IntelliCorp

Contribution to AI

Before the shells arrived, building a knowledge-based system meant writing the inference engine as well as the knowledge, so every project began by reconstructing machinery that MYCIN and DENDRAL had already worked out. KEE, first sold in 1983 and grown out of Lisp code originally written at Stanford to model DNA, packaged that machinery for anyone with a purchase order. Knowledge lived in frames — units with slots, and facets on those slots specifying inheritance, expected values and attached procedures — with rules represented as frames themselves, and both forward and backward chaining available. The more unusual feature was worlds: alternative sets of assumptions managed by an assumption-based truth maintenance system, so a programme could carry several incompatible hypotheses simultaneously and see which survived contradiction. Graphical browsers let an engineer inspect and edit the class lattice directly, which mattered because knowledge bases became unreadable well before they became large. KEEconnection joined frames to relational databases, and IBM licensed a version running under MVS, putting the technique where corporate data actually sat. The commercial lesson was harsher than the technical one. Delivery, not knowledge acquisition, was the binding constraint: Lisp machines cost up to $150,000 and served one user, and the late 1980s went into reimplementing the same ideas in C for ordinary hardware as Kappa. The subsequent drift into object modelling and the UML standards effort was the exit route most expert-systems vendors eventually took.

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