PERSON
Andrey Kolmogorov
The Russian mathematician who, in a slim 1933 monograph, put probability on three axioms and thereby gave statistical machine learning its grammar—and who, three decades later, asked how much information a single object contains, answering that the measure is the length of its shortest description and thus supplying the theoretical foundation for the claim that compression is comprehension.
Andrey Kolmogorov is the silent grammar of artificial intelligence. His 1933 axiomatisation of probability theory—three rules, a slim German monograph, the transformation of a floating philosophical argument into a rigorous mathematical structure—is the foundation on which every loss function, every posterior update, every softmax output in every
large language model rests, whether its engineers cite him or not. You cannot speak coherently about what a learning system does without speaking his language, and most practitioners speak it the way most of us speak prose: fluently, and without knowing the name of the man who set the rules. His second revolution arrived in the 1960s, when he turned from probability to the question of how much information a single object contains—not a stream of messages, not a statistical ensemble, but one string, one number, one thing. His answer was