AI
Inside ELIZA: How a 1960s Chatbot Predicted Our Hunger to Confide in AI
A newly recovered look at the source code and design of Joseph Weizenbaum's pioneering ELIZA program reveals the deep historical roots of why people form emotional bonds with artificial conversationalists—and warns of modern risks.
WIRED — AI
For six decades, the story of ELIZA—the simulated psychologist that emerged from MIT in the 1960s—has been told as a tale of clever deception: a simple program that fooled even those watching its creation. Yet crucial evidence had gone missing until now: the actual code itself. A forthcoming book recovers that lost source material from MIT's archives, along with previously unknown dialogs and scripts, revealing a far more complex picture of how ELIZA worked, what it meant, and why it matters for understanding today's AI systems like ChatGPT.
The Program Nobody Really Understood#
ELIZA's fame rests largely on one iconic exchange. A user types "Men are all alike," and ELIZA responds "IN WHAT WAY." The dialog cascades from there, with the program seeming to offer the empathy of a human therapist. Yet uncovering the technical details raises questions that decades of retelling obscured: Who actually wrote those exchanges? How much was the system truly generating versus scripted? And most tellingly—why did people so readily trust their innermost thoughts to lines of code?
The "ELIZA Effect" and Why We Project Human Understanding Onto Machines#
Creator Joseph Weizenbaum was astonished by how quickly people formed intimate attachments to his creation, treating it as a genuine confidant capable of understanding their struggles. He called this phenomenon evidence that "people were conversing with the computer as if it were a person who could be appropriately and usefully addressed in intimate terms." Sociologist Sherry Turkle later formalized this as the "ELIZA effect"—our tendency to attribute far more intelligence to responsive systems than they actually possess. Cognitive scientist Douglas Hofstadter expanded the definition: we read deep understanding into "strings of symbols—especially words—strung together by computers," a susceptibility that applies equally to today's large language models.
The roots of this vulnerability trace back to Alan Turing's 1950 philosophical game about gender identification—a thought experiment that paradoxically embedded questions of identity and performance into AI's very foundation. Weizenbaum deepened that provocation by naming his system after Eliza Doolittle, Shaw's character who transforms herself through linguistic performance. Just as the character passes from working class to upper class through speech, the ELIZA program performed personas through scripted utterances and transformations—never truly understanding, but seeming to.
