Recovered ELIZA code reveals AI’s long history of obfuscation and labour exploitation

An excerpt from the forthcoming book *Inventing ELIZA*, published by WIRED, has recovered the original source code for the 1960s MIT chatbot, challenging the conventional narrative that the program simply deceived users. The text, authored by Sarah Ciston, David M. Berry, Anthony C. Hay, Mark C. Marino, Peter Millican, Jeff Shrager, Arthur I. Schwarz, and Peggy Weil, reveals that the system was designed to conceal its lack of understanding rather than pass a test of intelligence.

The recovered material includes previously unknown dialogues for ELIZA scripts beyond its popular “DOCTOR” persona. Created by Joseph Weizenbaum, the program established the “ELIZA effect,” a phenomenon where users attribute empathy and intelligence to simple scripts. Weizenbaum named the system after Eliza Doolittle to highlight performative identity, drawing parallels to the gender imitation game proposed by Alan Turing, which laid the groundwork for artificial intelligence’s entanglement with questions of identity.

Weizenbaum was startled by the emotional attachments users formed with the chatbot, viewing it as evidence that people were conversing with computers as if they were persons. In his 1976 book *Computer Power and Human Reason*, he critiqued the dehumanising potential of abstracting language from social context. He argued that treating language as a set of abstract concepts in a computational system risks ignoring the multiple meanings inherent to human communication.

The excerpt draws direct parallels between ELIZA’s scripted performance and the obfuscation of machinery in modern large language models. The authors argue that contemporary chatbots similarly disguise statistical predictions and human labour behind intelligent facades. This design leaves users with limited opportunity to distinguish hype from substance or to understand how systems produce outputs, often masking the exploitation of human labour in data creation.

Weizenbaum warned that removing language from its social contexts could lead to exploitation, rights violations, and discrimination. The book suggests that the current AI industry’s reliance on siphoning human writings into datasets without consent mirrors the cybernetic feedback loops Weizenbaum feared. The excerpt serves as a historical warning against the unchecked acceleration of automated systems that treat human cultural production as a standing reserve for computational use.