The Chatbot That Explained Why Users Confide in ChatGPT

Over the past 60 years, ELIZA has left a significant mark on computation and culture, often depicted as the first example of what we now term chatbots, capable of engaging as an automated psychologist. The program, seemingly simple, is famous for “fooling” even the secretary who observed MIT professor Joseph Weizenbaum as he developed it. This is the narrative as it has unfolded.

Yet, within these narratives—despite numerous adaptations across various programming languages and fields of study, in educational settings and pop culture—one crucial element has been overlooked: the source code of the ELIZA program itself. Our new publication, Inventing ELIZA, retrieves this source code from the MIT Archives, providing a detailed analysis and discussion of it for the first time, along with newly discovered dialogs from ELIZA scripts beyond its well-known “DOCTOR” persona.

This research unveiled multiple versions of ELIZA: various program iterations designed to run diverse scripts or personas, developed through a range of technical innovations. Inventing ELIZA aims to rectify and deepen the understanding of ELIZA’s history and impact by delving into the misunderstandings, numerous revisions, and the missing code associated with it. In this excerpt, we explore one of ELIZA’s earliest exchanges, how it established the groundwork for human interactions with computers for years, and how this intricate program continues to resonate with the unbridled ambition of today’s AI landscape.

All men are similar.

IN WHAT RESPECT

They constantly nag us about various things.

CAN YOU PROVIDE A SPECIFIC INSTANCE

Well, my boyfriend insisted I come here.

YOUR BOYFRIEND INSISTED YOU COME HERE

He claims I’m often depressed.

I AM SORRY TO LEARN YOU ARE DEPRESSED

This dialog has been reproduced numerous times, inspiring programmers and authors to create many subsequent chatbots. However, upon closer examination of this exchange, more questions emerge: Who was this young woman? Was she a real person or merely a construct of ELIZA’s creator, Joseph Weizenbaum? How precisely did the ELIZA system formulate its responses, and to what extent were they altered? What made the system so effective in engaging individuals?

ELIZA, along with her “DOCTOR” persona, prompted a new way of thinking and raised concerns about human relationships with computers. Weizenbaum discussed this in his 1976 work Computer Power and Human Reason, bringing in philosophical, social, and political critiques. The distinct interaction presented by his program showed that emerging human-computer relationships would have significant repercussions, which he sought to explore and challenge. Witnessing its public reception, Weizenbaum was taken aback by how swiftly and emotionally people connected with ELIZA, interpreting it as “clear evidence that individuals interacted with the computer as if it were a person deserving of intimate communication.” The inclination to project empathy and infuse personal emotions into a computer mystified Weizenbaum. He was troubled by how much people associated rationality with computation, attributing understanding and intelligence to computer systems where none existed.

This phenomenon became known as the “ELIZA effect.” By 1991, the term began emerging in online discussions, but its usage had been around for decades prior. Sociologist Sherry Turkle defines “the ELIZA effect” as “our tendency to treat responsive computer programs as more intelligent than they actually are. Minimal interactivity leads us to project our own complexity onto a mere object.” Cognitive and computer scientist Douglas Hofstadter describes it as “people’s inclination to read far more understanding into strings of symbols—especially words—put together by computers than is justified,” which is readily applicable to today’s generative AI systems.

To appreciate the influence and challenge posed by ELIZA, we can refer to the famous dilemma posed by computer scientist Alan Turing in his essay “Computing Machinery and Intelligence,” where Turing questioned, “Can Machines Think?” His thought experiment was based on a parlor game not concerned with technology but with gender: A man and a woman are concealed in separate rooms, and an interrogator tries to determine their genders through a series of questions. The man attempts to deceive the interrogator by pretending to be a woman, while the woman strives to convince the interrogator of the “correct” information. Both assert they are the “real” woman, presenting a challenge to essentialist concepts of gender.