The Complete AI Timeline

If the history of artificial intelligence were squeezed into a single day, almost everything you have heard about would happen in the last few minutes. For more than two thousand years people imagined thinking machines. For nearly four centuries they built calculating ones. Only since the 1940s have they tried to make machines that learn, and only since 2022 have most people talked to one. This page lays the whole span out in order, so you can see the long dreams, the false starts, the quiet decades when the crucial ideas were found, and the sudden bursts when they finally paid off.

How to use this timeline

Each entry has a date, a headline and a short note that opens when you tap it, with a link to the page that tells that part of the story in full. Use the era chips above the list to show one period at a time. If you are new to the subject, start with AI History at a Glance, which gives the shape of the story before the detail.

A few conventions. Dates are the year an idea was published, a system was first shown, or an event happened; where work spanned several years the start or the best-known date is given and the note says so. Names are those credited in the original papers. Some credit is contested, especially for early neural-network ideas; Pioneers of AI discusses those disputes.

Nine eras

This guide’s framing Historians do not agree on one official way to divide AI’s history. The nine eras below are this guide’s working scheme, built around the shifts in what researchers believed would work: logic, then knowledge, then data, then scale. The boundaries are approximate, and some entries sit on a border (the Lighthill report of 1973 is filed under the first winter it helped cause).

• 1TO 1940Dreams & machinesMyths of artificial beings, automata, mechanical calculators, Babbage and Lovelace, and the logic of Boole, Gödel and Turing.
• 21940s–1956FoundationsElectronic computers, model neurons, cybernetics, information theory and Turing’s 1950 paper.
• 31956–1974Golden yearsThe field is named; programs prove theorems, play checkers and chat. Optimism and funding run high.
• 41974–1980First winterPromises outrun results; British and American funding is cut back.
• 51980–1993Expert systemsRule-based systems boom commercially, neural networks revive, then the boom busts into a second winter.
• 61993–2011Learning from dataStatistics and machine learning take over; the web supplies data; Deep Blue and Watson win in public.
• 72012–2016Deep learningGPUs and big datasets make many-layered networks work, from ImageNet to AlphaGo.
• 82017–2021TransformersAttention, pretraining and scaling produce BERT, GPT-3 and AlphaFold 2.
• 92022–Generative eraChatGPT, image generators, open-weight models, reasoning models, agents and the first AI laws.

The timeline

1. c. 700 BCEHesiod tells of Pandora, a woman made by the smith-god HephaestusGreek myth already imagined artificial beings; later stories added the bronze giant Talos who guarded Crete. Ancient dreams →
2. 1st c. CEHero of Alexandria describes programmable automataHis mechanical theatre was driven by weights and strings wound around pegged drums: an early form of stored instructions. Ancient dreams →
3. 1206Al-Jazari’s Book of Knowledge of Ingenious Mechanical DevicesThe engineer of the Artuqid court in upper Mesopotamia described water clocks, a musical automaton band and other machines. Ancient dreams →
4. c. 1305Ramon Llull’s Ars generalis ultimaLlull used rotating paper wheels to combine concepts mechanically, an idea that later inspired Leibniz. Ancient dreams →
5. 1642Blaise Pascal builds a mechanical calculatorThe Pascaline added and subtracted with geared wheels: arithmetic, once a mental skill, done by a machine. Ancient dreams →
6. 1673Leibniz demonstrates his stepped reckonerGottfried Wilhelm Leibniz showed his calculating machine to the Royal Society and dreamed of a calculus that would settle arguments by computation. Ancient dreams →
7. 1739Vaucanson’s “Digesting Duck”Jacques de Vaucanson’s automata, including a flute player (1738) and a duck that appeared to eat and digest, toured Europe. Ancient dreams →
8. 1770The “Mechanical Turk” chess automatonWolfgang von Kempelen’s chess-playing machine was a hoax with a human hidden inside, but it fed a long fascination with thinking machines. Ancient dreams →
9. 1818Mary Shelley’s FrankensteinThe novel set a lasting template for stories about creators who lose control of what they make. Ancient dreams →
10. 1822Charles Babbage proposes the Difference EngineA machine to calculate and print mathematical tables without human error. Babbage and Lovelace →
11. 1837Babbage describes the Analytical EngineWith a “store” and a “mill”, conditional branching and punched-card input, it was the first design for a general-purpose computer. Babbage and Lovelace →
12. 1843Ada Lovelace publishes her NotesHer notes on the Analytical Engine include a program for Bernoulli numbers and the observation that the Engine could manipulate symbols, not just numbers. Babbage and Lovelace →
13. 1854George Boole’s The Laws of ThoughtBoole turned logic into algebra; a century later his true/false arithmetic became the language of digital circuits. Babbage and Lovelace →
14. 1914Torres Quevedo demonstrates El AjedrecistaLeonardo Torres Quevedo’s electromechanical machine played a king-and-rook endgame against a human, unaided. Ancient dreams →
15. 1921Karel Čapek’s play R.U.R. premieresThe play, written in 1920, introduced the word “robot”, from a Czech word for forced labour. Ancient dreams →
16. 1931Kurt Gödel’s incompleteness theoremsGödel showed that any consistent formal system rich enough for arithmetic contains true statements it cannot prove. Alan Turing →
17. 1936Turing’s “On Computable Numbers”Alan Turing defined the Turing machine and showed some problems cannot be solved by any mechanical procedure; Alonzo Church reached a parallel result. Alan Turing →
18. 1943McCulloch and Pitts model the neuronWarren McCulloch and Walter Pitts showed that networks of simple on–off “neurons” could compute logical functions. Neurons, feedback and information →
19. 1945Von Neumann’s EDVAC reportJohn von Neumann’s draft described a stored-program computer, the architecture of almost every computer since. Neurons, feedback and information →
20. 1948Wiener’s Cybernetics and Shannon’s information theoryNorbert Wiener named the science of feedback and control; Claude Shannon’s “A Mathematical Theory of Communication” made information measurable in bits. Neurons, feedback and information →
21. 1949Donald Hebb’s The Organization of BehaviorHebb proposed that a connection between two neurons strengthens when one repeatedly helps to fire the other, a seed of learning in neural networks. Neurons, feedback and information →
22. 1950Turing’s “Computing Machinery and Intelligence”Turing proposed the imitation game and answered objections to machine thinking, from theology to Lady Lovelace. Alan Turing →
23. 1951Minsky and Edmonds build SNARCMarvin Minsky and Dean Edmonds built a learning machine from about 40 artificial neurons made with vacuum tubes. The Golden Years →
24. 1952Arthur Samuel’s checkers programRunning on the IBM 701, Samuel’s program later learned to improve its own play, an early landmark of machine learning. The Golden Years →
25. 1955The Dartmouth proposal coins “artificial intelligence”McCarthy, Minsky, Rochester and Shannon dated their funding proposal 31 August 1955. Dartmouth 1956 →
26. 1956The Logic TheoristAllen Newell, Herbert Simon and Cliff Shaw’s program proved theorems from Principia Mathematica, often called the first AI program. The Golden Years →
27. 1956The Dartmouth summer workshopA loose summer gathering at Dartmouth College gave the field its name and many of its founders. Dartmouth 1956 →
28. 1957The General Problem SolverNewell, Shaw and Simon’s GPS tried to separate general problem-solving strategy from knowledge of any one problem. The Golden Years →
29. 1958Rosenblatt’s perceptronFrank Rosenblatt’s learning machine drew bold press coverage; the hardware Mark I Perceptron followed. The Golden Years →
30. 1958John McCarthy creates LISPLISP became the main language of AI research for three decades. The Golden Years →
31. 1959Samuel popularises the term “machine learning”His paper on checkers described programming computers to learn from experience. The Golden Years →
32. 1961Unimate starts work at General MotorsThe first industrial robot arm handled hot die-cast parts on an assembly line in New Jersey. The Golden Years →
33. 1965Dendral begins at StanfordEdward Feigenbaum, Joshua Lederberg and Bruce Buchanan’s system inferred molecular structures, the first expert system. Expert systems →
34. 1965Ivakhnenko and Lapa’s multilayer learningIn Soviet Ukraine, Alexey Ivakhnenko’s Group Method of Data Handling trained networks layer by layer, an early form of deep learning. Pioneers of AI →
35. 1966Weizenbaum publishes ELIZAA simple pattern-matching program that imitated a psychotherapist convinced some users it understood them. The Golden Years →
36. 1966The ALPAC reportA US committee judged machine translation slow and costly, and funding for it collapsed. The First AI Winter →
37. 1966–72Shakey the robotSRI’s Shakey combined vision, planning and movement; its work gave us the A* search algorithm. The Golden Years →
38. 1969Minsky and Papert’s PerceptronsThe book proved the limits of single-layer perceptrons and helped turn funding away from neural networks. The First AI Winter →
39. 1970Linnainmaa’s reverse-mode differentiationIn a Helsinki master’s thesis, Seppo Linnainmaa described the method that underlies backpropagation. Pioneers of AI →
40. 1970Winograd’s SHRDLUTerry Winograd’s program held typed conversations about a simulated world of blocks, impressive but confined to its toy world. The Golden Years →
41. 1972PrologAlain Colmerauer and Philippe Roussel created the logic-programming language in Marseille. The Golden Years →
42. 1973The Lighthill reportSir James Lighthill’s review for the UK Science Research Council condemned AI’s failure to deliver, and British funding was cut sharply. The First AI Winter →
43. 1974Werbos describes backpropagation for neural networksPaul Werbos’s Harvard PhD thesis showed how to train multilayer networks; it went largely unnoticed for a decade. The connectionist revival →
44. 1976MYCIN describedEdward Shortliffe’s rule-based system for diagnosing blood infections performed as well as specialists in tests, but was never used routinely. Expert systems →
45. 1979The Stanford Cart crosses a roomHans Moravec’s camera-guided cart crossed a chair-filled room, taking about five hours: perception was hard. Lessons from History →
46. 1980XCON goes into use at DECJohn McDermott’s rule-based system configured computer orders, saving its maker millions a year and launching the expert-system boom. Expert systems →
47. 1980Searle’s Chinese Room argumentJohn Searle argued that manipulating symbols by rule is not the same as understanding them. Expert systems →
48. 1980Fukushima’s neocognitronKunihiko Fukushima’s layered vision network, published in English in 1980, anticipated convolutional networks. The connectionist revival →
49. 1982Japan’s Fifth Generation project beginsA ten-year national programme for logic-based “knowledge processing” computers; it triggered rival programmes in the US and Europe. Expert systems →
50. 1982Hopfield networksJohn Hopfield showed how a network of neurons could store memories as stable states, linking neural networks to physics. The connectionist revival →
51. 1984Cyc beginsDouglas Lenat set out to hand-code the common-sense knowledge that expert systems lacked. Expert systems →
52. 1985The Boltzmann machineDavid Ackley, Geoffrey Hinton and Terry Sejnowski described a stochastic network that learned from statistical physics principles. The connectionist revival →
53. 1986Backpropagation popularisedRumelhart, Hinton and Williams’s paper in Nature showed multilayer networks learning useful internal representations. The connectionist revival →
54. 1987The Lisp machine market collapsesCheaper general-purpose workstations undercut specialised AI hardware, a trigger for the second AI winter. Expert systems →
55. 1988Pearl’s Probabilistic Reasoning in Intelligent SystemsJudea Pearl’s book put Bayesian networks at the centre of reasoning under uncertainty. The statistical turn →
56. 1988Moravec’s paradoxIn Mind Children, Hans Moravec observed that skills easy for a toddler are hard for machines, and vice versa. Lessons from History →
57. 1989LeCun reads handwritten postcodesYann LeCun and colleagues at Bell Labs trained a convolutional network with backpropagation to read US zip codes. The connectionist revival →
58. 1992TD-GammonGerald Tesauro’s backgammon program taught itself by playing against itself, reaching near world-champion level. The statistical turn →
59. 1995Support-vector networksCorinna Cortes and Vladimir Vapnik’s paper made support-vector machines a leading machine-learning method for a decade. The statistical turn →
60. 1997Deep Blue beats KasparovIn May 1997 IBM’s chess machine won a six-game match against the world champion, by brute-force search plus expert tuning. The statistical turn →
61. 1997Long short-term memorySepp Hochreiter and Jürgen Schmidhuber’s LSTM let recurrent networks remember over long sequences. Attention Is All You Need →
62. 1998LeNet-5 and MNISTLeCun, Bottou, Bengio and Haffner’s paper on gradient-based document recognition became a deep-learning classic. The connectionist revival →
63. 2001Random forestsLeo Breiman’s ensembles of decision trees became a dependable workhorse of practical machine learning. The statistical turn →
64. 2003A neural probabilistic language modelYoshua Bengio and colleagues learned word embeddings and next-word prediction together, a direct ancestor of today’s LLMs. The statistical turn →
65. 2005Stanley wins the DARPA Grand ChallengeStanford’s car drove 212 km of desert without a driver, guided largely by machine learning. The statistical turn →
66. 2006Deep belief networksHinton, Osindero and Teh showed a way to train deep networks layer by layer, reviving interest in “deep learning”. The deep learning revolution →
67. 2009ImageNetFei-Fei Li’s team released a dataset that grew to over 14 million labelled images; its challenge began in 2010. The deep learning revolution →
68. 2011Watson wins Jeopardy!IBM’s question-answering system beat champions Ken Jennings and Brad Rutter in February 2011. The statistical turn →
69. 2011Siri launches on the iPhone 4SVoice assistants reached the mass market. The statistical turn →
70. 2012AlexNet wins ImageNetAlex Krizhevsky, Ilya Sutskever and Geoffrey Hinton’s GPU-trained network cut the error rate dramatically: the starting gun of the deep-learning era. The deep learning revolution →
71. 2013Word2vecTomas Mikolov and colleagues at Google made word embeddings fast and famous: king − man + woman ≈ queen. The deep learning revolution →
72. 2013DeepMind’s Atari playerA deep Q-network learned to play Atari games from raw pixels and the score. The deep learning revolution →
73. 2014Generative adversarial networksIan Goodfellow and colleagues pitted a generator against a discriminator, producing increasingly realistic images. The deep learning revolution →
74. 2014Sequence-to-sequence learning and attentionSutskever, Vinyals and Le translated with LSTMs; Bahdanau, Cho and Bengio added attention. Attention Is All You Need →
75. 2015ResNetKaiming He and colleagues at Microsoft Research trained networks over 150 layers deep using skip connections. The deep learning revolution →
76. 2015OpenAI foundedA non-profit research lab announced in December 2015 with the stated aim of advancing digital intelligence to benefit humanity as a whole. The Generative Era →
77. 2016AlphaGo beats Lee SedolIn March 2016 DeepMind’s program won 4–1 in Seoul, a decade earlier than many experts expected. The deep learning revolution →
78. 2016Neural machine translation in Google TranslateGoogle switched to an LSTM-based neural system, sharply improving quality. The deep learning revolution →
79. 2017“Attention Is All You Need”Vaswani and seven co-authors introduced the Transformer. Attention Is All You Need →
80. 2017AlphaZeroDeepMind’s single algorithm taught itself chess, shogi and Go from the rules alone. The deep learning revolution →
81. 2018GPT-1 and BERTOpenAI and Google showed that pretraining a Transformer on unlabelled text, then fine-tuning, beat task-specific designs. Attention Is All You Need →
82. 2018Gender ShadesJoy Buolamwini and Timnit Gebru found commercial face-analysis systems were far less accurate for darker-skinned women. Pioneers of AI →
83. 2019GPT-2 and staged releaseOpenAI’s 1.5-billion-parameter model was withheld at first over misuse worries, then released in full in November. Attention Is All You Need →
84. 2019Turing Award for deep learningBengio, Hinton and LeCun received the 2018 ACM Turing Award, announced in March 2019. Pioneers of AI →
85. 2019The Bitter LessonRich Sutton argued that general methods leveraging computation beat hand-built knowledge in the long run. Lessons from History →
86. 2020Scaling laws and GPT-3Kaplan and colleagues found loss falls predictably with scale; GPT-3 (175 billion parameters) learned tasks from examples in its prompt. Attention Is All You Need →
87. 2020Denoising diffusion modelsHo, Jain and Abbeel made diffusion a leading way to generate images. The Generative Era →
88. 2020AlphaFold 2 at CASP14DeepMind’s system predicted protein structures with accuracy comparable to experiment for many targets. Attention Is All You Need →
89. 2021CLIP and DALL·EOpenAI linked images and text, and generated pictures from prompts. The Generative Era →
90. 2021GitHub CopilotAn AI pair-programmer, built on OpenAI’s Codex, entered programmers’ editors as a technical preview. The Generative Era →
91. 2022The image generators go publicDALL·E 2 (April), Midjourney (July) and the openly released Stable Diffusion (August). The Generative Era →
92. 2022ChatGPTReleased on 30 November 2022; an analyst estimate put it at about 100 million monthly users two months later. The Generative Era →
93. 2023GPT-4, Claude, LLaMA and BardA burst of releases between February and March 2023 started an industry-wide race. The Generative Era →
94. 2023Bletchley Park AI Safety Summit28 countries and the EU signed a declaration on frontier-AI risk in November 2023. The Generative Era →
95. 2023GeminiGoogle DeepMind’s natively multimodal model family was introduced in December 2023. The Generative Era →
96. 2024The EU AI Act enters into forceThe first comprehensive AI law took effect on 1 August 2024, with obligations phasing in over several years. The Generative Era →
97. 2024Nobel Prizes for AIPhysics to Hopfield and Hinton; Chemistry to Baker, Hassabis and Jumper. The Generative Era →
98. 2024Reasoning modelsOpenAI’s o1 (September) spent extra computation “thinking” before answering. The Generative Era →
99. 2025Open-weight reasoningDeepSeek’s R1 (January) showed strong reasoning in a model anyone could download. The Generative Era →
100. 2025–26Agents and coding assistants spreadAI systems that take multi-step actions with tools, especially for software work, became mainstream products; reliability remains a work in progress. The Generative Era →

Entries from 2023 onwards describe a fast-moving field and are accurate as of 2026 As of 2026.

Patterns in the timeline

Scroll the full list and three patterns stand out. First, ideas usually arrive long before they work. Backpropagation was described in 1970 and 1974 but took off only after 1986, and deep networks needed until 2012 for enough data and GPU power. Second, progress clusters: quiet decades are followed by bursts, as in 1956–1966, 2012–2016 and 2022 onwards. Third, the approach in fashion keeps changing, from symbolic AI and logic to expert systems, to statistical machine learning, to deep learning and the Transformer, while ideas from the losing side often return later. Each AI winter followed a period when expectations ran ahead of results. Lessons from History asks what these patterns mean for today.

What happened next

The timeline is a list of events; history is made by people. Meet them, including many who rarely get the credit, in Pioneers of AI.

Key terms on this page

Sources & further reading

• McCarthy, J., Minsky, M., Rochester, N. & Shannon, C., A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence (1955).
• Turing, A., Computing Machinery and Intelligence, Mind 59 (1950).
• McCorduck, P., Machines Who Think (1979; 2nd edn 2004).
• Nilsson, N., The Quest for Artificial Intelligence (2010).
• Mitchell, M., Artificial Intelligence: A Guide for Thinking Humans (2019).
• Russell, S. & Norvig, P., Artificial Intelligence: A Modern Approach, 4th edn, chapter 1 history (2020).