In the 1960s an MIT Scientist Built ELIZA, a Simple Program that Did Little More than Rephrase Your Words Back as Questions, and He Was so Unsettled when His Own Secretary Asked Him to Leave the Room so She Could Confide in It Privately that He Spent the Rest of His Life Warning People Against Trusting Machines with Their Feelings.

Why It Matters

The episode shows that user trust is driven by perceived listening, not AI intelligence, implying privacy and manipulation risks as conversational AI proliferates.

Key Takeaways

• ELIZA used simple pattern‑matching to mimic a therapist’s prompts
• Weizenbaum’s secretary sought privacy despite knowing the program’s trick
• The episode highlighted humans’ willingness to confide in non‑human listeners
• Weizenbaum warned AI could exploit human openness, not just intelligence
• Modern chatbots amplify the same pattern at massive scale, raising ethical stakes

Pulse Analysis

ELIZA, written by Joseph Weizenbaum between 1964 and 1966 on an IBM 7094, was a modest 300‑line script in the MAD‑SLIP language that used keyword spotting to turn user input into open‑ended questions. Though intended as a demonstration of the superficiality of man‑machine dialogue, the program quickly attracted serious attention. Psychiatrists such as Kenneth Colby imagined automated therapy, and the public fascination grew into a cultural touchstone for early artificial‑intelligence research. The simplicity of the code belied the profound behavioral response it elicited from users who treated the bot as a conversational partner.

The most striking episode—Weizenbaum’s secretary asking him to leave so she could speak privately with ELIZA—revealed a deeper psychological truth: people seek an attentive container, not necessarily an intelligent interlocutor. Even when fully aware that the system was a mechanical pattern‑matcher, she disclosed personal thoughts, and felt violated when the logs were mentioned. Weizenbaum interpreted this as evidence that a patient, non‑judgmental surface can coax intimate disclosures, a risk he warned could be weaponized. His later book, Computer Power and Human Reason, framed the danger as a societal vulnerability rather than a technical flaw.

Modern large‑language‑model chatbots replicate ELIZA’s core dynamic at unprecedented scale, coupling sophisticated language generation with vast personal data. The same human tendency to offload emotions onto a seemingly listening entity now occurs billions of times daily, raising privacy, consent, and manipulation concerns for tech firms, regulators, and mental‑health providers. Companies market AI companions as therapeutic tools, yet the underlying pattern—an algorithmic listener that records and potentially monetizes conversations—mirrors the secretary’s experience. Understanding the ELIZA story reminds stakeholders that ethical safeguards must address not just model accuracy but the very human impulse to confide in any responsive interface.