The Confessional With No One Inside
In 2021 Jeff Shrager unearthed a 400‑line MAD‑SLIP source file among Weizenbaum’s papers at MIT, and an eight‑person “Team ELIZA” spent four years decoding it. Their work shows that ELIZA was a generic script interpreter, not a single “psychotherapist” bot, and that the iconic DOCTOR dialogue was only one of several scripts—including physics lessons and poetry recitations. The team also discovered that the original code used deterministic hashing rather than true randomness, and that sentence parsing included the conjunction “but,” a detail omitted from Weizenbaum’s 1966 paper. By pairing the recovered code with a rebuilt IBM 7094 time‑sharing environment—thanks to programmer Rupert Lane—they produced a faithful emulation that reproduces the historic conversation line‑for‑line, complete with a decision‑trace UI now available in a browser.
The revelation arrives at a moment when large language models dominate public imagination and debates about machine authority. ELIZA’s mythic status has long been invoked to illustrate how users anthropomorphize simple pattern‑matching systems, yet the new findings demonstrate that early AI researchers already built modular, script‑driven architectures that could be repurposed for diverse tasks. This nuance challenges the prevailing narrative that early chatbots were merely gimmicks, and it underscores a continuity between Weizenbaum’s interpreter and today’s plug‑in ecosystems for LLMs. Moreover, the project exemplifies “critical code studies,” blending technical reconstruction with cultural commentary—a methodology that could reshape how historians evaluate the provenance and impact of foundational software.
Looking ahead, the ELISA reconstruction invites scrutiny of other forgotten programs that may have shaped AI’s conceptual lineage. By exposing the exact mechanisms that gave ELIZA its persuasive veneer, the work fuels renewed calls for transparency in modern conversational agents, where opaque neural weights replace deterministic scripts. Researchers and policymakers should monitor whether similar archival digs surface additional “hidden scripts” that could inform safety guidelines or inspire new, lightweight chatbot frameworks. The project also signals a growing appetite for scholarly editions of historic code, suggesting that software archaeology may become a regular fixture in AI ethics and history discourse.
Key Takeaways
The recovered MAD‑SLIP source confirms ELIZA functioned as a multi‑script interpreter, overturning the long‑held belief that it was solely a Rogerian therapist simulation.
Deterministic hashing and a “but” parsing rule, absent from Weizenbaum’s original paper, are essential to reproducing the classic dialogue.
Team ELIZA’s emulation on a recreated IBM 7094 demonstrates that authentic execution of 1960s code is feasible with modern reconstruction techniques.
The findings highlight the need for deeper transparency in today’s AI systems, as early chatbot design already leveraged modular scripts to shape user perception.
About the Source
This analysis is based on reporting by HackerNoon. Here is a short excerpt for context:
What the first chatbot’s recovered source code reveals about why we trust machinesRead the original at HackerNoon