Guide
The AI timeline: key moments from 1950 to today
AI didn't appear overnight with ChatGPT. It's a 75-year story of big ideas, false starts, two "winters" and sudden breakthroughs. Here are the moments that mattered most, and why. Tap an era to jump to it, or filter by type.
The beginnings
1950–1969 · A handful of scientists ask whether machines could think, and give the field its name.
- 1950Research
Alan Turing asks "Can machines think?"
Turing's paper "Computing Machinery and Intelligence" proposes the imitation game, later called the Turing test.
Why it mattered: It framed the question the whole field has chased since.
- 1956Industry & policy
The term "artificial intelligence" is born
John McCarthy, Marvin Minsky and others hold a summer workshop at Dartmouth College under that name.
Why it mattered: Widely seen as the founding moment of AI as a field of research.
- 1958Hardware
The Perceptron learns from examples
Frank Rosenblatt builds a simple neural network that learns to recognise patterns.
Why it mattered: The ancestor of today's neural networks.
- 1966Products
ELIZA, the first chatbot
Joseph Weizenbaum at MIT builds a program that imitates a therapist using simple text patterns.
Why it mattered: People confided in it anyway, an early lesson in how easily we credit machines with understanding.
Winters and expert systems
1970–1996 · Big promises meet slow computers. Funding comes and goes, but key ideas are born.
- 1973Industry & policy
The first "AI winter" begins
The UK's Lighthill report criticises slow progress, and funding is cut in the UK and US.
Why it mattered: Showed the cost of over-promising, a pattern that repeats.
- 1980Products
Expert systems go to work
Rule-based programs such as XCON at Digital Equipment Corporation help companies configure orders and diagnose problems.
Why it mattered: AI's first real commercial success.
- 1986Research
Backpropagation makes neural networks trainable
Rumelhart, Hinton and Williams popularise a method for training networks with several layers.
Why it mattered: Still the core training technique behind deep learning today.
- 1987Industry & policy
The second AI winter
The market for specialised AI computers collapses and expert systems prove costly to maintain.
Why it mattered: Interest and funding drop for years.
- 1989Research
A neural network reads handwriting
Yann LeCun's team at Bell Labs trains a convolutional network to read handwritten digits; versions later read millions of cheques.
Why it mattered: An early practical win for neural networks.
Machines beat champions
1997–2011 · Computers win at chess and quiz shows, and the data and chips for modern AI arrive.
- 1997Big moments
Deep Blue beats the world chess champion
IBM's Deep Blue defeats Garry Kasparov in a six-game match.
Why it mattered: The first time a computer beat a reigning world champion at chess under match conditions.
- 2006Research
Deep learning gets its comeback
Geoffrey Hinton and colleagues show how to train deep networks layer by layer.
Why it mattered: Revived interest in neural networks after years in the wilderness.
- 2007Hardware
GPUs become general-purpose computers
Nvidia releases CUDA, letting graphics chips run any calculation in parallel.
Why it mattered: GPUs later made training huge AI models practical.
- 2009Research
ImageNet: 14 million labelled images
Fei-Fei Li's team releases a huge image dataset and, from 2010, a yearly recognition challenge.
Why it mattered: Gave researchers the data and the scoreboard that sparked the deep learning boom.
- 2011Big moments
IBM Watson wins Jeopardy!
Watson beats two champions of the US quiz show on live TV.
Why it mattered: Showed computers handling tricky natural-language questions.
- 2011Products
Siri puts an assistant in your pocket
Apple launches the Siri voice assistant on the iPhone 4S.
Why it mattered: Made talking to an AI assistant an everyday habit.
Deep learning takes off
2012–2019 · Neural networks trained on GPUs start beating every other approach.
- 2012Research
AlexNet wins ImageNet by a mile
A deep neural network trained on GPUs cuts the error rate dramatically in the ImageNet challenge.
Why it mattered: The moment the field switched to deep learning.
- 2014Research
GANs: machines that generate images
Ian Goodfellow introduces generative adversarial networks, two networks that compete to create realistic images.
Why it mattered: An early leap for generative AI.
- 2016Big moments
AlphaGo beats Lee Sedol at Go
DeepMind's AlphaGo wins 4–1 against one of the world's best Go players.
Why it mattered: Go was thought to be decades away from being mastered by computers.
- 2016Hardware
Chips built just for AI
Google reveals its Tensor Processing Unit (TPU), designed for neural network maths.
Why it mattered: Started the race for specialised AI chips.
- 2017Research
"Attention Is All You Need": the Transformer
Google researchers publish the Transformer, a new way for networks to process language.
Why it mattered: The design behind ChatGPT, Claude, Gemini and nearly every modern AI model.
- 2018Research
Pre-trained language models arrive
OpenAI's first GPT and Google's BERT learn from huge amounts of text, then adapt to many tasks.
Why it mattered: The "train once, use everywhere" idea behind foundation models.
- 2019Industry & policy
Turing Award for deep learning
Yoshua Bengio, Geoffrey Hinton and Yann LeCun receive computing's highest honour.
Why it mattered: Recognised deep learning as the field's defining breakthrough.
- 2019Big moments
GPT-2 is "too dangerous" to release at once
OpenAI releases its text generator in stages, citing misuse concerns.
Why it mattered: Started the public debate about releasing powerful AI.
AI goes mainstream
2020–2023 · Large language models and image generators reach hundreds of millions of people.
- 2020Research
GPT-3 shows the power of scale
OpenAI's 175-billion-parameter model writes, translates and answers questions from a few examples.
Why it mattered: Showed that bigger models trained on more text gain surprising new abilities.
ParametersLarge language model (LLM)Source: en.wikipedia.org
- 2020Research
AlphaFold 2 cracks protein folding
DeepMind's system predicts 3D protein structures with accuracy close to lab experiments.
Why it mattered: Solved a 50-year-old biology problem and changed drug research.
- 2021Products
GitHub Copilot writes code alongside you
An AI assistant inside code editors suggests whole lines and functions.
Why it mattered: The first mass-market AI tool for professional work.
- 2022Products
Anyone can make images from text
DALL·E 2, Midjourney and the openly released Stable Diffusion turn text prompts into pictures.
Why it mattered: Generative AI reaches artists, designers and the public.
- 2022Products
ChatGPT launches
OpenAI releases a free chatbot on 30 November; it reaches an estimated 100 million users within about two months.
Why it mattered: The moment AI became part of everyday life.
- 2023Products
GPT-4 can see as well as read
OpenAI's GPT-4 accepts images and scores highly on many professional and academic exams.
Why it mattered: Multimodal AI goes mainstream.
- 2023Products
Open-weight models take off
Meta releases its Llama models for researchers and companies to download and run.
Why it mattered: Powerful AI no longer only lives in a few companies' data centres.
- 2023Industry & policy
Governments meet on AI safety
The UK hosts the first global AI Safety Summit at Bletchley Park.
Why it mattered: AI risks become a matter for world leaders.
The agent era
2024–today · AI models start reasoning, using tools and doing tasks, not just answering questions.
- 2024Industry & policy
The EU passes the AI Act
The European Union adopts the first comprehensive law regulating AI by risk level.
Why it mattered: Sets rules many global companies now follow.
- 2024Research
Reasoning models think before answering
OpenAI's o1 takes time to work through problems step by step before replying.
Why it mattered: A big jump on maths, science and coding problems, followed by every major lab.
- 2024Industry & policy
Nobel Prizes go to AI pioneers
Physics honours John Hopfield and Geoffrey Hinton; Chemistry honours Demis Hassabis and John Jumper for AlphaFold, alongside David Baker.
Why it mattered: Science's top prizes recognise AI's impact.
- 2025Products
DeepSeek-R1 surprises the industry
Chinese lab DeepSeek releases a strong open reasoning model said to have cost far less to train than rivals.
Why it mattered: Shook assumptions about how much money top AI requires.
- 2025Products
Vibe coding and AI agents go mainstream
Andrej Karpathy coins "vibe coding" as people build apps by describing them to AI, and coding agents take on whole tasks.
Why it mattered: AI shifts from answering questions to doing work.
- 2026Products
GPT-6 rolls out to everyone
OpenAI brings its newest model family to all ChatGPT users, with answers that can include interactive visuals.
Why it mattered: The latest step in the race among AI labs.
How we chose these moments
A milestone makes the list if it changed what AI could do, how people used it, or what the field believed was possible. Dates and “firsts” in AI history are sometimes disputed, so every entry links to a source. Think we’ve missed something important? Tell us.