On a crisp December day in 2024, deep inside a virtual emulation of a 1960s mainframe, an antiquated line blinked back to life:
HOW DO YOU DO. PLEASE TELL ME YOUR PROBLEM.
This was ELIZA herself — not a replica. The first chatbot ever created, brought back to life on the same kind of computer it first ran on nearly six decades ago.
Wait, what’s ELIZA?
We’ve all gotten used to fancy AIs and chatbots nowadays, but what if I told you the first chatbot is around 60 years old?
Created between 1964 and 1966 by MIT computer scientist Joseph Weizenbaum, ELIZA was a simple program by today’s standards. It used basic pattern-matching and substitution to simulate a conversation — particularly the kind of open-ended questions favored by a Rogerian psychotherapist. When someone typed “I’m sad,” ELIZA might respond, “How long have you been sad?”
By the day’s standards, ELIZA was enchanting. Weizenbaum’s own secretary, drawn to its responses, once asked to be left alone with it. Others poured their hearts out to the machine. ELIZA was only known to a few people, but in that group, it became cultural phenomenon. But as it spread across early computing networks, something strange happened: the original version was lost.
Weizenbaum wrote ELIZA in MAD-SLIP, a mix of MAD (MIT’s Algebraic Decoder) and a list-processing library he had developed called SLIP. It ran on CTSS, the Compatible Time-Sharing System, hosted on MIT’s IBM 7094. The CTSS was a beast, an immense computer that cost nearly $3 million to run. It was also the first computer to implement password login.
CTSS ran on two tape units: one for the user and one for dumping the program in memory. The memory was 27 k words (36-bit words) for users, and 5 k words for the supervisor (operating system). This seems unimaginable by today’s standards, but this was able to run ELIZA.
How the First Chatbot Was Lost
The problem was that the machine was never connected to ARPAnet, the precursor of the internet. So, when other programmers began rewriting ELIZA in languages like Lisp, it was their versions that spread.
The Lisp version became the dominant strain, traveling rapidly through ARPAnet and embedding itself into the DNA of early AI research. Soon after, a BASIC version of ELIZA appeared in Creative Computing magazine in 1977, just as personal computers began arriving in American homes.
As a result, most people knew ELIZA either as a Lisp-based academic artifact or as a playful BASIC program typed into their Apple II. The original MAD-SLIP code faded into obscurity.
Until, that is, a team of digital sleuths decided to look for it.
The Unearthing of a Digital Fossil
In 2021, Jeff Shrager — who had written one of the first ELIZA clones back in the 1970s — convinced MIT archivist Myles Crowley to go digging. They found it in a box labeled “computer conversations”. Inside was a 1965 printout of the original source code.
But there was a problem. The code was incomplete, printed in fading ink, and in an arcane format created before ASCII existed. Some lines were abbreviated into cryptic fragments like W'R for WHENEVER. Others spanned punch cards, where missing spaces or typos could wreck entire routines.
The only way to test it was to revive CTSS itself — and that meant rebuilding a simulation of an IBM 7094. So, they did.
The team — Rupert Lane, Anthony Hay, Arthur Schwarz, David M. Berry, and Shrager — called themselves Team ELIZA. Together, they set out to resurrect not just ELIZA, but its entire ecosystem.
It was slow, painstaking work, and it almost failed because of a single bug: a single missing zero in line 1670 of a function. But when everything was fixed and sorted, ELIZA could finally speak.
Men are all alike.
IN WHAT WAY
The line echoed a transcript from Weizenbaum’s 1966 paper. ELIZA was alive again.
ELIZA can… learn?
Again, we’re talking about a simple chatbot from the ’60s. But ELIZA had one more surprise: it could learn.
Buried in the rediscovered code was a hidden “teacher mode,” invoked by typing +. This feature, barely mentioned in Weizenbaum’s published work, let users edit ELIZA’s script live — adding, removing, and modifying rules on the fly. The program could save these changes to disk, a kind of primitive persistence. It wasn’t quite machine learning — but it hinted at something more than just canned responses. In 1966, that was revolutionary.
ELIZA is a piece of computation history. She existed before the term “chatbot” existed. But this is a bit more than just a piece of history. She embodied ideas (symbolic reasoning, interactive computing, psychological modeling) that laid the groundwork for modern AI. And she exposed the human impulse to project feelings onto machines. Weizenbaum himself became a vocal critic of such projections, warning of “the computer as a psychological tool.”
Now more than ever, with AI truly taking off, it’s time to think about what ELIZA is. She seems very simple now but was enthralling 60 years ago. She seemed human but was, of course, just an algorithm. What does that say about our current AI systems?
Journal Reference: Rupert Lane et al, ELIZA Reanimated: The world’s first chatbot restored on the world’s first time sharing system, arXiv (2025). DOI: 10.48550/arxiv.2501.06707
