Six decades after Joseph Weizenbaum built ELIZA at MIT, a new book is finally doing what no prior account managed: reading the actual source code. Inventing ELIZA, co-authored by researchers who excavated the MIT Archives, recovers the original program and surfaces dialogs beyond ELIZA's famous "DOCTOR" persona — correcting a history that has been simplified, mythologized, and in key ways, just plain wrong.

The Code That Was Missing All Along

The conventional story of ELIZA positions it as the earliest chatbot, a program that could simulate a psychotherapist well enough to fool even Weizenbaum's own secretary. That framing has shaped how technologists talk about human-computer interaction for sixty years.

But without the source code, those accounts were always incomplete. What Inventing ELIZA reveals is that there were multiple ELIZAs — different program versions, multiple scripts and personas, built on a series of iterative technical innovations. The singular chatbot origin story is, in fact, plural.

The ELIZA Effect — and Why It Still Matters

The more consequential revelation isn't technical — it's psychological. Weizenbaum himself was disturbed by how quickly users formed emotional attachments to his creation. He described this as "clear evidence that people were conversing with the computer as if it were a person who could be appropriately and usefully addressed in intimate terms."

This observation gave rise to what became known as the "ELIZA effect" — a term that entered online discourse by 1991 but had been observed for decades before that.

Sociologist Sherry Turkle defines it as:

"Our more general tendency to treat responsive computer programs as more intelligent than they really are. Very small amounts of interactivity cause us to project our own complexity onto the undeserving object."

Cognitive scientist Douglas Hofstadter extended this to describe "the susceptibility of people to read far more understanding than is warranted into strings of symbols — especially words — strung together by computers." That framing maps almost perfectly onto how users relate to ChatGPT, Claude, and Gemini today.

Turing, Gender, and the Roots of AI Performance

The book situates ELIZA within a richer intellectual lineage than is usually acknowledged. Alan Turing's original imitation game — the thought experiment behind the Turing test — wasn't initially about machines at all. It was about gender: a man pretending to be a woman, an interrogator trying to determine which was which.

When Turing revised the game to substitute a machine for a man, he entangled artificial intelligence with questions of identity and performance from the start. Weizenbaum extended this by naming his program after Eliza Doolittle, the working-class character in Shaw's Pygmalion who is trained to pass as upper class through linguistic transformation.

"I chose the name 'Eliza' because, like G.B. Shaw's Eliza Doolittle of Pygmalion fame, the program could be taught to 'speak' increasingly well, although, also like Miss Doolittle, it was never quite clear whether or not it became smarter." — Joseph Weizenbaum

This isn't academic decoration. The authors argue that imitation, performance, and identity are structural features of AI systems — not incidental ones.

What Weizenbaum Actually Intended

Critically, Weizenbaum was explicit in his 1966 paper that ELIZA was never meant to pass the Turing test. He wrote that ELIZA's "principal objective" was "the concealment of its lack of understanding" — an exploration of human psychology, not a claim to machine intelligence.

That distinction matters enormously in 2025, when AI companies routinely market capability in terms that blur the line between performance and comprehension. ELIZA was designed to surface the human tendency toward over-attribution. Generative AI systems, whether intentionally or not, are engineered to exploit it.

Implications for Builders and Marketers

For founders and product teams building on top of large language models, the ELIZA effect is a live design variable, not historical trivia:

  • Users will anthropomorphize your AI product faster and more deeply than you expect — regardless of how carefully you disclaim its limitations
  • Emotional attachment forms quickly with minimal interactivity; this affects retention, but also creates ethical responsibilities around dependency
  • Persona design is load-bearing — ELIZA's "DOCTOR" script wasn't cosmetic, it was the mechanism through which people projected intelligence and empathy

The lesson Weizenbaum spent his career trying to communicate — that performing understanding is not the same as having it — is one the current generation of AI developers would do well to revisit.