Unearthed ELIZA Code Reveals 1960s Chatbot Had Multiple Personas and Context Memory
Updated
Updated · IEEE Spectrum · Jul 20
Unearthed ELIZA Code Reveals 1960s Chatbot Had Multiple Personas and Context Memory
1 articles · Updated · IEEE Spectrum · Jul 20
Summary
Researchers analyzing ELIZA’s original source code from MIT archives found the 1960s system was a broader conversational platform, not just the therapist bot most histories describe.
The code shows ELIZA separated its core engine from scripts, letting it adopt multiple personas with capabilities including script editing, contextual memory and, in later versions, conditional keyword matching.
Doctor—the famous psychotherapist persona—was only one script; others handled small talk, teaching, math, poetry, geography and more, with examples showing ELIZA could tutor students or solve arithmetic in plain language.
The findings, published in the 2026 MIT Press book "Inventing ELIZA," challenge Joseph Weizenbaum’s 10-page 1966 description by showing the implementation was more advanced than later reconstructions suggested.
That reappraisal recasts ELIZA as an early template for modern AI design—persona prompting, plug-in-like architectures and conversational memory—while highlighting how users helped create the illusion of understanding.
If ELIZA's unearthed code rewrites AI history, what other foundational truths might we have wrong?
As AI perfectly simulates empathy, how do we prevent attachment from becoming exploitation?
ELIZA Unveiled: The Surprising Sophistication, Historical Legacy, and Modern Lessons of the Original AI Chatbot
Overview
The recent discovery of ELIZA's original source code has transformed our understanding of early artificial intelligence. Researchers found that Joseph Weizenbaum’s work was much more technically sophisticated than previously believed, revealing advanced features that went far beyond the well-known DOCTOR script. This discovery exposed a significant gap between theoretical models and their real-world implementation, challenging long-held assumptions about ELIZA’s simplicity. Restoring the incomplete code required extensive technical effort, further highlighting the program’s complexity. Overall, the unearthing of ELIZA’s true capabilities has reshaped its historical legacy and deepened appreciation for early AI innovation.