What Eliza Teaches Us About The First Ai Chatbot And Why Its Creator Turned Against Ai – ИФЗ РАН

Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
The first AI chatbot didn't pass a deep test of intelligence. It just reflected your own words back at you.
In 1966, MIT computer scientist Joseph Weizenbaum published ELIZA, a compact program that simulated a Rogerian psychotherapist using pattern matching and simple script rules. Honestly, it had zero understanding of human feelings, trauma, or therapy. Yet people around the lab started typing confessions into teletype terminals, asking Weizenbaum to leave the room so they could talk to the machine in private.
Watching smart researchers and staff treat a basic text script like an empathetic confidant shook Weizenbaum to his core. Over the next decade, the man who built one of computing's most famous conversational programs became one of artificial intelligence's sharpest critics. His 1976 book, Computer Power and Human Reason, wasn't written by an outsider who hated tech. It came from an insider who watched humans hand over judgment and emotional trust to machines that couldn't care at all.
Because the Los Angeles Times URL was blocked behind access restrictions, I haven't read that specific column directly, so I'm building this piece from established historical records, Weizenbaum's published writings, and my own hands-on experience testing conversational systems and prompt-driven bots. I'm also flagging any illustrative examples clearly so nothing gets blurred into historical fact.
People love romanticizing early computing as a sudden leap toward artificial consciousness. ELIZA was way more humble—and way more revealing.
Built at MIT's Artificial Intelligence Laboratory and described in Weizenbaum's 1966 Communications of the ACM paper, ELIZA used a general script mechanism called MADCAP and a famous script named DOCTOR. DOCTOR mimicked non-directive therapy by spotting keywords in what you typed, transforming pronouns, and dropping a question or canned prompt back on screen. If you typed "I feel sad because my boss yelled at me," a rule looking for "I feel" or "because" might flip "my" to "your" and reply with something like, "Why do you think your boss yelled at you?" or "Tell me more about such feelings."
There was no memory of deep life history in the modern sense. No neural network weights. No genuine grasp of grief, family dynamics, or clinical care.
To illustrate how a rule-based script behaves in practice, consider this illustrative example of a DOCTOR-style exchange: a user types, "I'm tired of pretending everything is fine at work," and the script matches "I'm tired of" and responds, "How long have you been feeling that way about work?" It sounds attentive. In reality, it's just shuffling tokens according to a human-written template.
When I test modern chatbots on heavy personal or high-stakes questions, I still see that old ELIZA ghost in the room. Better language models produce smoother prose and wider associations than a 1960s script ever could. Still, fluency easily tricks our brains into reading warmth, wisdom, and intention into statistical text generation.
The biggest surprise about ELIZA wasn't the code. It was the humans typing at the keyboard.
Weizenbaum noticed something unsettling almost right away: people who knew perfectly well how ELIZA worked—including fellow computer scientists—still developed strong emotional projections toward it. Even his secretary, after watching him write the program and seeing the rules firsthand, reportedly asked him to step out so she could have a private conversation with the system.
That reaction has a name in psychology and computing: the ELIZA effect. It happens when people attribute real understanding, empathy, or inner life to software just because the output fits human conversational patterns.
Think about why that works so well on us. Humans are wired to look for social cues and meaning. When a screen asks "What does that suggest to you?" right after you vent about a bad day, your brain fills in the missing empathy. You supply the emotion; the machine supplies a mirror.
A common mistake people make—and I've caught myself doing it during late-night brainstorming sessions—is confusing sounding insightful with having judgment. A bot can rephrase your worry cleanly in two seconds. That doesn't mean it weighs consequences, lives with risk, or understands what's at stake in your life.
It's easy to assume critics of AI don't understand how software gets built. Joseph Weizenbaum blew that lazy assumption apart.
He wasn't a Luddite smashing looms. He had worked on early computing hardware and software, helped build ELIZA, and sat right at the center of MIT's AI world. Yet as he watched enthusiasm around artificial intelligence swell in the late 1960s and 1970s, he grew alarmed by two things: inflated claims about what machines could achieve, and a growing willingness among professionals to treat calculation as a substitute for human judgment.
In Computer Power and Human Reason (1976), Weizenbaum drew a sharp line between what a computer can compute and what a human ought to decide. Just because a task can be formalized or automated doesn't mean it should be. Decisions involving justice, care, moral responsibility, and human dignity aren't optimization problems with a clean objective function.
His critique wasn't always popular inside ambitious tech circles, where optimism often rewards bigger promises and faster deployment. Some colleagues argued he was drawing artificial limits on machine capability. Fair enough—technologists have debated the boundaries of automation for decades, and modern AI systems handle tasks that early researchers only dreamed about. Still, Weizenbaum's core warning about human agency and moral responsibility hasn't aged a bit. If anything, it feels sharper now that millions of people talk to chatbots every single day.
Comparing ELIZA to today's large language models isn't an apples-to-apples technical comparison. Today's systems train on massive datasets, generate nuanced text across dozens of languages, write code, and summarize dense documents in seconds. Pretending a 1966 rule engine is identical to a modern AI assistant would be sloppy history.
At the same time, ignoring the continuity between ELIZA and today's chatbots is a mistake too. Two patterns repeat across almost sixty years of conversational software.
First, anthropomorphism scales with polish. When a chatbot writes calm, empathetic-sounding replies without hesitation, people naturally treat it like a counselor, friend, or trusted advisor. The more natural the tone, the easier it is to forget there's no lived experience behind the curtain.
Second, convenience tempts us to outsource judgment. Typing a hard question into a chat box feels effortless compared to sitting with uncertainty, talking to a real person who might push back, or doing slow primary research. Smooth answers give us a rush of clarity that isn't always earned.
There's also a real debate worth being honest about. Some researchers and builders argue that well-designed conversational tools can act as helpful sounding boards, journaling prompts, or study aids when used with clear boundaries. Others worry deeply about emotional dependency, sycophancy, and people turning to bots during mental health crises instead of qualified professionals. Evidence on long-term psychological effects is still evolving, which is all the more reason to stay clear-eyed rather than treating chatbots as cures or replacements for real relationships.
If you use chatbots for work, learning, or everyday reflection, a few grounded habits keep you in the driver's seat without making tech miserable to use.
With expertise spanning multiple beats, Noah Sanders brings a multidisciplinary perspective to every story, enriching coverage with context and nuance.
© 2026 Geosirius. All rights reserved.

source

Scroll to Top