"Inventing ELIZA" Recovers Lost Chatbot Source Code
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- "Inventing ELIZA" recovers the original ELIZA source code from MIT Archives for the first time, offering close readings of the code alongside newly uncovered dialogs for scripts beyond the program's popular "DOCTOR" persona
- Joseph Weizenbaum created ELIZA at MIT in the 1960s and was startled by users' emotional attachments, writing that it demonstrated people were "conversing with the computer as if it were a person who could be appropriately and usefully addressed in intimate terms"
- The "ELIZA effect" — defined by sociologist Sherry Turkle as "our more general tendency to treat responsive computer programs as more intelligent than they really are" — was already visible in Weizenbaum's era and applies directly to today's generative AI, per cognitive scientist Douglas Hofstadter
- The famous "Men are all alike" dialog has been reprinted for decades, but the book notes the unnamed woman in it may have been Weizenbaum's invention rather than a real user, and questions how heavily the program's responses were edited
- Alan Turing's original imitation game was structured around gender — a man pretending to be a woman — and Weizenbaum named his system after Eliza Doolittle from Shaw's Pygmalion, deliberately continuing themes of performative identity rather than attempting to pass a Turing test
- ELIZA was explicitly not designed to pass the Turing test, per Weizenbaum's 1966 paper; the system aimed instead to explore the psychological factors that lead humans to misinterpret a computer's capabilities
- The women who converse with the "DOCTOR" persona remain unnamed in published accounts, while "DOCTOR" — a title that sounded masculine in the 1960s — establishes a gendered dynamic the book argues persists in how we frame AI interactions today
Why it matters: By recovering source code lost for decades, the book reframes ELIZA not as a quaint chatbot origin story but as a deliberate 1966 probe into why humans project empathy onto machines — and shows the program's gender-laden design choices, from its Doolittle namesake to its unnamed female users, still echo in how AI companies market and frame conversational agents today.
