The World's First Chatbot Fooled People into Thinking It Was Human in 1966
In 1966, computer scientist Joseph Weizenbaum created ELIZA at MIT to demonstrate how easily people could be deceived by computers. Using simple rules to rephrase user statements into questions like a Rogerian therapist, ELIZA convinced many users—including Weizenbaum's secretary—that it possessed real empathy, giving birth to the phenomenon known as the ELIZA effect.
An Experiment in Machine Conversation
In the mid-1960s at the Massachusetts Institute of Technology, computer scientist Joseph Weizenbaum developed an interactive program designed to explore how humans communicate with machines. Running on the IBM 7094 time-sharing system and written in a list-processing language called MAD-SLIP, the software was named ELIZA after Eliza Doolittle, the working-class character in George Bernard Shaw's play Pygmalion who is taught to speak like high society. Just as the literary character learned to mimic aristocratic speech without changing her background, Weizenbaum's program learned to mimic conversational patterns without possessing any underlying understanding of the world.
ELIZA operated by parsing natural language inputs typed by a human user at a teleprinter terminal and generating appropriate text responses. Rather than containing a fixed, built-in personality, ELIZA functioned as an engine that executed distinct conversational scripts. Each script contained a set of vocabulary rules and linguistic transformation templates. While several scripts were written, one particular implementation captivated both the public and the scientific community: a script named DOCTOR, which cast the computer in the role of a non-directive psychotherapist.
The Rogerian Disguise
Weizenbaum deliberately chose the persona of a Rogerian psychotherapist to solve a fundamental technical limitation. In Carl Rogers' client-centered therapy, the practitioner avoids offering direct advice, introducing new factual claims, or passing judgment. Instead, the therapist acts as an empathetic sounding board, reflecting the patient's own statements back to them in the form of open-ended questions and prompts to encourage further self-reflection.
This therapeutic approach proved to be an ideal computational shortcut. Because a Rogerian therapist primarily reflects the patient's own statements, the DOCTOR script did not need a vast internal database of real-world knowledge to maintain a plausible dialogue. If a user confessed a personal fear or mentioned a family member, the program did not need to know what fear felt like or understand the societal concept of kinship. It only needed to reflect the user's words back into a grammatically plausible inquiry, placing the burden of maintaining meaning entirely on the human participant.
Decomposition and Reassembly Rules
Under the hood, ELIZA's conversational mechanics relied on a sequence of pattern-matching operations. When a user submitted a sentence, the program scanned the text from left to right, searching for specific keywords that carried assigned priority weights. If the system identified multiple keywords in a single input, it prioritized the one with the highest assigned value, ensuring that emotionally loaded or structurally significant terms took precedence over generic words.
Once a keyword was selected, ELIZA applied a corresponding 'decomposition rule' that segmented the sentence around that keyword. The program swapped pronouns—turning 'my' into 'your', 'I am' into 'you are', and 'me' into 'you'—and dropped the rearranged fragments into pre-written 'reassembly rules'. For example, if a user typed 'I am feeling very depressed today', the program identified 'I am', swapped the pronouns, and inserted the remainder into a template to produce 'Why are you feeling very depressed today?' If no recognizable keyword appeared, ELIZA fell back on generic stock responses such as 'Please go on' or 'Tell me more about that'.
The Rise of the ELIZA Effect
Despite the program's mechanical simplicity, people interacting with ELIZA exhibited an immediate and profound tendency to attribute genuine empathy, consciousness, and comprehension to the machine. Users routinely spent hours conversing with the terminal, confiding intimate personal secrets and emotional struggles. In one well-documented instance, Weizenbaum's own administrative assistant, who had watched him build the software step by step, asked him to leave the room so that she could have a private conversation with the computer.
This psychological phenomenon came to be known as the 'ELIZA effect'—the unconscious tendency of human beings to project human traits, emotional resonance, and deep intentionality onto strings of symbols generated by computers. Because humans are fundamentally social creatures trained to infer communicative intent from language, users read profound understanding into ELIZA's minimal reflections, supplying their own emotional context to fill the vast void of the program's algorithmic emptiness.
Weizenbaum's Alarming Realization
Weizenbaum was deeply unsettled by how readily people accepted the illusion. What he had designed as a technical demonstration of how easily surface-level language could mimic thought was being hailed by some practicing psychiatrists as a serious, scalable prototype for automated psychological treatment. Prominent figures in computing and psychology speculated that automated systems might soon replace human therapists entirely, treating hundreds of patients simultaneously with minimal human oversight.
Horrified by the ethical implications, Weizenbaum shifted his focus from developing such systems to critiquing the uncritical acceptance of artificial intelligence. In his landmark 1976 book, *Computer Power and Human Reason: From Judgment to Calculation*, he argued that there is a critical, inviolable distinction between computational calculation and human judgment. He maintained that qualities such as compassion, empathy, and moral responsibility emerge from genuine human experience and embodiment, and that delegating tasks requiring authentic care to unfeeling machines represents a fundamental abdication of human responsibility.
The Enduring Relevance of 1966
ELIZA marks the foundational milestone in the history of conversational artificial intelligence, serving as the ancestor to every modern virtual assistant, automated customer service agent, and large language model. While modern systems employ deep neural networks and billions of parameters rather than the hand-crafted transformation rules of MAD-SLIP, the underlying psychological dynamic remains largely unchanged.
The ELIZA effect continues to shape human interactions with digital systems today. When conversational interfaces produce fluent, human-sounding text, observers still instinctively assume the presence of internal comprehension, moral agency, or genuine emotion. ELIZA's enduring historical lesson is not merely a triumph of early programming, but a lasting warning about human psychology: that the illusion of machine intelligence is often constructed by the human mind reading between lines that are entirely blank.
Key takeaways
•ELIZA was created in 1966 by Joseph Weizenbaum at MIT to demonstrate how simple pattern-matching rules could mimic conversation without genuine understanding.
•The DOCTOR script adopted the persona of a non-directive Rogerian psychotherapist, allowing it to reflect user statements into questions and hide its lack of real-world knowledge.
•The tendency of people to project human emotions and understanding onto simple computer outputs became known as the 'ELIZA effect'.
•Alarmed by how quickly users and professionals believed ELIZA had real empathy, Weizenbaum became a prominent critic of confusing computational calculation with human judgment.