.
- Ecological rationality shows that heuristics are not irrationality but context-dependent tools; a heuristic that fails in an artificial lab task may be optimal in its natural environment.
- System 1 versus System 2 framing is useful but does not fully capture why people reason correctly in some naturalistic contexts and fail in others.
- Rational irrationality: individuals can be epistemically irrational while being socially rational if false beliefs buy group membership.
Key takeaway
The debate between the heuristics-and-biases tradition and ecological rationality is not about whether humans are rational but about what "rational" means relative to the environment — a distinction that will drive the explanation of public irrationality later in the book.
Chapter 3 — Logic and Critical Thinking
Central question
What are the fundamental tools of deductive logic, what errors do they prevent, and why do people who can deploy them in naturalistic settings fail at formal logical puzzles?
Main argument
What logic is. Pinker introduces formal deductive logic as the study of arguments in which the conclusion follows necessarily from the premises, regardless of the content — the form is what matters. He covers categorical syllogisms (All A are B; C is A; therefore C is B), propositional logic (if–then, and, or, not), and the difference between valid arguments (the conclusion follows from the premises) and sound arguments (valid and the premises are true).
The Cognitive Reflection Test and motivated cognition. The CRT (e.g., "A bat and a ball cost $1.10. The bat costs $1 more than the ball. How much does the ball cost?") is designed so that a plausible but wrong answer comes to mind immediately. Getting the right answer requires overriding System 1. Pinker uses the CRT to illustrate how many everyday reasoning errors are not about logic but about failing to apply careful deliberation.
The Wason selection task and cheater detection. The Wason selection task asks subjects to identify which cards must be turned to test a rule like "If a card shows a vowel on one side, it has an even number on the other." Fewer than 10% solve it correctly in the abstract version. Yet when the same logical structure is embedded in a social contract — "If you are drinking alcohol, you must be over 18" — performance soars to over 70%. The lesson Pinker draws, following Cosmides and Tooby, is that human logical competence is domain-sensitive: we have evolved dedicated reasoning systems for social contracts and cheater detection that far outperform our general-purpose logical faculty.
Informal fallacies. Pinker catalogs the classical informal fallacies — ad hominem (attacking the person rather than the argument), appeal to authority, slippery slope, straw man, false dichotomy, appeal to nature, post hoc ergo propter hoc. These are not just rhetorical names; each represents a specific failure of argument structure that can corrupt reasoning even when no formal deductive rule is broken.
The Monty Hall problem revisited. Pinker walks through the full conditional probability argument for the Monty Hall problem: switching doors wins 2/3 of the time because the host's action (always revealing a goat behind one of the unchosen doors) transmits information. The lesson: people's intuitions are built for a world where events are independent; conditional probability requires formal bookkeeping that intuition resists.
The conjunction fallacy. The Linda problem: given a description of Linda as politically active and philosophically inclined, most people rate "Linda is a bank teller and a feminist activist" as more probable than "Linda is a bank teller" — violating the basic rule that P(A and B) ≤ P(A). Pinker notes this is partly a communication artifact (people interpret "bank teller" as implying "not feminist") but uses it to show how representativeness trumps probability calculus.
Key ideas
- Logical validity is purely about form: a valid argument with false premises can reach a false conclusion, and an invalid argument may happen to have a true conclusion.
- The abstract Wason task is failed by ~90% of people, but the same logical structure in a social-contract context is solved by ~70% — evidence for specialized cheater-detection machinery.
- Informal fallacies are named and categorized precisely because they are recurrent patterns of reasoning failure, not random noise.
- The conjunction fallacy shows that narrative coherence ("it all fits together") can override simple probability logic.
- Critical thinking is teachable, but only if students practice recognizing argument forms stripped of their rhetorical content.
Key takeaway
Logic competence is not domain-general: humans possess sophisticated reasoning in social-contract contexts and struggle in abstract formal ones — which is why teaching logic explicitly, not just modeling informal norms, is educationally important.
Chapter 4 — Probability and Randomness
Central question
What is probability, how should we reason under uncertainty, and why do humans systematically misjudge the likelihood of events?
Main argument
Five interpretations of probability. Pinker lays out the competing definitions: classical probability (ratio of favorable to total equally likely outcomes), frequentist probability (limiting frequency in a long run of trials), propensity probability (objective tendency of a setup to produce an outcome), subjective probability (a rational agent's degree of belief), and logical probability (the degree of support evidence gives a hypothesis). For everyday rational inference, the Bayesian subjective-belief interpretation proves the most useful because it applies to single events and updates on new evidence.
Basic probability calculus. The chapter covers the addition rule (P(A or B) = P(A) + P(B) − P(A and B) for events that are not mutually exclusive), the multiplication rule for independent events (P(A and B) = P(A) × P(B)), and the law of total probability. Pinker uses these not as mere formulas but to show where intuition reliably breaks down.
The gambler's fallacy and the law of large numbers. After a roulette wheel has landed on red six times in a row, most people feel black is "due" — the gambler's fallacy. In truth, each spin is independent. Pinker explains the correct principle: the law of large numbers guarantees that in the long run relative frequencies converge to probabilities, but short-run sequences are highly variable and have no "memory." The fallacy confuses the short run with the long run.
Availability heuristic and media distortions. People judge the frequency of events by how easily examples come to mind. Dramatic, vivid, recently covered events — plane crashes, shark attacks, terrorist murders — are over-represented in memory and therefore feel more common than they are. Car accidents, which kill vastly more people, feel less threatening because they generate less memorable coverage. Pinker connects this directly to public policy distortions: resources flow toward vivid threats (terrorism) and away from mundane but far more lethal ones (cardiovascular disease).
The birthday problem. In a room of 23 people, the probability that at least two share a birthday exceeds 50%. This is highly counterintuitive because people think in terms of pairwise comparisons between themselves and one other person, not the total number of possible pairings (253 pairs). The birthday problem shows that humans are poor intuitive combinatorialists.
Regression to the mean. When an extreme event occurs — a student aces an exam, a player hits a home run — the next observation tends to be closer to average, not because anything has changed but because the first event was partly lucky. The failure to account for regression to the mean produces systematic errors: praising a student after an exceptional performance and then attributing their subsequent average performance to complacency; designing flight-training feedback around confirmation that punishment works (when performance improves after harsh feedback, the improvement would have happened anyway due to regression).
Key ideas
- The five interpretations of probability are not interchangeable; the Bayesian subjective interpretation is most appropriate for updating beliefs on evidence.
- The gambler's fallacy (short-run correction) is distinct from the valid law of large numbers (long-run convergence).
- Availability heuristic errors create systematic misallocation of fear and policy attention toward vivid, memorable threats.
- Regression to the mean is ubiquitous and routinely misinterpreted as evidence of a causal intervention.
- People are poor combinatorialists, underestimating how quickly the number of possible pairings grows with group size.
Key takeaway
Human probability intuitions are systematically skewed by availability, by confusion between short-run and long-run statistics, and by the failure to account for regression to the mean — errors that formal probability calculus exists to correct.
Chapter 5 — Beliefs and Evidence: Bayesian Reasoning
Central question
How should rational agents update their beliefs when they receive new evidence — and what systematic errors arise when they fail to do so?
Main argument
Bayes's theorem. Pinker presents Bayes's theorem as the central engine of rational belief update:
P(hypothesis | evidence) = [P(evidence | hypothesis) × P(hypothesis)] / P(evidence)
In plain English: the probability that a hypothesis is true, given the evidence, equals the prior probability of the hypothesis times the likelihood of seeing that evidence if the hypothesis were true, divided by the probability of seeing that evidence overall. The posterior probability — your updated belief — is proportional to the prior times the likelihood.
Base-rate neglect: the medical test example. Consider a disease affecting 1% of the population. A test is 90% accurate (90% sensitivity, 90% specificity). You test positive. What is the probability you have the disease? Most people say 90%. The correct answer is roughly 8%. Because the disease is rare (prior = 0.01), the large population of healthy people generates many false positives (0.10 × 0.99 ≈ 10%) while the small diseased population generates relatively few true positives (0.90 × 0.01 = 0.9%). The Bayesian calculation reveals that the apparent accuracy of the test is massively diluted by the base rate. This example has direct implications for medical screening programs, criminal forensics, and security screening.
Natural frequencies as a teaching tool. Gigerenzer, Cosmides, and Tooby showed that base-rate reasoning improves dramatically when information is presented as natural frequencies ("out of 1,000 people, 10 have the disease; 9 of those will test positive; of the 990 who don't have the disease, 99 will also test positive") rather than probabilities. The natural-frequency format taps the same cognitive machinery that our ancestors used for counting — it makes Bayesian reasoning intuitive.
The prior matters: extraordinary claims require extraordinary evidence. Because Bayes's theorem weights the posterior by the prior, a claim that contradicts everything we know about physics (a homeopathic remedy that works by quantum resonance) requires overwhelming evidence to become believable — because the prior probability is extremely low. This is the formal underpinning of Carl Sagan's dictum. Conversely, a common-sense claim needs only modest confirming evidence.
The replication crisis as Bayesian failure. Pinker argues that much of the psychology replication crisis stems from researchers and readers failing to use appropriate priors. When a study tests a surprising, counterintuitive hypothesis (a power pose changes hormone levels; unconscious priming shapes voting behavior) and obtains a statistically significant result (p < 0.05), the reflex is to believe the hypothesis. But if the prior probability of the hypothesis was very low — say, 5% — even a genuine p < 0.05 result yields a posterior below 50% (due to false-positive risk). The machinery of null-hypothesis significance testing was designed without priors and therefore systematically over-sells surprising results.
Prior probabilities and epistemic norms. Pinker discusses how individuals' priors are shaped by their ideological communities, which creates "epistemic bubbles." A conservative and a progressive looking at the same ambiguous evidence on crime statistics will update differently because they started from different priors — and Bayes's theorem shows this divergence is formally legitimate as long as their priors reflect genuinely different background beliefs. The problem arises when priors are not formed from evidence but from tribal affiliation.
Key ideas
- Bayes's theorem: posterior probability is prior probability updated by the likelihood ratio of the evidence.
- Base-rate neglect is one of the most consequential reasoning errors in medicine, law, and security.
- Natural-frequency formats bypass the base-rate neglect problem by leveraging evolved counting cognition.
- Low priors mean that even a significant p-value may leave the hypothesis below 50% posterior — the Bayesian explanation for the replication crisis.
- The prior's origin matters: priors formed from evidence are epistemically defensible; priors inherited from tribal membership produce persistent divergence even in the face of identical evidence.
Key takeaway
Bayesian reasoning is the normative standard for rational belief update: multiply your prior by the likelihood ratio of the evidence, and the base rate can never be ignored without distorting the result.
Chapter 6 — Risk and Reward: Rational Choice and Expected Utility
Central question
How should a rational agent make decisions under uncertainty, and what are the limits of the classical expected-utility framework?
Main argument
Expected utility theory. John von Neumann and Oskar Morgenstern proved in 1944 that any agent whose preferences satisfy a small set of coherence axioms (completeness, transitivity, independence, continuity) will behave as if maximizing expected utility: the probability-weighted sum of the utilities of outcomes. The theory does not prescribe what goals to pursue — it prescribes consistency in how you pursue them. Rational choice is not about maximizing money or pleasure; it is about being coherent across choices.
The St. Petersburg paradox and diminishing marginal utility. The St. Petersburg paradox offers a coin-flipping game with infinite expected monetary value, yet no reasonable person would pay much to play. The resolution: utility is not linear in money — the marginal utility of an additional dollar decreases as wealth increases (diminishing marginal utility). Daniel Bernoulli's logarithmic utility function captures this: the rational agent maximizes expected log-wealth, not expected dollars. This is the formal justification for buying insurance and for risk aversion.
The Allais paradox and its significance. Maurice Allais constructed pairs of choices in which virtually everyone violates the independence axiom of expected utility theory: they prefer a guaranteed $1 million to a lottery in a way that is inconsistent with how they choose between two risky lotteries. This suggests that people weight certainty extra-heavily — they are not expected-utility maximizers. Kahneman and Tversky's prospect theory generalizes by adding loss aversion (losses loom larger than equivalent gains), probability weighting (overweighting small probabilities, underweighting large ones), and a reference-point dependence (people care about changes from a reference point, not absolute levels).
Loss aversion in practice. Loss aversion helps explain a range of economic puzzles: why people hold losing stocks too long (selling would crystallize the loss), why homeowners refuse to lower house prices below purchase price (a nominal loss), and why employees respond more to pay cuts than to equivalent shortfalls in expected bonuses.
Pascal's mugging. Pinker discusses the puzzle that expected-utility maximization, applied naively to extreme low-probability/high-stakes outcomes, can produce absurd recommendations: a mugger threatening "astronomical punishment" could in principle extort any amount, because no matter how small the probability is, the expected harm is infinite. This motivates bounded utility functions and the need for robustness criteria.
Rationality of risk preferences. Pinker defends von Neumann and Morgenstern against the charge that the Allais paradox debunks expected utility: the axioms are normatively compelling even if descriptively violated, because violating them makes you susceptible to Dutch books (betting arrangements that guarantee a loss). The chapter argues that learning expected-utility thinking reduces costly errors even if it does not describe actual behavior.
Key ideas
- Expected utility = Σ P(outcomei) × U(outcomei); a rational agent maximizes this quantity given their preferences.
- Diminishing marginal utility explains why risk aversion is rational and why insurance is rational for risk-averse agents.
- The Allais paradox shows that humans weight certainty extra-heavily, violating the independence axiom — captured by prospect theory.
- Loss aversion: losses are approximately twice as painful as equivalent gains are pleasurable.
- Pascal's mugging reveals that unbounded utility combined with tiny nonzero probabilities produces absurd results, motivating real-world caveats to the theory.
- Dutch books: an agent who violates coherence axioms can be made to accept combinations of bets that guarantee a loss.
Key takeaway
Expected utility theory is the normative benchmark for rational choice under uncertainty; human departures from it (especially loss aversion and probability weighting) are well-documented and costly in practice.
Chapter 7 — Hits and False Alarms: Signal Detection and Statistical Decision Theory
Central question
When decisions must be made under uncertainty about whether a signal is real or noise, how should a rational agent set their threshold — and what framework resolves the tradeoffs between false positives and false negatives?
Main argument
The signal detection framework. Signal detection theory (SDT), developed in the 1950s for radar operators and extended to psychology, starts with the insight that any detection decision involves two overlapping distributions: the distribution of evidence when a signal is absent (noise alone) and when it is present (signal plus noise). Because the distributions overlap, no criterion perfectly separates them — any threshold generates both misses (false negatives: signal present but not detected) and false alarms (false positives: no signal but detector fires).
The four outcomes: hits, misses, false alarms, correct rejections. Pinker lays out the 2×2 signal-detection matrix. A radiologist reading mammograms, a judge deciding guilt, a security screener, a spam filter — all face the same tradeoff: tightening the criterion reduces false alarms but increases misses; relaxing it reduces misses but increases false alarms. Neither error is free; their relative costs depend on context.
The ROC curve. The Receiver Operating Characteristic curve plots hit rate (true positive rate) against false-alarm rate as the criterion varies. A perfect detector has a point at (0, 1) — all hits, no false alarms. A random detector follows the diagonal. The area under the curve (AUC) measures discriminability, separating sensitivity from response bias. Pinker uses the ROC curve to show that many real controversies about whether a test "works" are actually debates about where to set the criterion, not about sensitivity.
d-prime: measuring sensitivity independently of bias. The quantity d' (d-prime) measures how well-separated the signal and noise distributions are, independently of the criterion. Two detectors with the same hit rate can have very different d-primes if one operates with a lax criterion. Many evaluations of medical tests, security systems, and judicial standards confuse criterion placement with sensitivity.
Courtroom applications. Pinker walks through how SDT applies to the criminal standard "beyond reasonable doubt." Setting a very high threshold for conviction (stringent criterion) minimizes false convictions (false alarms) but increases wrongful acquittals (misses). The famous Blackstone ratio ("better that ten guilty persons escape than one innocent suffer") is a statement about how society should set the criterion — not a claim about discriminability.
Calibration and prediction tournaments. Pinker discusses Philip Tetlock's work on forecasting: superforecasters who decompose problems, assign precise probabilities, track their calibration, and update regularly dramatically outperform experts. SDT connects naturally: good forecasters keep their posteriors well-separated from chance (high d') and set their expression criteria to match the actual costs of false positives vs. false negatives.
Key ideas
- Signal detection theory separates discriminability (d', or how well the signal and noise distributions are separated) from criterion (where you set the threshold).
- The ROC curve shows the full tradeoff between hit rate and false-alarm rate; the optimal point depends on the relative costs of the two error types.
- "Beyond reasonable doubt" is a criterion choice, not a scientific claim — its optimal placement depends on how society weights false conviction versus false acquittal.
- Many apparently controversial findings about test accuracy are actually disagreements about criterion placement, not about the underlying discriminability.
- Tetlock's superforecasters exemplify high d-prime and well-calibrated criteria applied systematically.
Key takeaway
Rational detection decisions require specifying both the discriminability of available evidence and the relative costs of false positives versus false negatives — conflating the two leads to systematic errors in medicine, law, and security policy.
Chapter 8 — Self and Others: Game Theory
Central question
When rational individuals interact strategically — where each person's outcome depends on the choices of others — what does rationality predict, and why do rational individuals so often end up in collectively bad outcomes?
Main argument
Zero-sum and non-zero-sum games. Pinker begins by distinguishing zero-sum games (my gain is exactly your loss: chess, poker, war over a fixed resource) from non-zero-sum games, where both players can gain or both can lose. Most of the interesting interactions in social life are non-zero-sum, which means cooperation and collective action are possible — but not guaranteed.
The prisoner's dilemma. Two suspects, unable to communicate, must independently choose to cooperate (stay silent) or defect (testify against the other). If both cooperate, both get a light sentence; if both defect, both get a heavy sentence; if one defects and the other cooperates, the defector goes free and the cooperator gets the worst sentence. Defection is the dominant strategy — it is better for each player regardless of what the other does — yet mutual defection leaves both worse off than mutual cooperation. This is the canonical illustration of how individually rational choices produce collectively irrational outcomes.
Nash equilibrium. A Nash equilibrium is a set of strategies from which no player has an incentive to deviate unilaterally. The prisoner's dilemma has a unique Nash equilibrium at mutual defection — which is Pareto inferior to mutual cooperation. Pinker uses this to show that rationality does not guarantee efficiency; the market, left to its own devices, can get stuck in Nash equilibria that leave welfare on the table.
Coordination games and multiple equilibria. In coordination games (drive on the left or the right?), any consistent standard is a Nash equilibrium. There is no prisoner's dilemma tension — both players prefer coordination — but without communication or convention, they may fail to coordinate. Conventions, institutions, and laws solve coordination problems by selecting an equilibrium.
The tragedy of the commons. Garrett Hardin's framing: each individual is rational to exploit a shared resource (a fishery, an aquifer, the atmosphere) as intensively as possible, because restraint simply leaves more for others to exploit. But if all reason this way, the commons is depleted. Pinker frames the tragedy of the commons as an iterated multiplayer prisoner's dilemma and notes that Elinor Ostrom's work shows real communities solve it through reputation, monitoring, and graduated sanctions — without requiring either privatization or government.
Iterated games and the evolution of cooperation. When the prisoner's dilemma is played repeatedly between the same parties, defection is no longer dominant. Robert Axelrod's famous computer tournaments showed that tit-for-tat (cooperate on the first move, then do whatever the other player did last round) is remarkably robust: it is nice (never defects first), provocable (retaliates immediately), forgiving (returns to cooperation after punishment), and clear (the strategy is transparent). Cooperation can evolve through reciprocity when interactions are repeated and the shadow of the future is long enough.
Volunteer dilemmas and social loafing. In a volunteer dilemma, one agent needs to take a costly action for the group to benefit. Each individual prefers that someone else pay the cost. The Nash equilibrium is mixed-strategy: each person volunteers with some probability that depends on the cost and the size of the group. Larger groups have a lower probability of volunteering per person — the bystander effect has a game-theoretic explanation.
Key ideas
- The prisoner's dilemma shows that individual rationality and collective rationality can diverge sharply: Nash equilibrium is not the same as Pareto optimality.
- Nash equilibrium: a strategy profile where no individual can improve their outcome by unilaterally changing their strategy.
- The tragedy of the commons is a multiplayer prisoner's dilemma, soluble by reputation, monitoring, and institutions (Ostrom).
- In repeated games, tit-for-tat achieves cooperation without altruism: it is individually rational when the shadow of the future is long.
- The bystander effect is a mathematical prediction of the volunteer dilemma, not merely a social-psychological curiosity.
Key takeaway
Game theory reveals the precise conditions under which self-interest produces collective welfare (coordination games, repeated interactions with reciprocity) and when it does not (one-shot prisoner's dilemmas, tragedies of the commons) — and what institutional solutions can shift the equilibrium.
Chapter 9 — Correlation and Causation
Central question
How can we distinguish genuine causal relationships from mere statistical correlations — and what methods allow us to draw causal conclusions from data?
Main argument
The correlation/causation distinction. Pinker opens with vivid examples of spurious correlations: countries with more TVs per capita have lower birth rates; Nicolas Cage films are correlated with swimming pool drownings. Correlation — a statistical association between two variables — is not causation, because a third variable might cause both, because time lags create confounds, or because the association is pure chance. Yet much of everyday reasoning, most media reporting, and considerable social policy is built on observing a correlation and concluding causation.
The causal inference revolution. Pinker introduces the framework developed by Judea Pearl and popularized by econometricians and epidemiologists. Causal reasoning requires specifying a causal model: a directed graph (DAG — directed acyclic graph) in which arrows represent hypothesized cause-effect relationships. The graph makes explicit which variables are causes, which are confounders, and which are colliders (variables caused by two others). Without specifying the causal structure, statistical controls can make inferences worse rather than better.
Confounders and selection bias. A confounder is a third variable that causes both the apparent cause and the apparent effect. Classic example: ice cream sales and drowning rates are both caused by summer weather (a confounder). Controlling for the confounder eliminates the spurious correlation. Selection bias — a systematic difference in who is observed — is another major source of spurious associations: hospital patients who have both disease A and disease B may be overrepresented compared to the general population (Berkson's paradox).
Randomized controlled trials as the gold standard. By randomly assigning participants to treatment and control conditions, randomization ensures that on average, all confounders — known and unknown — are equally distributed across groups. The only systematic difference between groups is the treatment. Pinker argues this makes RCTs the gold standard for causal inference in medicine, education, and policy — and advocates for extending the use of RCTs in public policy (evidence-based policy).
Natural experiments and quasi-experimental methods. Where RCTs are impossible or unethical, researchers use natural experiments (cases where assignment to treatment was effectively random for reasons unrelated to the outcome), regression discontinuity designs (comparing individuals just above and just below a threshold), instrumental variables (a variable that affects the treatment but not the outcome except through the treatment), and difference-in-differences analyses. Pinker illustrates each with real-world applications: the Oregon Medicaid lottery, Card and Krueger's minimum-wage study, the Vietnam draft lottery.
The hierarchy of evidence. Not all causal evidence is equal. Pinker presents an informal hierarchy: randomized controlled trials > natural experiments > instrumental variables > regression discontinuity > observational studies with many controls > single case studies > anecdotes. Moving up the hierarchy requires correspondingly stronger claims. Policies based on anecdote or uncontrolled observation — "traditional wisdom," the highest-paid person's opinion (HiPPO) — are epistemically vulnerable even when they contain a kernel of truth.
Key ideas
- Correlation is not causation; the three possible explanations for a correlation are direct causation, reverse causation, and confounding.
- Directed acyclic graphs (DAGs) formalize causal models and reveal which variables to control for (confounders) and which to avoid controlling for (colliders).
- Randomized controlled trials balance all confounders by design — they are the most rigorous causal method available.
- Natural experiments and quasi-experimental methods extract causal information from observational data when randomization is impossible.
- Evidence-based policy means replacing dogma, folklore, and HiPPO with RCTs and natural experiments wherever feasible.
Key takeaway
Causal inference requires specifying a causal model — identifying confounders and ruling out alternative explanations — and the randomized experiment is the gold standard for doing so, with quasi-experimental methods as the best available alternative when randomization is not possible.
Chapter 10 — What's Wrong with People?
Central question
If the tools of rational thought are available and teachable, why do so many people — including educated, intelligent people — hold irrational beliefs, believe in conspiracy theories, deny scientific consensus, and maintain factually false views?
Main argument
The epistemological paradox. Pinker opens with a puzzle: the tools of rational thought (logic, probability, Bayesian inference, causal analysis) have been publicly available for centuries. Yet in the age of universal education and the internet, conspiracy theories, pseudoscience, and motivated political reasoning seem to flourish. The chapter offers an explanation grounded not in stupidity but in the social functions of belief.
Motivated reasoning: reasoning like a lawyer, not a scientist. The first mechanism is motivated reasoning: using the faculty of reason not to discover truth but to construct a defense of a predetermined conclusion. A lawyer builds the best possible case for their client regardless of guilt; a motivated reasoner does the same for their favorite belief. The evidence is gathered selectively, counterarguments are dismissed with more scrutiny than supporting arguments, and the internal sense of having "thought it through" is entirely maintained — reasoning has occurred, but in the wrong direction.
Myside bias: asymmetric critical thinking. Building on Keith Stanovich's research, Pinker distinguishes myside bias from the general biases of Chapter 2. Myside bias is specifically the tendency to apply rigorous critical scrutiny to evidence that opposes a favored view and to accept evidence that supports it with minimal scrutiny. Studies show that intelligent people are more susceptible to myside bias than less intelligent people: greater cognitive ability provides more tools for rationalization. This is sometimes called the "smart idiot" effect — those with higher analytical skills construct more sophisticated defenses of false beliefs.
The reality mindset vs. the mythology mindset. Pinker distinguishes two modes of engaging with belief. The reality mindset treats beliefs as representations of the world: they are subject to revision, they generate predictions, and they must cohere with evidence. The mythology mindset treats beliefs as social commitments: they signal group membership, affirm shared values, and bind communities together. Many beliefs operate simultaneously in both registers — but when they are in the mythology register, they become immune to empirical challenge, because their function is not to represent facts but to express identity.
The tragedy of the commons in epistemology. Pinker frames the failure of public rationality as a collective-action problem. Individual rational agents can be collectively irrational if truth-seeking is a public good that benefits everyone but whose production depends on costly individual effort. Holding a politically incorrect belief, correcting a popular myth, or publicly disavowing a tribal narrative all impose social costs on the individual while providing diffuse benefits to the epistemic commons. This creates a prisoner's dilemma: rational individuals underproduce rational public discourse.
Cognitive-biases-as-features: the evolutionary case. Pinker discusses the argument that motivated reasoning and myside bias are not design flaws but adaptations. Hugo Mercier and Dan Sperber's "argumentative theory of reasoning" holds that human reason evolved not for individual truth-seeking but for social argumentation — to convince others and to defend against being convinced by bad arguments. On this view, confirmation bias is not a bug in a truth-seeking machine but a feature of a social advocacy machine.
Conspiratorial thinking and the tribal epistemology. Conspiracy theories satisfy multiple psychological needs simultaneously: they provide a simple causal story for complex events, confer insider status on believers, and position the believer as a courageous truth-seeker against a corrupt establishment. Pinker notes that many conspiracy theories are unfalsifiable by design — any disconfirming evidence can be reframed as evidence of the conspiracy's reach. The "tribal epistemology" Pinker diagnoses is a version of myside bias scaled up to group level: entire political communities treat information sources as trustworthy or untrustworthy based on tribal affiliation rather than track record.
Key ideas
- Motivated reasoning inverts the epistemic process: the conclusion is fixed, and reason is recruited to justify it.
- Myside bias is especially pronounced among high-intelligence individuals, who use their reasoning skills to construct more sophisticated rationalizations.
- The reality/mythology distinction explains why some beliefs are impervious to evidence: they are not meant to represent facts but to signal membership.
- The tragedy of the commons in epistemology: individual rationality (social conformity, tribal signaling) is collectively irrational (degrades public epistemic standards).
- The argumentative theory of reasoning (Mercier and Sperber): reason evolved for social persuasion, not individual truth-seeking — which explains many of its characteristic features.
- Conspiracy theories are self-sealing: the structure of the belief immunizes it against falsification.
Key takeaway
Human irrationality in public life is not primarily a cognitive deficit but a social one: beliefs serve tribal-signaling and identity functions that take precedence over epistemic functions, producing individually rational but collectively catastrophic departures from truth-seeking.
Chapter 11 — Why Rationality Matters
Central question
What is the case for making rationality a social priority — for institutions, education, and culture — and what does the evidence show about what happens when societies do and do not apply rational methods?
Main argument
Rationality and personal flourishing. Pinker cites Wändi Bruine de Bruin's research showing that individuals who score better on measures of rational thinking (calibrated Bayesian reasoning, avoidance of framing effects, deliberate decision processes) have measurably better life outcomes: better financial decisions, better health behaviors, fewer legal problems, more stable relationships. This is not a trivial correlation — the effect persists after controlling for general intelligence and socioeconomic background.
The historical case: Enlightenment arguments that changed the world. Pinker argues that many of the greatest moral advances of the past three centuries were driven by arguments, not just moral sentiment. Cesare Beccaria's argument against torture and cruel punishment in 1764 — grounded in consequentialist reasoning about deterrence and proportionality — contributed directly to the reform of criminal law. The arguments against slavery advanced by Enlightenment thinkers applied the principle of impartial consideration: if we object to being enslaved ourselves, consistency requires we object to enslaving others. The same logical structure — identify the morally relevant properties, apply them impartially — drove arguments for women's suffrage and civil rights. Rationality, applied to ethics, expands the moral circle.
The expanding circle and utilitarian reasoning. Pinker draws on Peter Singer's "expanding circle" to argue that rational moral reasoning — asking what principle I would accept if I did not know which person I would be — tends to expand the set of beings to whom moral consideration is owed. This is not mere sentimentalism; it is an application of the impartiality and consistency requirements that are internal to rationality itself.
Science as institutionalized rationality. Science is not just a collection of facts; it is a set of social and epistemic practices — peer review, replication, open publication, statistical disclosure — designed to correct for the biases of individual scientists. Pinker argues that the institutions of science embody collective rationality in the same way that legal institutions embody collective justice. The extraordinary achievements of science — vaccine development, materials science, the germ theory of disease — are the fruits of applying rational methods systematically over time.
Democracy as institutionalized rationality. A functioning liberal democracy applies similar logic: competing arguments are aired, evidence is assessed (imperfectly), power is checked, and leaders can be removed without violence. Pinker connects this to the book's broader argument: the threats to democracy from disinformation, conspiracy thinking, and tribal epistemology are threats to the institutionalized rational discourse that democracy depends on.
Progress as the product of rationality. Pinker closes with the empirical ledger from his earlier books (The Better Angels of Our Nature, Enlightenment Now): reductions in extreme poverty, famine, child mortality, and interpersonal violence over the past two centuries. He attributes this progress not to luck but to the application of rational methods — evidence-based medicine, agricultural science, public health engineering, rational governance — and argues that continued progress depends on defending and extending those institutions.
The responsibility of the educated. Pinker ends with a challenge: those who understand the tools of rational thought have an obligation to apply them publicly, to model epistemic norms, and to resist tribal epistemology even when it is socially costly. Rationality, like democracy, is not self-sustaining; it requires active maintenance.
Key ideas
- Bruine de Bruin's research: better-reasoned individuals have better life outcomes even controlling for IQ — rationality has personal returns beyond general intelligence.
- The arguments that abolished torture, slavery, and oppression were formal logical arguments appealing to impartiality and consistency, not merely emotional appeals.
- Science and liberal democracy are institutional embodiments of rational discourse — they impose checks on individual biases that individuals alone cannot correct.
- The expanding moral circle is driven partly by rational argumentation: consistency and impartiality demand widening the set of beings to whom rights are extended.
- Progress (declining poverty, violence, disease) is the empirical result of institutionalized rational methods applied over two centuries.
- Tribal epistemology threatens not just individual rationality but the institutional infrastructure — science, democracy, journalism — that collective rationality depends on.
Key takeaway
Rationality matters at every scale: individually, it improves the quality of decisions and the quality of life; institutionally, it is the mechanism through which science, democracy, and moral progress become possible — which is why defending rational discourse is both a personal and a civic obligation.
The book's overall argument
Chapter 1 (How Rational an Animal?) — establishes the central puzzle: humans are capable of extraordinary feats of rational inference (the San hunters, mathematics, science) yet fall into systematic errors in formal reasoning tasks, raising the question of what rationality is and when it is applied.
Chapter 2 (Rationality and Irrationality) — surveys the major frameworks — heuristics-and-biases, ecological rationality, dual-process theory — and introduces rational irrationality: the insight that holding socially useful false beliefs can be individually rational, previewing the social-function explanation of public irrationality.
Chapter 3 (Logic and Critical Thinking) — introduces the first formal tool: deductive logic and informal fallacy detection; demonstrates that human logical competence is domain-sensitive (social contracts: high; abstract formal logic: low) and therefore that explicit instruction matters.
Chapter 4 (Probability and Randomness) — introduces probabilistic reasoning as the second tool; shows that the gambler's fallacy, availability heuristic, and regression-to-the-mean errors are systematic and correctible once the formal structure is understood.
Chapter 5 (Beliefs and Evidence: Bayesian Reasoning) — introduces Bayesian inference as the normative standard for updating beliefs on evidence; demonstrates that base-rate neglect is the root of medical misdiagnosis, false-positive science, and the replication crisis.
Chapter 6 (Risk and Reward: Rational Choice and Expected Utility) — introduces expected utility theory as the framework for decision under uncertainty; catalogs the well-documented human departures from it (loss aversion, the Allais paradox) and defends the normative status of the theory despite these violations.
Chapter 7 (Hits and False Alarms: Signal Detection and Statistical Decision Theory) — introduces signal detection theory; shows that rational detection decisions require specifying both discriminability and the relative costs of false positives versus false negatives — a requirement routinely violated in medicine, law, and security.
Chapter 8 (Self and Others: Game Theory) — introduces game theory; shows that individually rational choices can be collectively disastrous (prisoner's dilemma, tragedy of the commons) but that repeated interaction, reciprocity, and institutions can rescue collective rationality.
Chapter 9 (Correlation and Causation) — introduces causal inference; argues for randomized controlled trials and quasi-experimental methods as the rational tools for causal knowledge, and for evidence-based policy as their institutional application.
Chapter 10 (What's Wrong with People?) — synthesizes the tools of the preceding chapters to explain the persistence of public irrationality: motivated reasoning and myside bias turn the tools of reason against truth-seeking; the reality/mythology distinction explains why certain beliefs are impervious to evidence; the tragedy of the commons explains why individually rational behavior degrades collective epistemic standards.
Chapter 11 (Why Rationality Matters) — completes the argument by showing what is at stake: individual well-being, moral progress, and the institutional infrastructure of science and democracy all depend on maintaining rational discourse — making the defense of rationality a personal and civic imperative.
Common misunderstandings
Misunderstanding: Pinker claims humans are generally rational.
The book's argument is more precise: humans are rational in specific ecological niches (social contracts, physical causation, recurrent foraging problems) and irrational in others (abstract probability, base rates, long-run statistical thinking). The claim is not global rationality but context-dependent competence.
Misunderstanding: The book is purely a criticism of cognitive biases, in the tradition of Kahneman's Thinking, Fast and Slow.
The book is partly that, but it also presents a significant corrective from Gigerenzer's ecological rationality program — arguing that many apparent biases are context-appropriate heuristics. And unlike the biases literature, Pinker's primary focus is on why rationality fails in public and political life, not just in individual laboratory tasks.
Misunderstanding: Myside bias is just ordinary political partisanship.
Pinker follows Stanovich in treating myside bias as a distinct cognitive phenomenon: asymmetric critical thinking based on outcome preference. It is different from general bias, from low intelligence, and from mere disagreement. Critically, it is stronger in more intelligent individuals, which is why it is not solved by education alone.
Misunderstanding: Pinker's defense of rationality amounts to scientism — reducing all questions to empirical inquiry.
Pinker explicitly limits the claim: rational tools are means-end instruments for achieving goals, not sources of ultimate values. The book does not claim science can determine what goals to pursue — that is a separate question. It claims that once goals are specified, rational methods are the best available means for achieving them.
Misunderstanding: The book argues that irrationality is purely an individual problem fixable by better education.
The book's most important argument is structural: irrationality in public life is a collective-action problem — a tragedy of the commons in the epistemic domain. Individual education is necessary but not sufficient; institutional solutions (science, peer review, democratic accountability, fact-checking norms) are also required.
Central paradox / key insight
The central paradox of the book is the rationality gap: humans built Euclidean geometry, calculus, quantum mechanics, and the double-blind clinical trial — achievements requiring extreme rational precision — yet remain susceptible to horoscopes, conspiracy theories, and the gambler's fallacy. How can the species that created formal logic believe in homeopathy?
Pinker's resolution has two parts. First, rational competence is domain-specific: the same species can be brilliantly rational in the niches evolution shaped (tracking game, reading social cues, building tools) and poor in domains that require formal decontextualized reasoning. Second — and more importantly — holding irrational public beliefs is often individually rational: it signals tribal loyalty, avoids social punishment, and maintains group membership. The tragedy is that individually rational epistemic behavior produces collectively irrational public discourse.
The problem is not that humans lack the capacity for rational thought — the existence of science, law, and mathematics demonstrates otherwise. The problem is that rational thought is hard, socially costly, and often in competition with the much easier and socially rewarded option of believing what your tribe believes.
Important concepts
Rationality
The use of knowledge and reasoning to achieve goals; specifically, forming accurate beliefs given evidence and making coherent decisions given those beliefs. Pinker's definition is means-end: rationality is always relative to goals and evidence, not a standalone faculty.
Ecological rationality
Gigerenzer's framework: heuristics that appear biased in formal laboratory settings may be optimally adapted to the statistical structure of natural environments. A fast-and-frugal heuristic is ecologically rational if it performs well in the environment it evolved for.
Bayesian reasoning
Updating beliefs by multiplying prior probability by the likelihood ratio of new evidence. Bayes's theorem: P(H|E) = [P(E|H) × P(H)] / P(E). The posterior probability tracks the prior plus the strength of the evidence.
Base-rate neglect
The error of ignoring the prior probability of a hypothesis when computing its posterior probability — the most common Bayesian failure, responsible for medical misdiagnosis, false-positive science, and security screening errors.
Expected utility
The probability-weighted sum of the utilities of outcomes: EU = Σ P(outcomei) × U(outcomei). Rational choice under uncertainty consists in maximizing expected utility, according to von Neumann and Morgenstern.
Loss aversion
The finding (from prospect theory) that losses are approximately twice as painful as equivalent gains are pleasurable. A key departure from expected-utility maximization that explains risk aversion, the endowment effect, and many financial anomalies.
Signal detection theory (SDT)
A framework separating the discriminability of a signal from noise (d-prime) from the criterion set to detect it. Any detection decision involves a tradeoff between false positives (false alarms) and false negatives (misses); the optimal criterion depends on the relative costs.
Prisoner's dilemma
A game in which two players each have a dominant strategy (defect) that leads to a mutually inferior outcome compared to mutual cooperation. The Nash equilibrium is Pareto-inferior — the canonical demonstration that individual rationality and collective welfare can diverge.
Nash equilibrium
A strategy profile in which no player can improve their payoff by unilaterally changing their strategy, given the strategies of the other players. A stable outcome of strategic interaction, but not necessarily a good one.
Tragedy of the commons
A situation in which each individual's rational overexploitation of a shared resource leads to its depletion, harming all users. A multiplayer prisoner's dilemma, soluble by institutions that change individual incentives.
Motivated reasoning
Using the faculty of reason to defend a predetermined conclusion rather than to discover truth — reasoning like a lawyer for the defense rather than like a scientist investigating a hypothesis.
Myside bias
The tendency to apply critical scrutiny asymmetrically: demanding high standards of evidence for claims that contradict preferred beliefs and accepting weak evidence for claims that support them. Distinct from general bias and — paradoxically — stronger in higher-intelligence individuals.
Reality mindset vs. mythology mindset
The reality mindset treats beliefs as representations of the world, subject to revision by evidence. The mythology mindset treats beliefs as social commitments, expressing group identity and values. Many public beliefs operate in the mythology register and are therefore impervious to factual refutation.
Rational irrationality
Bryan Caplan's concept: an agent can be epistemically irrational (holding false beliefs) while being individually rational if the social benefits of false belief (group membership, signaling loyalty) outweigh the epistemic costs. Explains why rational individuals produce irrational public discourse.
Directed acyclic graph (DAG)
A formal representation of a causal model: nodes are variables, directed edges represent causal hypotheses, and the absence of cycles enforces that causes precede effects. Used in Judea Pearl's causal inference framework to identify confounders, colliders, and the correct statistical controls.
Tit-for-tat
Robert Axelrod's winning strategy in iterated prisoner's dilemma tournaments: cooperate on the first move, then mirror the other player's previous move. Tit-for-tat is nice, provocable, forgiving, and clear — properties that make cooperation evolutionarily stable when interactions are repeated.
References and Web Links
Primary book and edition information
- Pinker, Steven. Rationality: What It Is, Why It Seems Scarce, Why It Matters. Viking/Penguin Random House, New York, 2021.
Background and overview
- Wikipedia: Rationality (book)
- Harvard Gazette excerpt from the book (Chapter 1)
- Pinker's Harvard course on Rationality (GENED 1066) — course page with syllabus
Key ideas and source works
- Kahneman, Daniel. Thinking, Fast and Slow. Farrar, Straus and Giroux, 2011. (The dual-process framework Pinker engages throughout.)
- Gigerenzer, Gerd. Rationality for Mortals. Oxford University Press, 2008. (Ecological rationality and fast-and-frugal heuristics.)
- Tversky, Amos and Daniel Kahneman. "Judgment under Uncertainty: Heuristics and Biases." Science, 1974.
- Axelrod, Robert. The Evolution of Cooperation. Basic Books, 1984. (Tit-for-tat and iterated prisoner's dilemma.)
- Stanovich, Keith. The Bias That Divides Us. MIT Press, 2021. (Myside bias, the source for Chapter 10.)
- Tetlock, Philip and Dan Gardner. Superforecasting. Crown, 2015. (Calibrated forecasting connected to signal detection.)
- Pearl, Judea and Dana Mackenzie. The Book of Why. Basic Books, 2018. (Causal inference and DAGs, underlying Chapter 9.)
- Mercier, Hugo and Dan Sperber. The Enigma of Reason. Harvard University Press, 2017. (The argumentative theory of reasoning.)
Reviews and critical discussions
- LessWrong: A review of Steven Pinker's new book on rationality
- New Statesman: Steven Pinker and the problem with rationality
- Psychology Today: Steven Pinker on Rationality
- Statistical Modeling blog (Andrew Gelman): I like Steven Pinker's new book
- Quillette: The Need for Rationality in a Hostile World
Additional chapter summaries and study resources
These are secondary summaries and should be used alongside, rather than instead of, the original book.