Artificial Intelligence
From the Turing Test to ChatGPT — the machines that learned to think, every milestone sourced.
A timeline of artificial intelligence, from Alan Turing's 1950 question of whether machines can think to the generative-AI boom of the 2020s. It traces the field's founding at Dartmouth, its booms and 'winters,' the neural-network breakthroughs of deep learning, landmark victories like Deep Blue and AlphaGo, and the large language models — GPT, ChatGPT, GPT-4 — that brought AI to the world. Every event is backed by content-verified sources, including the original research papers.
Events
- 1950Reputable sourceWell documented
Turing's Imitation Game
In his paper 'Computing Machinery and Intelligence,' Alan Turing set aside the question 'Can machines think?' as too vague, and proposed instead an 'imitation game': if an interrogator conversing by text cannot reliably tell a machine from a human, the machine should count as intelligent. It became known as the Turing Test.
Why it matters: Turing's paper framed the central question of artificial intelligence and gave the field its most famous benchmark, decades before the technology existed to attempt it.
Sources - 1956Reputable sourceWell documented
The Dartmouth Workshop
A summer research project at Dartmouth College brought together John McCarthy, Marvin Minsky, Nathaniel Rochester, Claude Shannon, and others to explore whether machines could be made to simulate learning and intelligence. The 1955 proposal contained the first use of the phrase 'artificial intelligence.'
Why it matters: Often called the birthplace of AI, the Dartmouth workshop named and launched the field as a distinct area of research.
SourcesRelated timelines- The Internet & Computing → — The wider story of computing
- 1958Reputable sourceWell documented
The Perceptron
Frank Rosenblatt built the Perceptron, an early artificial neural network that could learn to recognize simple patterns such as letters. It was greeted with enormous enthusiasm — one report called it the first serious rival to the human brain — but in 1969 Minsky and Papert showed a single-layer perceptron could not compute basic functions like XOR.
Why it matters: The Perceptron pioneered the trainable neural network, the idea at the heart of today's AI — even though its early limitations helped trigger the first collapse of the field.
- 1966Peer-reviewedWell documented
ELIZA, the First Chatbot
Joseph Weizenbaum of MIT wrote ELIZA, a program that imitated a psychotherapist by matching patterns in a user's typed words and reflecting them back as questions — all in about 200 lines of code. Weizenbaum was disturbed to find that users formed real emotional attachments to it.
Why it matters: ELIZA was the first program widely seen as conversing in natural language, and the reaction it provoked foreshadowed how readily people would attribute understanding to machines.
- 1966–1972Peer-reviewedWell documented
Shakey the Robot
Built at the Stanford Research Institute, Shakey was the first mobile robot able to perceive its surroundings and reason about its own actions. It could plan routes, move objects, and recover from errors, driven by a planning program called STRIPS.
Why it matters: Shakey combined perception, planning, and action in one machine for the first time, and its methods influenced robotics and AI for decades.
- 1973Peer-reviewedWell documented
The First AI Winter
After years of grand promises that failed to materialize, the British government's 1973 Lighthill Report concluded that AI research had not delivered its promised breakthroughs. Funding was slashed in Britain and the United States, ushering in a prolonged downturn known as an 'AI winter.'
Why it matters: The first AI winter showed a recurring pattern of hype followed by disillusionment that would shape the field's boom-and-bust history.
Sources - 1980sPeer-reviewedWell documented
Expert Systems and the 1980s AI Boom
AI revived in the 1980s around 'expert systems' — programs that encoded the rules of human specialists to solve narrow problems, pioneered by systems like Stanford's MYCIN for diagnosing infections. Corporations invested heavily, and a commercial industry grew up around the technology before it, too, hit a downturn.
Why it matters: Expert systems were AI's first major commercial success, proving the technology could be useful — while their brittleness set up the second AI winter.
- 1986Peer-reviewedWell documented
Backpropagation
In a landmark Nature paper, David Rumelhart, Geoffrey Hinton, and Ronald Williams showed how the 'back-propagation' algorithm could train multi-layer neural networks, letting hidden layers learn useful internal representations of a problem.
Why it matters: Backpropagation solved the training problem that had crippled neural networks and remains the foundational algorithm behind modern deep learning.
- May 1997Peer-reviewedWell documented
Deep Blue Defeats Kasparov
In a six-game rematch in New York, IBM's Deep Blue became the first computer to defeat a reigning world chess champion, Garry Kasparov, under standard tournament conditions, winning 3.5–2.5. The machine used hundreds of processors and custom chips to evaluate millions of positions per second.
Why it matters: Deep Blue's victory was a symbolic milestone — a computer beating the best human at a game long seen as a pinnacle of human intellect.
- February 2011Primary sourceWell documented
IBM Watson Wins Jeopardy!
IBM's Watson, a question-answering system able to interpret natural-language clues, defeated the quiz show Jeopardy!'s two greatest champions, Ken Jennings and Brad Rutter, in a televised match. The room-sized machine drew on thousands of processor cores and a vast store of text.
Why it matters: Watson showed that computers could understand and answer everyday human questions far better than traditional search, hinting at the natural-language AI to come.
Sources- IBM. Watson, Jeopardy! champion · primary
- 2012Peer-reviewedWell documented
AlexNet and the Deep Learning Revolution
A deep neural network called AlexNet, built by Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton and trained on graphics processors, won the ImageNet image-recognition challenge by a huge margin, roughly halving the previous error rate.
Why it matters: AlexNet's dramatic win convinced the field that deep neural networks and GPUs worked, igniting the deep learning revolution that powers modern AI.
Sources - 2014Peer-reviewedWell documented
Generative Adversarial Networks
Ian Goodfellow and colleagues introduced Generative Adversarial Networks (GANs), in which two neural networks — a generator and a discriminator — are trained against each other, one trying to create realistic data and the other trying to spot fakes.
Why it matters: GANs made it possible for machines to generate startlingly realistic images and helped launch the field of generative AI.
Sources - March 2016Peer-reviewedWell documented
AlphaGo Defeats Lee Sedol
DeepMind's AlphaGo defeated Lee Sedol, one of the world's greatest players of Go, four games to one in a match in Seoul. Go, with more possible positions than atoms in the universe, had been considered far beyond the reach of computers for decades to come.
Why it matters: AlphaGo's win — combining deep neural networks with reinforcement learning, and producing moves no human had considered — was a landmark that showed AI could master intuition-heavy tasks.
Sources - 2017Peer-reviewedWell documented
The Transformer
Researchers at Google published 'Attention Is All You Need,' introducing the Transformer — a neural network architecture built entirely on 'attention' mechanisms, dispensing with the recurrence used before. It trained faster and handled long-range context far better.
Why it matters: The Transformer became the foundation of virtually every modern large language model, from BERT to GPT — arguably the single most consequential AI architecture of the era.
Sources - 2018Peer-reviewedWell documented
BERT and the Rise of Large Language Models
Google researchers released BERT, a Transformer-based model pre-trained on huge amounts of text to understand language bidirectionally — reading context from both directions at once. It set new records across a wide range of language tasks.
Why it matters: BERT showed the power of pre-training large models on unlabeled text, the recipe that would drive the explosion of large language models.
- 2020Peer-reviewedWell documented
GPT-3
OpenAI unveiled GPT-3, a language model with 175 billion parameters. Trained simply to predict the next word across enormous amounts of text, it could perform many tasks — translation, question-answering, even basic coding — from a few examples given in the prompt, without task-specific training.
Why it matters: GPT-3 demonstrated that scale alone could produce surprisingly general abilities, setting the stage for the generative AI boom.
Sources - 2021Primary sourceWell documented
DALL·E and AI Image Generation
OpenAI introduced DALL·E, a model that generates original images from written descriptions, combining unrelated ideas into plausible pictures. It and successors like DALL·E 2 brought text-to-image generation into the mainstream.
Why it matters: AI image generators put creative, generative AI directly into the public's hands and raised new questions about art, copyright, and authenticity.
Sources - 2021Peer-reviewedWell documented
AlphaFold Solves Protein Folding
DeepMind's AlphaFold predicted the three-dimensional structures of proteins from their amino-acid sequences with accuracy rivalling laboratory experiments — cracking a 50-year-old grand challenge in biology.
Why it matters: AlphaFold showed AI could make genuine scientific breakthroughs, and its open database of predicted structures has transformed biology and drug discovery.
- November 30, 2022Primary sourceWell documented
ChatGPT Launches
OpenAI released ChatGPT, a conversational interface to a GPT-3.5 language model, fine-tuned with human feedback to follow instructions and answer follow-up questions in dialogue. Free and easy to use, it reached an estimated hundred million users within two months.
Why it matters: ChatGPT put powerful generative AI in front of the general public for the first time, touching off a worldwide boom in AI investment, adoption, and debate.
Sources- OpenAI. Introducing ChatGPT · primary
- March 2023Primary sourceWell documented
GPT-4 and the Generative AI Boom
OpenAI released GPT-4, a larger, more capable multimodal model that could accept images as well as text and scored in the top ranks of human test-takers on many professional and academic exams, including a simulated bar exam. It arrived amid an explosion of rival models and a fierce debate over AI's risks and regulation.
Why it matters: GPT-4 marked the arrival of generative AI as a mainstream technology reshaping work, education, and creativity — and intensified urgent questions about safety, jobs, and control.
Sources- OpenAI. GPT-4 · primary