1952: Ivy Mike, and the computers it took to build it
At dawn on 1 November 1952, local time, the United States detonated "Ivy Mike" over Enewetak Atoll in the Pacific — the first full-scale test of a thermonuclear weapon, built on the Teller–Ulam design. It was a feat of physics and engineering, not of artificial intelligence. But the calculations behind it quietly shaped the machines that would later learn to think, and even to play.
Modelling how neutrons multiply and scatter inside a fission-fusion device defied neat equations; the mathematics was too tangled for pen and paper. In 1946, mathematicians Stanislaw Ulam and John von Neumann proposed an alternative: simulate the process thousands of times with random numbers and let statistics reveal the answer. They called it the Monte Carlo method, after the casino town, and first ran it on the ENIAC in 1948 — among the earliest scientific uses of electronic computers for anything other than ballistics tables.
The appetite for ever more Monte Carlo runs, feeding directly into the design verified by Ivy Mike, pushed Los Alamos to build its own stored-program machine: MANIAC I, completed in 1952 on von Neumann's architectural blueprint. It chewed through the neutron-diffusion simulations that the bomb's designers needed.
Once MANIAC was occasionally free of weapons work, its builders turned it toward lighter questions. In 1956, a Los Alamos team programmed it to play chess on a simplified six-by-six board — one of the first times a computer searched through possible moves to choose its own, a direct ancestor of every chess and Go engine since.
That lineage is not a coincidence of place but of method. The same trick — sampling randomly among countless possible futures to estimate which path leads somewhere good — now sits inside modern reinforcement learning and the Monte Carlo Tree Search that let AlphaGo and AlphaZero out-calculate human champions. A statistical shortcut built to model a fireball over the Pacific became, decades later, one of the quiet engines of game-playing AI.