In the summer of 1654, a French aristocrat with a gambling problem wrote to the mathematician Blaise Pascal. If two players are interrupted in the middle of a game of chance, how should the pot be divided? The intuitive answers contradicted each other. Pascal turned the puzzle over with Pierre de Fermat in a brief and brilliant correspondence, and between them they invented the mathematical theory of probability — the discipline that, three centuries later, would underwrite insurance, quantum mechanics, machine learning, public-health policy, and the entire empirical method of modern science. The gambling problem was solved in a few weeks. The conceptual reorganization took a hundred years.
Modern probability rests on three axioms set out by Andrey Kolmogorov in 1933: probabilities are non-negative numbers; the probability of the entire sample space is 1; and the probabilities of disjoint events add. From this minimum, the rest unspools: conditional probability (the chance of A given that B has happened), independence (when conditioning on B doesn't change A's probability), random variables (numerical functions of outcomes), expectation (the long-run average of a random variable), variance (its spread). Formally, probability is a measure on a sigma-algebra of subsets of the sample space — a connection that lets calculus, integration, and the full machinery of analysis enter the room. Two interpretations of the numbers have coexisted since the eighteenth century. Frequentists read a probability as a long-run relative frequency: an inherent property of a repeatable experiment. Bayesians read it as a degree of belief, a number that an agent updates as evidence arrives. The math is the same; the philosophical commitments differ. For most of the twentieth century, frequentism dominated applied statistics. The Bayesian revival of the past forty years — driven by cheap computation and the realization that many real problems aren't repeatable — has reshaped fields from genetics to AI. The deeper achievement of the theory is that it gave humans a precise language for what they don't know: not just I'm uncertain, but here is the shape of my uncertainty, and here is how it should change in light of evidence.
Probability is now the operating language of empirical science: p-values in clinical trials, posterior distributions in machine learning, Monte Carlo simulation in physics, risk models in finance, probabilistic forecasting in epidemiology and meteorology. The replication crisis in social science was, at root, a probability-literacy crisis — a generation of researchers running null-hypothesis tests they did not deeply understand. Whether ordinary public reasoning ever becomes more probabilistic — whether people learn to think in distributions rather than point estimates — remains an open question of public reasoning, with consequences from elections to pandemics.