The History of AI

Grundbegriffe der Wahrscheinlichkeitsrechnung

Era 1 · The foundations (1763–1942) › Probability and statistics

Lineage
Data and learning
Kind
Publication

Who

Moscow State University

What happened

Andrey Kolmogorov, a professor at Moscow University, publishes Grundbegriffe der Wahrscheinlichkeitsrechnung with Springer in Berlin, a book of about sixty pages that grounds probability on a handful of axioms.

Why it mattered

It defines probability as a measure on a set of possible events. From then on, probability is one more branch of mathematics, with the same foundation that statistics and machine learning use.

The full story

At the start of the twentieth century, probability was used in physics, in insurance, and in statistics, but it had no definition everyone accepted. Laplace's, favorable cases over possible cases, worked only when all the cases were equally likely. In 1900 Hilbert had included the task of giving probability a set of axioms in the sixth of the 1900 problems, and several mathematicians had tried without winning general acceptance.

Andrey Kolmogorov, thirty years old and teaching in Moscow, published the proposal that ended up being accepted in 1933. The central idea is to treat probability as a measure, just like length or area. There is a set of possible outcomes; events are subsets of it; and probability assigns each event a number between 0 and 1, so that the certain event gets 1 and the probability of several mutually exclusive events is the sum of their separate probabilities. With that and an axiom of continuity, the rest of the theory follows.

The book also gave precise definitions of conditional probability and expectation, and put random processes, such as Markov chains, on a firm footing. Glenn Shafer and Vladimir Vovk, who studied its sources, stress that almost all the pieces existed beforehand; what was new was assembling them into a short, complete system.

The English translation, Foundations of the Theory of Probability, came out in 1950. Ever since, it has been the starting point of probability textbooks, and measure theory is the language in which the results of statistical learning are written.

Under the hood

There is a set Ω of possible outcomes and a family of its subsets, the events. A probability P assigns a number to each event and satisfies three conditions: P(A) ≥ 0; P(Ω) = 1; and if A₁, A₂, … have no elements in common, the probability of their union is the sum of their probabilities.

Conditional probability is defined as P(A|B) = P(A ∩ B) / P(B), and Bayes's theorem follows from that definition in two lines. A random variable is a function on Ω, and its expectation is an integral with respect to P.

Builds on

  • Hilbert's 23 problems (August 8, 1900)

    Inherited, with no proven direct link

    The sixth problem asks for axioms for probability; Kolmogorov's are the accepted answer

  • Théorie analytique des probabilités (1812)

    Inherited, with no proven direct link

    The axioms replace Laplace's classical definition, favorable cases over possible cases, with a measure on a set of events

Sources

How to cite

“Grundbegriffe der Wahrscheinlichkeitsrechnung.” The History of AI, revised November 13, 2026. https://thehistoryofai.com/en/milestone/kolmogorov-axioms

This entry was last reviewed on November 13, 2026.

Open in the interactive timeline