Index
All 243 milestones in the history of AI, by era and by field, from 1763 to today, each with its entry, the people behind it, and its sources.
Era 1 · The foundations (1763–1942)
AI does not yet exist as a discipline. Six lines of work advance separately and produce the pieces that will later be combined: formal logic, statistical inference, the neuron, the programmable machine, the notion of computation, and the logic circuit.
Logic and the foundations of mathematics
- The Mathematical Analysis of Logic (1847)
- An Investigation of the Laws of Thought (1854)
- Begriffsschrift (1879)
- Hilbert's 23 problems (August 8, 1900)
- Principia Mathematica (1910)
- The Entscheidungsproblem (1928)
- The incompleteness theorems (1931)
Probability and statistics
- Bayes's essay (December 23, 1763)
- The method of least squares (1805)
- Théorie analytique des probabilités (1812)
- Markov chains (1906)
- Maximum likelihood (April 19, 1922)
- Grundbegriffe der Wahrscheinlichkeitsrechnung (1933)
Neuroscience and the psychology of learning
- Golgi's stain (1873)
- The neuron doctrine (May 1888)
- The word "neuron" (1891)
- The word "synapse" (1897)
- The law of effect (1898)
- The conditioned reflex (April 23, 1903)
- Nobel Prize for Golgi and Cajal (1906)
- The all-or-none principle in nerve (February 1914)
- Recurrent neural circuits (August 1933)
Calculating machines and programmable computing
- The Analytical Engine (1837)
- Ada Lovelace's notes (August 1843)
- El Ajedrecista and the Essays on Automatics (1912)
- Z3 (May 12, 1941)
Computability
- The lambda calculus and an unsolvable problem (April 1936)
- Finite Combinatory Processes (September 1936)
- On Computable Numbers (November 30, 1936)
Information and circuits
- Transmission of Information (July 1928)
- Kotelnikov's sampling theorem (1933)
- Nakashima's theory of relay circuits (September 1935)
- Shannon's thesis on relay circuits (1937)
- Shestakov's algebra of relay circuits (1938)
Era 2 · AI is born (1943–1979)
In 1943, logic and neuroscience meet for the first time in a mathematical model of the neuron. In 1955 the field gets its name. For two decades, two programs coexist, the symbolic and the connectionist, until the cutbacks of 1966 to 1974 stall them both.
Cybernetics, information, and electronic computers
- Behavior, Purpose and Teleology (January 1943)
- Colossus (January 1944)
- First Draft of a Report on the EDVAC (June 30, 1945)
- ENIAC (December 10, 1945)
- The Macy conferences (March 8, 1946)
- Cybernetics (1948)
- The Manchester Baby (June 21, 1948)
- A Mathematical Theory of Communication (July 1948)
- Dynamic Programming (1957)
- Some Moral and Technical Consequences of Automation (May 6, 1960)
- Speculations Concerning the First Ultraintelligent Machine (1965)
Neural networks and learning
- The McCulloch-Pitts neuron (December 1943)
- Hebb's rule (1949)
- SNARC (1951)
- Stochastic approximation (September 1951)
- Samuel's checkers program (1952)
- The perceptron (1957)
- Hubel, Wiesel, and the visual cortex (October 1959)
- ADALINE (1960)
- Forerunners of backpropagation (October 1960)
- GMDH (1965)
- A multilayer network trained with stochastic gradient descent (June 1967)
- Vapnik-Chervonenkis theory (1968)
- Perceptrons (1969)
- Reverse-mode automatic differentiation (1970)
- Werbos's dissertation (1974)
- Genetic algorithms (1975)
- The neocognitron (October 1979)
Symbolic AI
- Intelligent Machinery (1948)
- Programming a Computer for Playing Chess (March 1950)
- Computing Machinery and Intelligence (October 1950)
- The Georgetown-IBM experiment (January 7, 1954)
- Logic Theorist (1955)
- The Dartmouth proposal (August 31, 1955)
- The Dartmouth workshop (June 1956)
- General Problem Solver (1957)
- LISP (1958)
- The Mechanisation of Thought Processes symposium (November 24, 1958)
- DENDRAL (1965)
- The resolution principle (January 1965)
- ELIZA (January 1966)
- The A* algorithm (1968)
- SHRDLU (1968)
- The frame problem (1969)
- STRIPS (December 1971)
- Prolog (1972)
- MYCIN (1972)
- Inverse document frequency (1972)
- Wu's method (1977)
The first winter
- The ALPAC report (November 1966)
- The Mansfield Amendment (1969)
- The Lighthill report (1973)
- The DARPA cuts (1974)
Era 3 · From the expert boom to the statistical turn (1980–2011)
Expert systems take symbolic AI into business, and their market collapses in 1987. Meanwhile, probability, neural networks, and reinforcement learning mature with almost no public attention. By the end of the era, the two missing ingredients arrive: programmable GPUs and massive datasets.
Expert systems and the second winter
- R1 (XCON) (1980)
- Searle's Chinese room (September 1980)
- The Fifth Generation project begins (1982)
- Cyc (1984)
- The "AI winter" warning (1984)
- The collapse of the LISP machine market (1987)
- The Strategic Computing Initiative is cut back (1987)
- Fuzzy control on the Sendai subway (July 15, 1987)
- The Fifth Generation project ends (March 1992)
- Deep Blue beats Kasparov (May 11, 1997)
Statistical learning and probabilistic models
- Learned decision trees (1984)
- Valiant's PAC learning (November 1984)
- Bayesian networks (1985)
- IBM's statistical machine translation (1988)
- Hidden Markov models in speech (February 1989)
- AdaBoost (1995)
- Random forests (1995)
- Support vector machines (September 1995)
- The neural probabilistic language model (February 2003)
Neural networks
- Kohonen's self-organizing maps (1982)
- The Hopfield network (April 1982)
- The Boltzmann machine (January 1985)
- Backpropagation by Rumelhart, Hinton, and Williams (October 1986)
- LeCun's convolutional network (1989)
- LSTM (November 1997)
- LeNet-5 and MNIST (1998)
- CIFAR's neural computation program (2004)
- Deep belief nets (July 2006)
- Deep networks in speech recognition (August 2011)
Reinforcement learning
- Temporal differences (August 1988)
- Q-learning (1989)
- TD-Gammon (1992)
- Convergence of Q-learning (May 1992)
- Reinforcement Learning: An Introduction (1998)
- Monte Carlo tree search (May 2006)
Robotics and perception
- Dickmanns's robot cars (1986)
- Brooks's subsumption architecture (1986)
- ALVINN (1989)
- SIFT and the Viola–Jones face detector (September 1999)
- Stanley wins the DARPA Grand Challenge (October 8, 2005)
- Watson wins Jeopardy! (February 14, 2011)
- Siri (October 4, 2011)
Hardware and data
- CUDA (February 15, 2007)
- ImageNet (June 2009)
- Deep networks trained on GPUs (June 2009)
- DeepMind is founded (2010)
- DanNet (2011)
- Google Brain is born (2011)
Era 4 · Deep learning and foundation models (2012–2021)
In 2012, AlexNet shows that deep networks, GPUs, and massive data beat hand-designed methods. In 2017, the Transformer offers an architecture that scales. By 2021, the field is talking about "foundation models": a single pretrained model that adapts to many tasks.
Deep learning
- AlexNet (September 30, 2012)
- word2vec (January 16, 2013)
- Sequence to sequence (June 3, 2014)
- The attention mechanism (September 1, 2014)
- ResNet (December 10, 2015)
Transformers and language models
- The Transformer (June 12, 2017)
- ULMFiT (January 18, 2018)
- ELMo, GPT-1, and BERT (February 15, 2018)
- GPT-2 (February 14, 2019)
- Scaling laws (January 23, 2020)
- GPT-3 (May 28, 2020)
- Vision Transformer (October 22, 2020)
- Scale reaches China and Korea (April 26, 2021)
- GitHub Copilot (June 29, 2021)
- The term "foundation model" (August 16, 2021)
Deep reinforcement learning
- DQN (December 19, 2013)
- AlphaGo beats Fan Hui (October 5, 2015)
- AlphaGo beats Lee Sedol (March 9, 2016)
- PPO (July 20, 2017)
- AlphaGo Zero and AlphaZero (October 19, 2017)
- OpenAI Five beats the Dota 2 world champions (April 13, 2019)
- AlphaStar (October 30, 2019)
Generative image models
- Variational autoencoder (December 20, 2013)
- Generative adversarial networks (June 10, 2014)
- The first diffusion model (March 12, 2015)
- Score-based generative models (July 12, 2019)
- DDPM (June 19, 2020)
- DALL·E and CLIP (January 5, 2021)
AI for science
- AlphaFold 1 (December 2018)
- AlphaFold 2 (November 2020)
- AlphaFold 2 in Nature and an open database (July 15, 2021)
- RoseTTAFold (July 15, 2021)
Alignment
- Reinforcement learning from human preferences (June 12, 2017)
- Bias and accountability in AI (February 23, 2018)
- RLHF for language models (September 18, 2019)
Infrastructure and institutions
- Google acquires DeepMind (January 26, 2014)
- TensorFlow, Chainer, and PyTorch (November 9, 2015)
- OpenAI is announced (December 11, 2015)
- TPU (May 18, 2016)
- Governments enter AI (March 2017)
- OpenAI adopts a capped-profit model (March 11, 2019)
- Turing Award for Bengio, Hinton, and LeCun (March 27, 2019)
- Anthropic is founded (January 2021)
Era 5 · Generative AI, reasoning, and agents (2022–2026)
ChatGPT turns language models into a mass-market product on November 30, 2022. Since then, progress has run along three lines: models that reason before they answer, models that act with tools, and a race for compute that has become a matter of state.
Frontier models
- ChatGPT (November 30, 2022)
- LLaMA (February 24, 2023)
- GPT-4 and Claude (March 14, 2023)
- Gemini 1.0 (December 6, 2023)
- Mixtral 8x7B (December 11, 2023)
- Claude 3 (March 4, 2024)
- GPT-4o (May 13, 2024)
- DeepSeek-V3 (December 26, 2024)
- Grok 3 (February 17, 2025)
- Qwen3 (April 29, 2025)
- Kimi K2 (July 11, 2025)
- GPT-5 (August 7, 2025)
- Gemini 3 Pro (November 18, 2025)
- Claude Opus 4.5 (November 24, 2025)
- Claude Opus 4.6 (February 5, 2026)
- Gemini 3.1 Pro (February 19, 2026)
- GPT-5.4 (March 5, 2026)
- Claude Mythos Preview and Project Glasswing (April 7, 2026)
- Muse Spark (April 8, 2026)
- GPT-5.5 (April 23, 2026)
- DeepSeek V4 (April 24, 2026)
- Claude Fable 5 and Claude Mythos 5 (June 9, 2026)
- Claude Opus 5 (July 24, 2026)
- GPT-6 Astra (September 3, 2026)
- Claude Opus 5.5 (September 22, 2026)
- Gemini 4 Argon (September 30, 2026)
Alignment and reasoning
- Chain of thought (January 28, 2022)
- InstructGPT (March 4, 2022)
- Constitutional AI (December 15, 2022)
- Tree of thoughts (May 17, 2023)
- DPO (May 29, 2023)
- Interpretable features in Claude 3 Sonnet (May 21, 2024)
- OpenAI o1 (September 12, 2024)
- Alignment faking (December 18, 2024)
- DeepSeek-R1 (January 20, 2025)
- Claude 3.7 Sonnet (February 24, 2025)
- Gemini 2.5 Pro (March 25, 2025)
- Gold at the International Mathematical Olympiad (July 19, 2025)
Agents
- ReAct (October 6, 2022)
- Toolformer (February 9, 2023)
- Generative agents (April 7, 2023)
- SWE-bench (October 10, 2023)
- Computer use in Claude (October 22, 2024)
- Model Context Protocol (November 25, 2024)
- Operator, Deep Research, and Claude Code (January 23, 2025)
- The time horizon of agents (March 18, 2025)
- ChatGPT agent (July 17, 2025)
Images, video, and the physical world
- DALL·E 2 (April 6, 2022)
- Stable Diffusion (August 22, 2022)
- Sora (February 15, 2024)
- Gemini Robotics (March 12, 2025)
- Genie 3 (August 5, 2025)
AI for science and recognition
- AlphaFold 3 (May 8, 2024)
- Nobel Prize in Physics for Hopfield and Hinton (October 8, 2024)
- Nobel Prize in Chemistry for Hassabis, Jumper, and Baker (October 9, 2024)
- Turing Award for Barto and Sutton (March 5, 2025)
- AlphaEvolve (May 14, 2025)
Infrastructure and compute
- Chinchilla (March 29, 2022)
- Chip export controls on China (October 7, 2022)
- xAI's Colossus (September 2024)
- Stargate (January 21, 2025)
- The expansion of Colossus (December 30, 2025)