A crucial piece of artificial intelligence history has finally been recovered. The original source code of ELIZA, the famous chatbot developed by Professor Joseph Weizenbaum at MIT in the 1960s, has resurfaced from the university’s archives thanks to the work of the authors of the book Inventing ELIZA. For decades, researchers had access only to partial descriptions and transcriptions of conversations, but now the original program is available for in-depth analysis, offering new insights into the birth of human-machine interaction.
The discovery of the code in MIT archives
The book Inventing ELIZA presents for the first time a detailed reading of the source code, which had remained inaccessible for over fifty years. The discovery has made it possible to understand the multiple versions of ELIZA, each designed to run different scripts or personas. Among them, the most famous is the “DOCTOR” persona, a virtual therapist capable of simulating a psychological conversation. The finding revealed how Weizenbaum developed a series of technical innovations to make the program appear more intelligent than it actually was.
Sponsored Protocol
How ELIZA foreshadowed modern chatbots
The most famous dialogue of ELIZA, in which a young woman confides her depression to the virtual therapist, has inspired generations of programmers and writers. However, the story behind that dialogue is more complex. Weizenbaum was startled by how quickly people formed emotional attachments to the machine, a phenomenon he himself described as “clear evidence that people were conversing with the computer as if it were a real person.” This tendency, known as the “ELIZA effect,” is now at the core of how chatbots like ChatGPT function, often leading users to overestimate their intelligence. Connecting ELIZA to modern AI assistants is also the concept of performative identity: the name itself is a tribute to Eliza Doolittle, the protagonist of George Bernard Shaw’s Pygmalion, who learns to speak like an aristocrat. Similarly, ELIZA “learns” to play a role through repetitive linguistic patterns, without possessing true understanding.
Sponsored Protocol
The ELIZA effect and lessons for today’s AI
The term “ELIZA effect” was coined to describe the human tendency to attribute intelligence to programs that are actually devoid of it. Sociologist Sherry Turkle defines it as “our general tendency to treat responsive computer programs as more intelligent than they really are.” This lesson is more relevant than ever today, with the explosion of generative models. Some modern tools, like the AI spending dashboard launched by 1Password, allow tracking token usage on platforms such as Anthropic, Cursor, and OpenAI, but the illusion of understanding persists. The recovery of ELIZA’s code not only sheds light on the field’s origins but also serves as a warning: behind every seemingly empathetic chatbot lies a set of mechanical rules. As Weizenbaum wrote, “ELIZA thraws away most of its inputs” and its principal objective is to conceal its lack of understanding. A caution that resonates today as the AI industry pushes toward ever more sophisticated models.
Sponsored Protocol
Source: https://www.wired.com/story/inventing-eliza-book-excerpt-chatbot
