ELIZA's rediscovered source code rewrites chatbot history

The kindly Rogerian therapist that lives in every AI history book turns out to have been running on a much more elaborate machine than the 1966 paper let on. IEEE Spectrum's adaptation of a forthcoming MIT Press book reports that the original ELIZA source code, unearthed from MIT's archives after decades untouched, shows Joseph Weizenbaum's program was architecturally richer than the 10-page paper he published in the January 1966 Communications of the ACM ever documented.

The piece, adapted from "Inventing ELIZA: How the First Chatbot Shaped the Future of AI" (MIT Press, 2026), argues that ELIZA "was not merely a simple pattern-matching chatbot but can be better understood as a sophisticated platform designed for multiple 'personas,' or scripts, with a complex set of capabilities, including script editing and contextual memory." Beyond the famous Doctor script that mimicked Rogerian therapy, the recovered materials describe scripts for mathematics, poetry, color, paradoxes, synchronization, and relativity. Edwin F. Taylor at MIT's Education Research Center built teaching scripts including Intrvw, Canvec, FVP1, and Arithm, using what the authors call conditional keyword matching so the system could branch on what had already been discussed.

Why this matters if you do not care about 1960s computing history: the popular narrative around the "ELIZA effect," the fallacy that a trivial program could seem sentient, has shaped how a generation of critics talks about modern language models. If the founding case study was itself misunderstood as pure pattern matching when it was actually a scripting platform with contextual memory, then the shorthand lesson drawn from it, that these systems are simpler than they look, needs a footnote.

The honest caveat is that the reporting does not tell you when the code was found, who located it, or the programming language it was written in, and it does not give a line-by-line comparison with the 1966 paper. What it does give you is a reason to treat the primary source, not the received wisdom, as the reference. For historians, educators, and anyone teaching the origins of human-computer interaction, that is the point worth taking away.