Fooled by Randomness: Complete Glossary of Key Concepts
Fooled by Randomness is dense with concepts that don't have obvious names — ideas Taleb is assembling for the first time, or applying to domains where they haven't appeared before. This glossary collects every major term, with definitions short enough to actually use.
The concepts are organized into two groups: the statistical and structural ideas that explain how randomness works, and the cognitive and behavioral ideas that explain why humans systematically misread it.
The Shape of Randomness
These terms describe how random processes actually work — what they produce, what they hide, and how they mislead.
Alternative Histories — The full distribution of outcomes that could have resulted from a decision, not just the one that did. Any realized outcome is one draw from the distribution. Evaluating a decision by its outcome alone ignores the alternatives that were equally possible.
Survivorship Bias — The error of drawing conclusions from a sample that includes only winners, because the losers have dropped out and become invisible. The surviving companies, traders, and strategies look better than the full initial cohort because the failures aren't in the data anymore.
Ergodicity — The property of a random process where time averages and ensemble averages converge. For performance evaluation: ergodic processes absorb lucky runs over time, which is why "Masters of the Universe" consistently end up in diminished circumstances after their lucky sequences exhaust.
Regression to the Mean — The tendency for extreme outcomes to be followed by less extreme ones. Since extreme observations oversample from favorable variance, subsequent independent draws from the same distribution trend back toward average. There is no jinx — only arithmetic.
Skewness — The asymmetry of a distribution's tails. Negative skewness means rare, large losses. Many strategies with smooth, consistent returns are negatively skewed — they win frequently in small amounts and lose rarely in catastrophic ones. Win rate tells you nothing about expected value without knowing the skew.
The Rare Event (Peso Problem) — The systematic underestimation of low-probability, high-impact events. Calm periods make the rare event seem more unlikely and encourage exposure buildup — which makes the eventual event more damaging when it arrives.
Path Dependence — The property where outcomes depend on the sequence of events, not just current conditions. Small early differences in luck, timing, or visibility compound through feedback loops into large final differences. QWERTY, Microsoft, and career trajectories are all path-dependent.
Nonlinearity — In nonlinear systems, small differences in input produce large differences in output. Winner-take-all market structures are nonlinear: marginally better-positioned players capture disproportionate outcomes. Early luck that tips the position has outsized influence.
The Cognitive Dimension
These terms describe why human cognition systematically misreads the statistical patterns above.
Probability Blindness — The cognitive inability to experience a probability distribution. The brain visualizes one scenario at a time, not the probability-weighted mixture of all outcomes. Decisions made "intuitively" about probabilistic situations are actually scenario-based, which systematically distorts the weighting.
Attribution Bias — The tendency to attribute successes to skill and failures to luck. In domains with meaningful randomness, this breaks feedback loops: the model that produced a good outcome doesn't get updated because it's assumed to have been correct; the model that produced a bad outcome doesn't get updated because the failure is attributed to circumstances.
False Causality (Gambler's Ticks) — The automatic association the brain forms between any salient antecedent and a subsequent positive outcome. Like Skinner's pigeons, traders develop rituals connected to favorable outcomes through pure conjunction. The rituals persist because the Skinnerian mechanism is hardware-level and doesn't respond to knowing about it.
Wittgenstein's Ruler — When you use a ruler to measure something, you're simultaneously using the thing to measure the ruler. If the ruler is poorly calibrated, the measurement mostly tells you about the ruler. Applied to information sources: a low-credibility source's claim mostly tells you about the source, not the subject.
The Statistical Tools
Monte Carlo Simulation — A method that generates large numbers of random paths from a probability model to estimate the distribution of outcomes. As a thinking tool, it shows how many "brilliant" track records can be produced by luck alone — establishing the null hypothesis before evaluating any specific performance.
Noise vs. Signal — Signal is real information; noise is random variation. The brain can't distinguish them in real time. Critically, the ratio of signal to noise depends on how often you look: checking investments daily produces almost entirely noise; checking annually produces mostly signal from the same underlying portfolio.
The Philosophical Foundations
The Problem of Induction — No finite set of confirming observations proves a universal rule. One disconfirming observation falsifies it. Taleb's application: track records of "safety" don't prove safety. They prove the catastrophic event hasn't appeared in the sample yet.
Popperian Asymmetry — The logical asymmetry that one disconfirming observation falsifies a theory, while infinite confirming observations cannot prove it. In investment terms: state your exit condition before entering a position. Pre-commitment forces falsifiability; post-hoc reasoning defeats it.
The Pascal Principle — When outcomes involve catastrophic, irreversible magnitude, expected-value reasoning breaks down. The arithmetic of expected value assumes repeated play; ruin ends the series. Do not risk ruin for any finite gain, regardless of the probability.
For the full framework connecting these concepts, start with: - Fooled by Randomness: How Luck Masquerades as Skill - Living With Randomness