IBM (700 Series Launch)

Contribution to AI

Before the 700 series, a serious machine was a one-off: ENIAC, Whirlwind, the IAS computer and their cousins each had their own instruction set, and a program written for one ran nowhere else. Building nineteen identical 701s and then a hundred and twenty-three 704s made a program into something transferable — an artefact you could post to another site and expect to work. The 704 customers who founded SHARE in 1955 were swapping routines within months, and researchers acquired a common machine to argue about. The particulars mattered too. Floating-point arithmetic and index registers in hardware, and magnetic core replacing electrostatic storage, made numerical and symbolic work practical on the same box; FORTRAN arrived on it in 1957 and LISP the year after, and the car and cdr that survive in every Lisp descend from the address and decrement fields of a 704 word. Nearly every first attempt ran on these machines: Samuel's checkers player from 1952, later the version that improved by playing itself, Bernstein's chess programme, Rosenblatt's perceptron simulations at Cornell Aeronautical Laboratory, and Gelernter's geometry prover, which proved its first theorem in 1959. IBM was meanwhile careful in public, discouraging salesmen from suggesting the machines could think — so it supplied the field's physical substrate while declining to claim the field, even as Rochester, who designed the 701's arithmetic unit, co-signed the Dartmouth proposal.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).