Inference Corporation

Contribution to AI

The knowledge-engineering boom needed tooling, and ART was one of the handful of products that fixed what a commercial expert-system shell looked like. It extended OPS5's forward-chaining production rules — matched by the Rete algorithm — with backward chaining, frame-style structures and a truth-maintenance layer whose viewpoints let a knowledge base hold several mutually inconsistent hypothetical worlds at once and withdraw conclusions when the assumptions behind them failed. It ran on Lisp machines and was priced for corporate buyers, which produced its most durable consequence indirectly: NASA's Johnson Space Center, wanting that style of programming in C on ordinary hardware, began CLIPS in 1985 with its syntax deliberately modelled on a subset of ART's and written without access to Inference's code. CLIPS was distributed at nominal cost, taught widely, and its descendants — Jess among them — carried that parenthesised rule notation into business rules engines still running today. The second act counted for as much. CBR Express took case-based reasoning, until then largely an academic argument about memory and analogy pursued at Yale and Georgia Tech, and made it a help-desk product; Compaq's SMART system and the QuickSource printer diagnostic shipped to customers were both built on it. A library of past cases proved cheaper to maintain than a rule base, which is why customer support outlasted the expert-systems bust as symbolic AI's steadiest commercial market.

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