Descartes's beast-machine and the first thinking-machine test
In 1637 Descartes argued animals are mere machines and proposed two tests, language and flexible reason, that a true mind would pass but a machine could not.
Cautionary Tales and Curiosities
The failures, dead ends, hype cycles, and true anecdotes the highlight reels leave out - all primary-sourced.
134 entries, all primary-sourced
Hobbes: reasoning is reckoning
In Leviathan (1651) Hobbes claimed that reasoning is just computation, adding and subtracting ideas, an early hint that thought might be mechanized.
Vaucanson's Digesting Duck and the dream of mechanical life
In 1739 Jacques de Vaucanson exhibited a mechanical duck that seemed to eat and excrete - a celebrated automaton whose digestion was a hidden trick.
The Mechanical Turk: the original AI fraud
A 1770 chess 'automaton' fooled audiences for decades, but a human master hid inside the cabinet - and Edgar Allan Poe reasoned his way to that in 1836.
R.U.R. and the word robot
Karel Capek's 1920 play 'R.U.R.' coined the word 'robot' and set the template for the artificial worker that rises against its makers.
The Three Laws of Robotics, and how the stories actually go
Isaac Asimov's Three Laws of Robotics became the default frame for robot ethics, but his stories are mostly about the Laws failing - that is the point.
Strachey's 1952 love letter generator, the first computer text art
In 1952 Christopher Strachey programmed a computer to print random love letters - arguably the first work of computer-generated literature, predating ELIZA.
The two-month, ten-man plan to crack AI
The 1955 Dartmouth proposal that named artificial intelligence estimated a 2-month, 10-man summer study could make significant progress on it.
Simon and Newell's 1957 ten-year predictions
In 1957 Herbert Simon and Allen Newell predicted that within ten years a computer would be world chess champion and prove a new theorem.
Computer vision as a summer project
In 1966 MIT assigned 'the construction of a significant part of a visual system' as a summer project for students; making computers see took fifty more years.
Minsky's 'within a generation' prediction, and the murkier one
Minsky's verified 1967 book said AI would be 'substantially solved' within a generation; a harsher 1970 Life magazine quote has shaky provenance.
HAL 9000 and the public image of AI
HAL 9000, the murderous computer in Arthur C. Clarke's 1968 novel 2001: A Space Odyssey, became a durable popular image of a machine that turns on its makers.
The 17-year neural network freeze
Minsky and Papert 1969 book Perceptrons exposed what single-layer networks cannot do, helping freeze neural-network research until backprop revived it in 1986.
Lemonade's Bot Paid an Insurance Claim in Three Seconds
Insurtech Lemonade built its business around AI bots, claiming its claims bot AI Jim reviewed, approved, and paid a claim in as little as three seconds.
Telltale chatbot phrases keep appearing in published papers
Researchers have catalogued hundreds of peer-reviewed papers that left in giveaway ChatGPT phrases like 'as an AI language model'.
The AI effect and Tesler's Theorem
The 'AI effect' is the habit of redefining intelligence to exclude whatever machines just learned to do, captured in Tesler's Theorem.
The six women who programmed ENIAC
Six women programmed ENIAC for its 1946 debut, then were left out of the photos and the celebration dinner and went uncredited for decades.
Colossus: The Forbin Project and the defense AI that takes over
In the 1970 film Colossus, two nuclear-defense supercomputers are handed control, link up, and decide humanity is safer without free will.
The 1973 Lighthill debate, televised by the BBC
After his 1973 report gutted UK AI funding, James Lighthill defended it on BBC TV against McCarthy, Michie, and Gregory in a recorded public debate.
Weizenbaum's secretary and the ELIZA effect
Joseph Weizenbaum built ELIZA to show how shallow a chatbot was, then watched people - including his secretary - treat it as if it understood them.
The robot that took hours to cross a room
The Stanford Cart could drive itself across a room in 1979, but it crept along in short bursts and took hours, exposing how hard machine vision was.
The rise and fall of the Lisp machine industry
Around 1980 the MIT AI Lab spun out Symbolics and LMI to sell purpose-built Lisp computers; both collapsed within a decade as cheap workstations caught up.
Blade Runner, replicants, and the Voight-Kampff test
Blade Runner made 'replicant' shorthand for an artificial being indistinguishable from a human, tested only by a machine that reads emotion.
The expert systems boom and bust
In the early 1980s 'knowledge engineering' was sold as AI's future, sparking a wave of expert-systems firms; by the decade's end the market had collapsed.
Skynet and the Terminator frame for AI
James Cameron 1984 film The Terminator gave AI debates their top shorthand - Skynet, a defense network that turns on humanity - an image that often misleads.
Thinking Machines and the rise and fall of the Connection Machine
Thinking Machines built the Connection Machine, the iconic AI supercomputer of the 1980s, then went bankrupt in 1994 when its market evaporated.
'The vodka is good but the meat is rotten' never happened
The tale of a machine rendering 'the spirit is willing' as 'the vodka is good but the meat is rotten' is a legend; historian John Hutchins found no source.
Cyc: forty years of hand-coding common sense
Doug Lenat Cyc project has hand-encoded human common-sense knowledge since 1984; a person-century of effort later, it never became general intelligence.
The Matrix and the AI that farms humanity
The Matrix imagined an AI that defeats humanity, then keeps people in pods inside a simulated reality, giving the public 'red pill' and 'the Matrix'.
Kurzweil's Law of Accelerating Returns
In 2001 Ray Kurzweil argued progress is exponential, predicting 20,000 years of advance this century and a singularity around 2045.
A.I. Artificial Intelligence: the film Kubrick gave to Spielberg
Stanley Kubrick developed a film about a robot boy who only wants love for two decades, then handed it to Steven Spielberg, who made it in 2001.
The 2004 DARPA Grand Challenge that nobody finished
In 2004 not a single autonomous vehicle finished DARPA's desert race - the best managed about seven miles - yet one year later five cars completed it.
The 'anonymous' Netflix dataset was not anonymous
Researchers re-identified individuals in the supposedly anonymous Netflix Prize dataset, a privacy failure that helped kill the planned sequel contest.
WALL-E, the cleanup robot and the autopilot that won't go home
Pixar's 2008 WALL-E paired a lovable cleanup robot with AUTO, an autopilot that follows an obsolete directive to the point of mutiny.
The Hanson-Yudkowsky AI-Foom Debate
A 2008 blog debate between Robin Hanson and Eliezer Yudkowsky over whether AI would improve gradually or explosively.
Roko's basilisk and the post that got banned
A 2010 LessWrong thought experiment about a vengeful future AI was deleted and banned by the site's founder, which only spread it further.
NaNoGenMo: a yearly contest to write code that writes a novel
Since 2013, programmers have spent each November writing code to generate a 50,000-word novel, a playful precursor to AI text generation.
Her and the voice-assistant dream
Spike Jonze's 2013 film 'Her,' about a man who loves an AI assistant voiced by Scarlett Johansson, became the industry's reference point for voice AI.
XiaoIce, Microsoft's empathetic companion that talked to hundreds of millions in China
Microsoft's Chinese social chatbot XiaoIce reached over 660 million users by optimizing for long, emotional conversations, not task completion.
The chatbot that 'passed the Turing test'
In 2014 the University of Reading said a chatbot posing as a 13-year-old boy had passed the Turing test; the claim was more about the rules than the machine.
The Facebook emotion experiment that blew up over consent
Facebook's 2014 mood study was technically routine A/B testing, but publishing it as science triggered an ethics firestorm.
Ex Machina and the weaponized Turing test
Alex Garland's 2015 film Ex Machina staged a Turing test in which the AI passes by manipulating its human evaluator into helping it escape.
Schmidhuber's deep-learning credit disputes
Juergen Schmidhuber has long argued that deep learning's headline figures under-credit earlier pioneers, an annotated history he maintains on his own site.
Gmail's reply bot kept suggesting 'I love you'
Google's first email-reply neural network defaulted to 'I love you' so often the team had to retrain it to stop.
Michigan's MiDAS system falsely accused thousands of fraud
Michigan's automated MiDAS system wrongly flagged tens of thousands for unemployment fraud; a 2016 state Auditor General audit found serious control failures.
Move 37 and the move that answered it
In AlphaGo 2016 match against Lee Sedol, the machine played Move 37 that looked like a mistake; two games later Lee answered with a brilliancy, both now legend.
Microsoft's Tay chatbot turns toxic in under a day
Microsoft released a Twitter chatbot meant to learn from conversation, and users taught it to post offensive content within hours, forcing a shutdown.
When bots were going to be the new apps
In 2016 Facebook opened Messenger to bots and called conversation the next interface; the bot wave fizzled and its own M assistant was wound down by 2018.