The Mechanical Turk hid a human chess player inside
The 18th-century chess automaton known as the Mechanical Turk was operated by a human concealed in its cabinet, as Edgar Allan Poe argued in 1836.
Facts
Atomic, verifiable facts - every one tied to a primary source.
268 entries, all primary-sourced
Samuel Butler warned of evolving machines in an 1863 article
Samuel Butler's 1863 article 'Darwin among the Machines' applied evolution to machines and became the seed of Erewhon's Book of the Machines.
Hollerith's tabulator captured the census trial data in 5.5 hours where rivals needed 44 to 55
In the 1888 Census Bureau competition, Herman Hollerith's electric tabulator captured the trial data in 5.5 hours where rivals needed roughly 44 and 55 hours.
Principia Mathematica does not prove that 1 plus 1 equals 2 until well into the first volume
Whitehead and Russell's Principia Mathematica was so rigorous that the proof that one plus one equals two does not arrive until well into the first volume.
Torres Quevedo's El Ajedrecista was the first true game-playing machine
Around 1912 Leonardo Torres Quevedo built El Ajedrecista, an automaton that played the king-and-rook chess endgame on its own - the first game-playing machine.
The word 'robot' was coined for a 1921 play and comes from the Czech for forced labor
Karel Capek's 1921 play 'R.U.R.' introduced the word 'robot,' built from the Czech 'robota' meaning forced labor; Capek credited his brother Josef for it.
Hollerith's punched-card company became IBM
Herman Hollerith founded the Tabulating Machine Company in 1896 to sell his punched-card system; through mergers that firm became, in 1924, IBM.
Cinema's first major robot was the false Maria in Metropolis
The Maschinenmensch robot in Fritz Lang's 1927 Metropolis is regarded as the first blockbuster robot and the template for later film machines.
Godel proved incompleteness by building a sentence that says of itself 'this is not provable'
To prove his 1931 first incompleteness theorem, Kurt Godel encoded formulas as numbers and built a sentence that says of itself 'this is not provable.'
Church's lambda calculus became a root of computer science and shaped Lisp
Alonzo Church's 1936 lambda calculus, built to prove an unsolvability result, became a deep root of computer science and shaped functional programming and Lisp.
Konrad Zuse's Z3 was built from about 2,000 telephone relays and read its program from punched film
The Z3, presented in Berlin in May 1941 and regarded as the first program-controlled automatic computer, used about 2,000 telephone relays and punched film.
ENIAC used roughly 18,000 vacuum tubes
ENIAC, unveiled in February 1946 as the first general-purpose electronic digital computer, used roughly 18,000 vacuum tubes and weighed about thirty tons.
Grey Walter's 1940s robot tortoises found their own charging station
Grey Walter's late-1940s tortoise robots used phototaxis to steer back to a recharging station when their batteries ran low.
Asimov's Three Laws of Robotics appear in 'I, Robot'
Asimov Three Laws of Robotics are stated in his 1950 collection I, Robot, and the stories - starting with Runaround - show the Laws producing harmful behavior.
Shannon estimated chess has about 10^120 possible games
In his 1950 chess paper, Claude Shannon estimated the number of possible chess games at around 10 to the 120th power.
Turing predicted that by about 2000 machines would fool an average judge for five minutes
In his 1950 paper, Alan Turing predicted that by around 2000 machines would play the imitation game well enough to fool an average judge for five minutes.
Turing argued machines could think on BBC radio in 1951
On 15 May 1951 the BBC broadcast Alan Turing's talk 'Can Digital Computers Think?', a year after his famous paper on the same question.
Bell Labs' Audrey only recognized digits well for one trained voice
Audrey hit 97 to 99 percent digit accuracy, but only after its circuit was tuned to a single speaker's voice.
The Dartmouth proposal estimated AI as a two-month, ten-man study
The 1955 proposal for the Dartmouth summer project asked for 'a 2 month, 10 man study of artificial intelligence' to be carried out in the summer of 1956.
The term 'artificial intelligence' was coined in a 1955 funding proposal
The phrase 'artificial intelligence' was introduced in the August 1955 Dartmouth proposal written by McCarthy, Minsky, Rochester, and Shannon.
The General Problem Solver introduced means-ends analysis
Newell, Shaw, and Simon GPS introduced means-ends analysis: pick an operator that reduces the biggest gap to the goal, then repeat - echoed in modern AI agents.
PLATO had online forums and chat decades before the internet
The University of Illinois PLATO education system pioneered message boards, email, chat, and multiplayer games long before the modern internet.
The Mark I Perceptron was a room-sized machine with 400 photocells
Rosenblatt's Mark I Perceptron was physical hardware - a 20-by-20 grid of 400 photocells wired to motor-driven weights - not just an algorithm.
Lisp treats program code and data as the same kind of object
John McCarthy's 1960 Lisp treated code and data as the same object, and a short function could interpret any Lisp expression - defining the language in itself.
The first industrial robot patent was titled 'Programmed Article Transfer'
George Devol's robot patent, US 2,988,237, was filed in 1954, granted in 1961, and titled 'Programmed Article Transfer' - it became the basis for the Unimate.
The IBM Shoebox could understand just 16 spoken words
IBM's 1962 Shoebox recognized 16 words - the digits zero to nine plus six commands like plus, minus, and total.
DENDRAL is widely regarded as the first expert system
DENDRAL, begun at Stanford around 1965 to identify molecules from mass-spectrometer data, is widely regarded as the first expert system.
I.J. Good coined the term intelligence explosion
The phrase intelligence explosion comes from I.J. Good's 1965 paper on the first ultraintelligent machine.
ALPAC found no predictable prospect of useful machine translation
The 1966 ALPAC report found 'no immediate or predictable prospect of useful machine translation,' leading the US government to cut MT funding.
MIT once treated building computer vision as a summer project
A 1966 MIT memo proposed building 'a significant part of a visual system' as a summer project; reliable computer vision took roughly fifty more years.
HAL 9000 debuted in Clarke's 1968 novel '2001: A Space Odyssey'
The artificial intelligence HAL 9000 appeared in Arthur C. Clarke's novel '2001: A Space Odyssey,' published in 1968 alongside Stanley Kubrick's film.
87% of Americans are identifiable by ZIP, gender, and birth date
Latanya Sweeney found 87% of the US population is uniquely identifiable from just ZIP code, gender, and date of birth.
A 1.3B InstructGPT was preferred over the 175B GPT-3
Human evaluators preferred the 1.3-billion-parameter InstructGPT over the 175-billion-parameter GPT-3, despite about 100x fewer parameters.
A 2022 survey put a 50% chance of human-level AI at 2059
AI Impacts' 2022 survey of 738 machine-learning researchers gave a median 2059 forecast for high-level machine intelligence.
A tribunal rejected Air Canada's claim its chatbot was a separate entity
In Moffatt v. Air Canada, the Civil Resolution Tribunal ruled the airline is responsible for what its website chatbot says, awarding the customer $650.88.
AI training compute doubled every 3.4 months after 2012
OpenAI's 2018 'AI and Compute' found the largest training runs doubled compute every 3.4 months since 2012 - over 300,000 times growth, beating Moore's Law.
AlexNet was trained on two consumer GPUs
AlexNet, the 2012 network that launched deep learning, was trained on two NVIDIA GTX 580 GPUs - the same class of graphics card sold to gamers.
AlphaZero reached superhuman chess in 24 hours
Starting from random play and knowing only the rules, AlphaZero reached a superhuman level of chess and shogi within 24 hours of self-play training.
Amazon billed Mechanical Turk as 'Artificial Artificial Intelligence'
Amazon's 2005 Mechanical Turk launch billed it as 'Artificial Artificial Intelligence' - software calling on hidden human workers to do what machines could not.
An MIT-OpenAI trial on emotional ChatGPT use ran four weeks with 981 people
A 2025 MIT Media Lab and OpenAI randomized trial studied 981 participants over four weeks and more than 300,000 messages with ChatGPT.
Anthropic is led by siblings Dario and Daniela Amodei
Anthropic was founded and is led by siblings Dario Amodei (CEO) and Daniela Amodei (President), former OpenAI researchers.
Arthur Samuel's 1959 checkers paper and machine learning
Arthur Samuel's 1959 IBM paper on a self-improving checkers program is widely credited as the origin of the term 'machine learning'.
Asimov's Three Laws of Robotics first appeared in Runaround in 1942
Isaac Asimov first stated the Three Laws of Robotics as an explicit code in his story Runaround, in the March 1942 issue of Astounding Science Fiction.
Books3 held roughly 200,000 pirated e-books
The Books3 AI training dataset was a collection of about 200,000 illegal e-books before a rights-holder group had it taken offline in 2023.
California suspended Cruise's robotaxi permits in 2023
California's DMV suspended Cruise's driverless permits in 2023 after one of its robotaxis dragged a pedestrian.
Chinchilla rule: double the data for every doubling of model size
DeepMind's Chinchilla paper found that for compute-optimal training, the number of training tokens should double each time model size doubles.
Chinchilla: double the model, double the data
DeepMind's 2022 Chinchilla paper found compute-optimal training needs model size and training tokens to scale equally - double the model, double the data.
CLIP was trained on 400 million image-text pairs
OpenAI's CLIP learned from a dataset of 400 million image-and-text pairs collected from the internet.