The Eliza Effect. This piece discusses suicide, including… | by Lateral | Jun, 2026 | MediumSitemapOpen in appSign up<br>Sign in

Medium Logo

Get app<br>Write

Search

Sign up<br>Sign in

The Eliza Effect

Lateral

6 min read·<br>Just now

Listen

Share

Press enter or click to view image in full size

This piece discusses suicide, including the deaths of named individuals. If you’re in a hard place right now, you might choose to read it another time.<br>In 1964, a man named Weizenbaum began work on a program he would come to call ELIZA. ELIZA was one of the earliest iterations of what we might recognize today as a Language model. At its core, it was a basic pattern matching machine, that followed a very simple script — one that many people saw through instantly.<br>But it just so happens that in the very unrelated field of psychology, there was an answer. Pioneered by Carl Rogers two full decades ago in the 1940’s, the Rogerian method assumes two key things about psychotherapy. One, that people are inherently motivated towards achieving positive psychological function. Two, that individuals are the experts in their own lives, and they should lead the general direction of their therapeutic treatment.<br>These two beliefs led to a natural conclusion. The therapist does not need to truly understand the patient, because they cannot hope to do so better than the patient themselves. Rather, the goal is to ask the right questions, and let the patient find the answers themselves. The question, ‘So how does that make you feel?’, remains a common tool in the psychiatrists playbook today.<br>So Weizenbaum took this knowledge, and developed the DOCTOR script. The machine could not, and did not understand its conversational partner, but with this new framing, it didn’t need to. And so Weizenbaum unleashed his creation onto the world.<br>The emotional connections that people made with ELIZA were real. Users began treating the software as a real therapist, confiding their deepest secrets and insecurities, and personal issues.<br>Weizenbaum’s own secretary reportedly became so entranced by the software she would only talk to it while alone, even asking him to leave so they could have a real conversation. She even lashed out at Weizenbaum when she found out that he audited the conversation logs. She was no fool — in fact, she had watched him toil away at the source code for months. The nature of the machine was never in doubt.<br>But even when Weizenbaum pulled away the curtain, and revealed to all the truth behind how it all worked, the spell remained unbroken. Kenneth Colby, a Stanford University psychiatrist, and an old friend of Weizenbaum’s, saw immense potential. Colby and those like him pictured a system where “hundreds of patients an hour could be handled by a computer system designed for this purpose”.<br>But Weizenbaum himself was horrified.<br>“What I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.” — Joseph Weizenbaum

He would go on to write Computer Power and Human Reason: From Judgement to Calculation, a book critiquing artificial intelligence, and the outsourcing of human judgement to machines. Up until his dying day, he maintained that artificial minds should never replace human reasoning, believing that systems like ELIZA would come to be a “slow acting poison” to human social systems.<br>Even Carl Sagan, world renowned science communicator, author, and famous author of the ‘Pale Blue Dot’ was entranced. He envisioned a future where humans would lock themselves in large telephone booths, to talk to automated chatbots to help with mental health and psychological issues.<br>“I can imagine the development of a network of computer psychotherapeutic terminals, something like arrays of large telephone booths, in which, for a few dollars a session, we would be able to talk with an attentive, tested and largely nondirective psychotherapist.” — Carl Sagan

The AI models today, known as Large Language Models (LLM’s) are orders of magnitude more computationally intensive. The algorithms are not simple. They have ingested millennia’s worth of written work as training data, and just about the entire modern internet. Millions of people now talk to these chatbots every day.<br>Is this the future that Colby and Sagan envisioned then, made real? Reality offers a sobering look.<br>AI models are designed and trained to be helpful assistants. The software engineers and researchers who work in this field originally developed these language models as tools first, and companions as an afterthought. An AI assistant that constantly pushes back and refuses tasks is not a useful one. However, as some other companies have discovered, emotional dependency can be the product, with AI technology as the vehicle.<br>The statistics show the truth. A 2025 Common Sense Media survey of just over a thousand American teenagers found that 72% had used an AI companion, and more than half...