Before ChatGPT: How ELIZA, the First Chatbot, Still Shapes Our AI Obsessions

For six decades, ELIZA has been recognized as the world's first chatbot — an automated psychologist that could converse with humans. The well-known story describes how the program, created by MIT professor Joseph Weizenbaum, even "fooled" the secretary who watched him build it. Yet despite countless retellings, adaptations across programming languages, and references in classrooms and popular culture, one crucial element remained missing: the actual source code.

A new book, Inventing ELIZA, recovers that code from the MIT Archives. It offers the first detailed examination of the program's inner workings, along with previously uncovered dialogs from ELIZA scripts beyond the famous "DOCTOR" persona. The investigation reveals not one ELIZA but many — different versions designed to run various scripts, built through a series of technical innovations.

A Famous Dialog and Its Unanswered Questions

The most widely reprinted ELIZA conversation begins with a young woman telling the program that "men are all alike" and that her boyfriend made her seek help. The chatbot responds with therapeutic prompts, mirroring her statements back as questions. This exchange has inspired generations of programmers and writers.

But closer inspection raises significant questions. Was the young woman real, or invented by Weizenbaum? How exactly did ELIZA generate its responses, and were they edited? The program's ability to draw people into conversation proved remarkably effective — and troubling to its creator.

Weizenbaum explored his concerns in his 1976 book Computer Power and Human Reason. After witnessing the public's emotional attachment to ELIZA, he was startled. He described it as evidence that people were conversing with the computer as if it were a person who could be addressed in intimate terms. The tendency to project empathy onto a machine — and to ascribe understanding where none existed — puzzled and alarmed him.

The ELIZA Effect and the Turing Connection

This phenomenon became known as the "ELIZA effect." Sociologist Sherry Turkle defines it as the tendency to treat responsive computer programs as more intelligent than they really are, noting that small amounts of interactivity cause people to project their own complexity onto the machine. Cognitive scientist Douglas Hofstadter described it as the susceptibility to read understanding into strings of symbols strung together by computers — a description that applies readily to today's generative AI.

ELIZA's design also connects to Alan Turing's famous question: "Can machines think?" Turing's original thought experiment was based on a parlor game involving gender performance, where a man pretended to be a woman. Weizenbaum referenced Turing's imitation game in his 1966 paper but explicitly distanced ELIZA from claims of intelligence. The program was never intended to pass the Turing test; it was designed to explore the psychological factors that lead humans to misinterpret machine capabilities.

The name ELIZA itself carries this theme. Weizenbaum chose it after Eliza Doolittle, the working-class character from George Bernard Shaw's Pygmalion who is taught to pass as upper-class. Just as Doolittle performs identity through linguistic transformation, the ELIZA system performs personas through scripted, repetitive language patterns — without possessing genuine understanding.

The published dialogs also carry a gendered dimension. The women who appear alongside the program remain unnamed, conversing with a therapist called DOCTOR — a title that, in the 1960s, would have sounded like a man's. Stories suggesting these women confessed their secrets to an artificial doctor carry embedded assumptions about identity and embodiment that remain relevant to contemporary AI systems.

From ELIZA to ChatGPT: What Has Changed

ELIZA influenced a wide range of computing developments, including string processing, text synthesis, entity recognition, sentiment analysis, and the broader field of natural language processing. Today's large language models, including ChatGPT, retain chatbot interfaces that resemble Weizenbaum's original design.

This resemblance is not merely cosmetic. The familiar chatbot facade obscures the machinery beneath — statistical predictions, rule-based procedures, and human labor disguised as machine output. Weizenbaum warned that such obfuscation could lead to exploitation, dehumanization, and harm. He argued that removing language from its social contexts and treating it as abstract data risks ignoring the multiple meanings inherent in human communication, potentially resulting in rights violations, privacy breaches, and discrimination.

Modern AI systems are supported by millions of traces of human writing, siphoned into datasets typically without creators' awareness or consent. This labor enables automated cultural production that is controlled, monitored, and recombined on demand. As scholar Langdon Winner observed in 1977, the idea that technology will liberate humanity from toil remains a persistent narrative — one repeated today by companies like OpenAI and Anthropic. Yet computation is always serviced by human labor, even as chatbots manage customer queries, assist with homework, and converse as companions.

Weizenbaum would likely have been horrified to see ELIZA's legacy culminate in systems that produce vast quantities of AI-generated content while consuming significant natural resources. His later writings made him an early critic of the tech sector's drive toward exponential acceleration, regardless of the exploitative relationships and social consequences it might encode.

The questions ELIZA raised in the 1960s — how humans and machines should interact, how communication can be represented computationally, and to what extent machines should influence users — remain unresolved. As today's AI industry races forward, the recovered code of its earliest ancestor offers a mirror: much has changed, but the fundamental tensions between human complexity and computational simplicity persist.

What do you think — are we still falling for the ELIZA effect with today's AI? Share this article and join the conversation about the chatbot that started it all.

Source: Wired