AI 4 November 2026 5 MIN READ
ELIZA, Deep Blue and AlphaGo
Three machines that surprised their makers: a 1966 script that only reflected, a chess engine that calculated and a Go player with a famous move 37.
THE SHORT ANSWER
ELIZA in 1966 only reflected sentences back, yet users felt understood. Deep Blue beat Kasparov in 1997 by brute-force search, not learning. AlphaGo beat Lee Sedol 4–1 in 2016, and a move by each player was rated a 1-in-10,000 choice. Three machines, three kinds of surprise.
On this page
ELIZA, 1966: a program that only reflected
ELIZA was written by Joseph Weizenbaum at MIT in a language called MAD-SLIP for an IBM 7094, and described in Communications of the ACM in January 1966. It did not understand anything. Input sentences were analysed with decomposition rules triggered by keywords, and the answers were built with reassembly rules. The famous therapist was only one script, called DOCTOR, which mimics a Rogerian psychotherapist by reflecting a person’s statements back as questions. Weizenbaum had planned other scripts too, for a travel agent, a mathematician and a poet.
The reaction surprised him. In the 1966 paper he reported that some subjects were very hard to convince that the program was not human. In his 1976 book, as relayed by Smithsonian Magazine, he recalled that his secretary asked him to leave the room so she could carry on talking to ELIZA in private, and he concluded that extremely short exposures to a simple program could induce powerful delusional thinking in quite normal people. We could not open the book, so those two details come through the Smithsonian.
Two myths are worth correcting. ELIZA was never published with the paper, only the DOCTOR script was, and it was not written in Lisp. The Lisp versions were re-creations from the paper’s description. And the code was found again: in 2021, Jeff Shrager asked an MIT archivist to search Weizenbaum’s papers, and a folder labelled “Computer Conversations” held a printout of ELIZA’s original source. Weizenbaum’s estate agreed to release it into the public domain. In January 2025 the team published “ELIZA Reanimated”, running the original program on a restored operating system emulated on an IBM 7094. The recovered code does not follow the algorithm in the 1966 article. It treats “but” as a sentence delimiter, for instance, which the paper never mentions. Sources disagree on the code’s size: 411 lines according to the ACM, or a main program of about 230 lines according to ELIZAGEN.
Deep Blue, 1997: brute force, not learning
The 1996 match in Philadelphia was won by Garry Kasparov 4–2, but Deep Blue won the first game, the first time a computer beat a reigning world champion under regular time controls. In May 1997 in New York the rematch went as follows: Kasparov won game 1, Deep Blue game 2, games 3 to 5 were drawn, and Deep Blue won game 6 in 19 moves, for a match score of 3.5–2.5.
IBM says plainly that this was not machine learning: Deep Blue got its strength from brute-force computing power. Between the matches, the team improved its endgame databases and its evaluation function and hired grandmasters as advisers. Its designers describe a 30-node IBM system with 480 single-chip chess engines, which searched an average of 126 million positions a second over longer searches during the 1997 match. IBM’s history page gives different figures, so the hardware numbers depend on the source.
A well-known story says a bug made Deep Blue play a random move at the end of game 1 that Kasparov read as superhuman. It rests on one interview, in which Murray Campbell told Nate Silver, so treat it as an anecdote. Afterwards Kasparov accused IBM of cheating and asked for a rematch. IBM declined and retired the machine.
AlphaGo, 2016: the 1-in-10,000 moves
In March 2016 in Seoul, DeepMind’s AlphaGo beat Lee Sedol, a 9-dan professional with 18 world titles, by 4 games to 1. The games were played on 9, 10, 12, 13 and 15 March. AlphaGo combined a policy network to choose moves with a value network to predict the winner, trained first on human expert games and then through self-play, with tree search. Five months earlier, it had beaten the European champion Fan Hui 5–0.
Two moves became famous. In game 2, AlphaGo’s move 37, a shoulder hit on the right side of the board, had a 1 in 10,000 chance of being played by a human, by DeepMind’s estimate. One 9-dan commentator called it a very strange move and the other thought it was a mistake. It turned out to be the turning point of the game.
In game 4, Lee Sedol answered with his own move 78, a wedge on the 11-11 point, which DeepMind also rated at 1 in 10,000 for a human. It was Lee’s only win. After it AlphaGo’s play fell apart in a string of blunders, and DeepMind’s own commentary admits that “we still do not know why”. Note that the 1-in-10,000 figure is an estimate from AlphaGo’s model, not a measured frequency.
Why it matters
Three machines, three different ideas: a script that reflects, a search that calculates, a network that learns. People reacted to each of them in ways their makers did not expect.
Our take: the surprise was in the people. Each time, the machine did something narrow. What changed was what humans read into it.
Do you agree?
Take it to Chevlume: write your view, add your sources and let others answer. The closed alpha is by invitation.
Keep reading
The Mechanical Turk: A Chess Machine Hoax
A chess machine from 1770 that was really a hidden person. How the trick worked, who it beat, and how its name ended up at Amazon.
AI / EXPLAINEROpen Source vs Open Weights in AI
What 'open' really means for AI models: the Open Source AI Definition, open weights, restrictive licences and why the difference changes what you may do.
AI / EXPLAINERWhat Is an AI Agent, Really?
An agent is a model that chooses its own steps and uses tools in a loop. How that differs from a workflow, how it is tested and how it fails.
Sources
- ELIZA—A Computer Program For the Study of Natural Language Communication Between Man and Machine (Weizenbaum, 1966; Wayback copy of Harvard page) · Communications of the ACM (Wayback copy of Harvard page) · Accessed
- The Many Lives of ELIZA · Communications of the ACM (news, Alex Wright) · Accessed
- Why the Computer Scientist Behind the World’s First Chatbot Dedicated His Life to Publicizing the Threat Posed by A.I. · Smithsonian Magazine · Accessed
- Joseph Weizenbaum's Original ELIZA · ELIZAGEN (curated by Jeff Shrager) · Accessed
- ELIZA Reanimated: The world’s first chatbot restored on the world’s first time sharing system · arXiv (Lane, Hay, Schwarz, Berry, Shrager) · Accessed
- Deep Blue · IBM · Accessed
- Deep Blue · Artificial Intelligence 134 (2002), Campbell, Hoane, Hsu (IBM) · Accessed
- Deep Blue (Britannica) · Encyclopaedia Britannica · Accessed
- Did Deep Blue beat Kasparov because of a computer bug? · WIRED (Klint Finley) · Accessed
- Kasparov versus Deep Blue 1997 · Chessprogramming wiki · Accessed
- What we learned in Seoul with AlphaGo · Google (The Keyword) · Accessed
- Google DeepMind Challenge Match - Lee Sedol v AlphaGo - match report · British Go Association · Accessed
- In Two Moves, AlphaGo and Lee Sedol Redefined the Future · WIRED (Cade Metz) · Accessed
- AlphaGo · Google DeepMind · Accessed
- Challenge Match Game 4: “Endurance” (game commentary) · Google DeepMind · Accessed
- Mastering the game of Go with deep neural networks and tree search · Nature (Silver et al.) · Accessed
Published
Discuss on X