Xerox PARC

Contribution to AI

The Alto's real bequest to artificial intelligence was not the desktop metaphor but the premise that one researcher could have an entire machine with a screen worth drawing on. When Warren Teitelman and Daniel Bobrow brought Lisp over from BBN, that premise produced Interlisp-D, in which the structure editor, the debugger, the cross-referencer and the running program all sat resident in a single image, so a knowledge base could be pulled apart while it was still executing. Xerox sold the result as AI workstations — the Dolphin, Dorado and Dandelion — and the system took the ACM Software System Award in 1992 for what it had established about programming environments. Representation research ran alongside it. KRL, published by Bobrow and Winograd in 1977, tried to bolt procedures onto frames; LOOPS later fused rules, objects and Lisp into one language for building expert systems, and that work fed directly into CLOS, the object standard Bobrow helped chair into Common Lisp. Johan de Kleer and John Seely Brown's qualitative physics showed that a program could reason about how a device behaves without any numbers, and de Kleer's assumption-based truth maintenance system let a reasoner hold several contradictory worlds at once, which is what model-based diagnosis of faulty circuits and spacecraft still rests on. The sharpest correction also came from inside: Lucy Suchman's fieldwork on people failing to operate a copier argued that plans guide action rather than govern it, and the planning literature has been answering her since.

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