Digital Equipment Corporation (DEC)

Contribution to AI

Hardware normally shapes research by accident, but the 36-bit machines shaped it almost by design. The PDP-6, and the PDP-10 that inherited its architecture, used words split into two 18-bit halves, so a Lisp cons cell fitted in a single word and single instructions could pull out its car and its cdr. Lisp had been an awkward guest on IBM hardware; now it had a machine that suited it, and MACLISP, Interlisp and MACSYMA were all built on that foundation. Just as important was the ownership model. A laboratory could buy its own computer and put an operating system of its own devising on it — ITS at MIT, WAITS at Stanford — which meant the machine ran the hours researchers kept and the culture that grew around it was local rather than administered. The earlier PDP-1 at MIT had already established the habit of sitting at a computer rather than submitting to it. The company also supplied the field's most-quoted commercial proof: R1, later XCON, written by John McDermott in OPS5 and configuring VAX orders in production from 1980, grew to roughly 2,500 rules and was reckoned to save tens of millions of dollars a year. Every expert-systems business case of the decade cited it, and its maintenance burden later became the standard argument against rules.

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