1894: Aleksandr Khinchin is born

On 19 July 1894, Aleksandr Yakovlevich Khinchin was born in Kondrovo, in the Kaluga region of the Russian Empire. In a career shaped by revolution and war, he became one of the principal architects of modern probability theory. The concepts he helped make precise—laws of large numbers, stationary processes, and entropy—later migrated, almost unnoticed, into the foundations of machine learning and artificial intelligence.

Working closely with Andrey Kolmogorov in the 1920s and 1930s, Khinchin supplied rigorous proofs that turned probability from a collection of clever techniques into a coherent mathematical discipline. He established broad conditions under which the law of large numbers holds and proved the law of the iterated logarithm, which describes the fine-scale growth of random fluctuations. He also laid groundwork for the theory of stationary stochastic processes—models of signals whose statistical character does not change with time. These ideas gave engineers a language for noise, uncertainty, and long-run averages.

Khinchin’s reach extended into information theory. In 1953 he published an axiomatic characterization showing that Shannon entropy is essentially the unique measure satisfying a short list of natural properties. That same entropy function reappears everywhere in contemporary AI: as the cross-entropy loss that trains neural networks, as a splitting criterion in decision trees, and inside the variational objectives of generative models.

The route from Khinchin’s theorems to today’s systems is indirect yet unmistakable. Early artificial intelligence leaned heavily on logic and symbolic rules. From the 1980s onward, statistical methods became dominant precisely because they could absorb the messiness of real data. Whenever a practitioner invokes a concentration inequality, samples a Markov chain, or minimizes a Kullback–Leibler divergence, the intellectual ground cleared by Khinchin and his contemporaries is being used.

Even everyday vocabulary reveals the debt. Terms such as stationarity, ergodicity, and entropy rate passed from the probability literature he helped create into the working lexicon of machine-learning researchers. On this day we therefore mark not a dazzling robot demonstration, but the birth of a mathematician whose insistence on rigor quietly equipped later generations to build systems that learn under uncertainty—the central predicament of modern AI.