Timeline of Major Events
Key moments in the history of AI, from 1943 to 2026.
The DawnThe Golden AgeThe First WinterThe Expert Systems EraThe Second WinterThe Rise of Statistical LearningThe Eve of Deep LearningThe Vision RevolutionThe Transformer EraThe Generative AI Wave
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
Turing publishes *On Computable Numbers*, introducing the Turing machine and laying the mathematical foundation for computability and modern computing.McCulloch and Pitts publish the first mathematical model of neurons, showing that brain activity can be described in mathematical terms.Wiener publishes *Cybernetics*, arguing that machines and living organisms share common feedback mechanisms.Hebb publishes *The Organization of Behavior*, formulating Hebbian learning: "neurons that fire together, wire together."Turing publishes *Computing Machinery and Intelligence*, proposing the "imitation game" (the Turing test).Shannon publishes *Programming a Computer for Playing Chess*, opening the field of computer game-playing research.Minsky and Edmonds build SNARC, the first hardware neural network in history.Samuel develops a checkers-playing program at IBM, pioneering the field of machine learning.Newell, Simon, and Shaw build Logic Theorist, the first AI program able to prove mathematical theorems automatically.The Dartmouth workshop convenes; McCarthy coins the term "artificial intelligence," giving birth to AI as a discipline.Rosenblatt invents the Perceptron, hailed worldwide as a learnable artificial brain.McCarthy invents the LISP programming language, which becomes the standard tool of AI research.Widrow and Hoff propose ADALINE and the LMS algorithm, carrying perceptron ideas into engineering practice.McCarthy and Minsky co-found the AI Group at MIT, which later grows into the MIT AI Lab.Samuel's checkers program defeats a strong amateur, popularizing the concept of "machine learning."McCarthy leaves MIT to found the Stanford AI Lab (SAIL), splitting U.S. AI research into East and West Coast camps.Feigenbaum develops DENDRAL at Stanford, the first successful expert system.Weizenbaum creates the ELIZA chatbot; users astonishingly form emotional attachments to a mere program.The ALPAC report questions the feasibility of machine translation, leading to deep cuts in U.S. MT funding.Minsky and Papert publish *Perceptrons*, mathematically proving the limits of single-layer networks and pushing neural networks into a deep freeze.Colmerauer and colleagues develop Prolog in Marseille, making logic programming the standard-bearer of European AI.Winograd completes SHRDLU at MIT, manipulating a blocks world through natural-language commands.The Lighthill Report sharply criticizes AI for failing to deliver on its promises, and the UK becomes the first to slash funding.DARPA cuts AI project budgets, spreading the "AI winter" across the U.S. and pushing many researchers into other fields.Minsky publishes *A Framework for Representing Knowledge*, introducing "frames" into cognitive modeling.Weizenbaum publishes *Computer Power and Human Reason*, a systematic reflection on the ethical limits of AI.Hans Moravec's Stanford Cart performs the first autonomous outdoor obstacle-avoidance navigation.Carnegie Mellon establishes the Robotics Institute, the first dedicated robotics research institute in the United States.The XCON/R1 expert system goes into production at DEC, saving roughly $40 million a year and proving AI's commercial value.Japan launches the Fifth Generation Computer project with $850 million in funding, triggering alarm and imitation in the U.S. and Europe.Hopfield publishes his paper on Hopfield networks, bringing statistical physics into neural networks and reigniting connectionist research.Japan establishes the Institute for New Generation Computer Technology (ICOT) to coordinate the Fifth Generation project.Lenat launches the Cyc project at MCC, an audacious attempt to encode all of human common sense into a machine.Hinton and Sejnowski propose the Boltzmann machine, introducing probabilistic inference into neural networks.Rumelhart, Hinton, and Williams publish the backpropagation algorithm in *Nature*, finally making multilayer neural networks trainable.Sejnowski and Rosenberg's NETtalk learns to read English aloud, becoming a classic demonstration of connectionism.Pearl publishes *Heuristics: Intelligent Search Strategies*, laying the groundwork for Bayesian networks and causal reasoning.The AAAI Spring Symposium focuses on the limits of expert systems, marking the start of an industry-wide reckoning.The Lisp machine market collapses; specialized AI hardware companies fold one after another and the expert-system bubble bursts.LeCun builds the LeNet convolutional network at Bell Labs, applying backpropagation to handwritten digit recognition for the first time.Watkins proposes Q-Learning at Cambridge, laying the foundation for modern reinforcement learning.Elman proposes the Simple Recurrent Network (SRN), bringing the time dimension into neural networks.Japan officially declares its Fifth Generation Computer project a failure, having met none of its original goals.*IEEE Spectrum* declares AI to be in its "second winter," and researchers retreat behind quieter labels such as "machine learning."Cortes and Vapnik publish their paper on support vector machines (SVMs), bringing statistical learning theory to its peak.Breiman publishes his paper on Bagging, crystallizing the idea of ensemble learning.IBM's Deep Blue defeats world chess champion Kasparov, shocking the world and dominating front pages.Hochreiter and Schmidhuber publish LSTM, a long short-term memory network that solves the vanishing-gradient problem.LeCun and colleagues publish LeNet-5, establishing the template for the modern convolutional neural network.Page and Brin found Google, with PageRank bringing graph algorithms onto the center stage of the internet.NVIDIA releases the GeForce 256 and coins the term "GPU," planting the seed for the later deep-learning compute revolution.Pearl publishes *Causality*, putting causal inference back on the mainstream research agenda.Breiman publishes his Random Forests paper, turning ensemble learning into the workhorse algorithm of industry.iRobot launches the Roomba vacuum, the first service robot to genuinely enter the home.The first DARPA Grand Challenge for autonomous vehicles is held in the Mojave Desert; every entrant fails to finish.Stanford's Stanley wins the DARPA Grand Challenge, ushering autonomous driving into an engineering era.Hinton and Salakhutdinov propose deep belief networks, bringing the term "deep learning" into the mainstream.Andrew Ng joins Stanford and begins exploring large-scale neural network training on GPUs.Geoffrey Hinton leads the neural network revival at the University of Toronto, work that later spawns the Vector Institute.NVIDIA releases CUDA, turning GPUs into a general-purpose computing platform and paving the way for deep learning.Fei-Fei Li's team releases the ImageNet dataset, with 14 million hand-labeled images planting the seeds of the deep-learning revolution.Andrew Ng and Raina publish *Large-scale Deep Unsupervised Learning Using GPUs*, demonstrating the huge advantage of GPU-based neural network training.Microsoft releases the Kinect, bringing depth sensing and human pose recognition to the consumer market.IBM Watson beats the human champions on *Jeopardy!*, bringing the idea of cognitive computing into public view.Andrew Ng and Daphne Koller co-found Coursera, taking Stanford's machine learning class to a global audience.The Google Brain project quietly launches inside Google X under Andrew Ng and Jeff Dean.Apple launches Siri, putting AI into the pockets of hundreds of millions of users.Google Brain trains a neural network on 16,000 CPU cores that learns to recognize "cats" from YouTube videos on its own.Yoshua Bengio and colleagues expand MILA at the University of Montreal, completing the Canadian "deep learning trinity."AlexNet wins the ImageNet competition by slashing error rates by 10 percentage points, igniting the deep-learning revolution.Mikolov and colleagues at Google publish Word2Vec, making word embeddings a standard NLP tool.DeepMind publishes *Playing Atari with Deep Reinforcement Learning*, with Deep Q-Networks teaching AI to play video games.Google acquires DeepMind for around $600 million, kicking off a global AI acquisition race among tech giants.Goodfellow, in a flash of inspiration at a Montreal bar, invents Generative Adversarial Networks (GANs), opening the era of generative AI.Baidu founds its Institute of Deep Learning (IDL) with Andrew Ng as Chief Scientist, marking China's tech giants entering the AI arena.Sutskever and colleagues publish the Seq2Seq paper, unifying translation and generation tasks under an encoder-decoder architecture.Bahdanau and colleagues introduce the attention mechanism into neural machine translation for the first time.Google open-sources TensorFlow, democratizing deep learning tooling.Kaiming He (何恺明) and colleagues publish ResNet; residual connections push networks past 100 layers and set new records on ImageNet and COCO.OpenAI is announced in San Francisco; Musk and Altman pledge $1 billion to counter "AI being monopolized by tech giants."AlphaGo defeats world Go champion Lee Sedol 4–1, conquering humanity's most complex board game.Facebook AI Research open-sources PyTorch, whose dynamic-graph style quickly wins over the research community.Eight Google researchers publish *Attention Is All You Need*; the Transformer architecture is born and changes everything.China's State Council issues the *New Generation AI Development Plan*, setting the goal of becoming the world's AI innovation hub by 2030.DeepMind releases AlphaZero, surpassing humans at Go, chess, and shogi through self-play alone.OpenAI releases GPT-1, proposing the "generative pre-training plus fine-tuning" recipe.The ACM awards the Turing Award to Hinton, Bengio, and LeCun for their contributions to deep learning.Google releases BERT; bidirectional pre-training sweeps 11 NLP benchmarks.OpenAI announces GPT-2 (1.5B parameters) but delays the full release as "too dangerous," sparking the open-source vs. safety debate.Zhipu AI is founded out of Tsinghua University's KEG lab, kicking off China's LLM startup wave.Cohere is founded in Toronto, taking Transformer-based commercialization into the enterprise market.Microsoft invests $1 billion in OpenAI and signs an exclusive Azure compute partnership.OpenAI releases GPT-3 (175B parameters); few-shot learning shows what looks like emergent capability.Hugging Face closes a Series B; the Transformers library becomes the de facto open-source standard.DeepMind's AlphaFold 2 solves the protein-folding problem at CASP14, hailed as "the biggest breakthrough in 50 years."OpenAI launches DALL·E, putting text-to-image generation in the public eye for the first time.Former OpenAI researchers Dario and Daniela Amodei co-found Anthropic, making AI safety the company's core mission.Inflection, Character.AI and a wave of new AI startups emerge, crystallizing the "AI native" startup paradigm.GitHub launches the Copilot preview, formally putting AI to work as a coding assistant for developers.DeepMind releases AlphaCode, performing at the level of an average competitive programmer.Midjourney enters public beta; diffusion models push AI art onto social platforms.Stability AI open-sources Stable Diffusion, blowing open the entire diffusion-model ecosystem.Anthropic internally rolls out the first version of Claude, opening the Constitutional AI line of work.ChatGPT launches on November 30; reaching 100 million users in two months, it becomes the fastest-growing consumer app in history.Microsoft launches Bing Chat (powered by GPT-4), opening a search war with Google.Meta releases LLaMA with weights available on request, igniting the open-source LLM movement.Musk registers a company under the name X.AI Corp, the embryonic form of xAI.OpenAI releases GPT-4, whose multimodal capabilities stun the industry.Anthropic releases Claude 1, emphasizing long context and safe, steerable behavior.Hinton resigns from Google and publicly warns of the risks of general AI.France's Mistral AI is founded, becoming Europe's standard-bearer for open-source large language models.China's top tech firms release Qwen, ERNIE Bot, Hunyuan and other LLMs in quick succession, opening the "battle of a hundred models."Anthropic releases Claude 2, with a 100K-token context window as its headline feature.xAI is officially founded, with Musk launching a three-way race against OpenAI and Anthropic.Meta releases LLaMA 2 under a free commercial license, finally bringing open-source LLMs into the enterprise.U.S. President Biden signs Executive Order 14110, bringing frontier models under national-security oversight.OpenAI's board abruptly fires Altman; he returns five days later under pressure from staff and investors.OpenAI releases the Sora video model, extending diffusion models and Transformers into the time dimension.Anthropic releases the Claude 3 family; Opus surpasses GPT-4 on several benchmarks.OpenAI releases GPT-4o with native multimodality and low-latency voice interaction.The European Union formally adopts the AI Act, the world's first comprehensive AI law.NVIDIA becomes one of the world's most valuable companies as AI compute demand reshapes the semiconductor landscape.OpenAI releases the o1 family of reasoning models, surfacing the "test-time compute" research direction.Hinton wins the Nobel Prize in Physics while Hassabis and Jumper win the Nobel in Chemistry — the first time AI researchers take Nobels on two fronts in the same year.DeepSeek releases the R1 reasoning model, approaching o1-level performance at a fraction of the cost and igniting a global open-source debate.The U.S., Japan, the UAE and partners announce the Stargate project, planning AI compute infrastructure on the $500 billion scale.Trump signs an executive order revoking Biden's EO 14110, pivoting U.S. AI governance toward an accelerationist stance.General-purpose agent products such as Manus and Operator go live, taking the "AI employee" concept into commercial use.xAI acquires X (formerly Twitter), building an integrated "model + social + data" platform.Anthropic releases the Claude 4 family, with agentic and coding workflows becoming its flagship capabilities.AI coding tools such as Cursor, Devin, and the Codex CLI enter mainstream developer workflows.OpenAI releases GPT-5, further unifying reasoning, tool use, and multimodal capabilities.More than 70 countries and regions worldwide have enacted AI governance frameworks, with AI safety and industrial policy deeply intertwined.AI is deeply embedded in healthcare, education, manufacturing, and research, with embodied intelligence and AI for Science emerging as the new main battlegrounds.