The Last Invention/AI timeline
A timeline of artificial intelligence
Eighty years from a mathematical sketch of a neuron to a product used by hundreds of millions of people. These are the dates that mattered in the history of AI, and why.
Foundations · 1943–1956
The first artificial neuron
Warren McCulloch and Walter Pitts publish a mathematical model of how a nerve cell could compute. Every neural network since descends from it.
Turing asks the question
Alan Turing’s paper “Computing Machinery and Intelligence” opens by asking whether machines can think, then replaces the question with a test: can a machine hold a conversation that passes for human?
The field gets a name
John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon organise a summer workshop at Dartmouth College. McCarthy’s proposal for it introduces the term “artificial intelligence”.
Early optimism and the first winter · 1958–1973
The perceptron
Frank Rosenblatt demonstrates a machine that learns to recognise simple patterns from examples instead of being programmed with rules.
The last invention
I. J. Good argues that a machine more intelligent than any person would design still better machines, setting off an intelligence explosion.
ELIZA
Joseph Weizenbaum’s chatbot at MIT imitates a therapist using simple pattern matching. Some users confide in it as if it understood them, which alarms its creator.
The limits of perceptrons
Minsky and Seymour Papert show mathematically what single-layer neural networks cannot do. Interest and funding move away from neural networks for more than a decade.
The first AI winter
A report for the British government by James Lighthill concludes that AI has not delivered on its promises. Funding is cut in Britain, and American funders pull back as well.
Expert systems and the second winter · 1980–1987
Expert systems go to work
Digital Equipment Corporation uses a rule-based program called XCON to configure computer orders. Companies spend the decade building systems that capture expert knowledge as rules.
Backpropagation
David Rumelhart, Geoffrey Hinton and Ronald Williams show how networks with several layers can be trained efficiently. It becomes the standard way to train neural networks.
The second AI winter
The market for specialised AI computers collapses and expert systems prove costly to maintain. Funding dries up again.
Machines start winning · 1997–2011
Deep Blue beats Kasparov
IBM’s chess computer defeats the world champion Garry Kasparov over six games in New York.
Watson wins Jeopardy!
IBM’s question-answering system beats two of the quiz show’s most successful champions on television.
The deep learning era · 2012–2016
AlexNet
A deep neural network built by Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton wins the ImageNet image-recognition contest by a wide margin. Within a few years deep learning dominates AI research.
Superintelligence, the book
Nick Bostrom’s Superintelligence makes the risks of machines smarter than people a mainstream subject. What the term means.
AlphaGo beats Lee Sedol
DeepMind’s program wins four games to one in Seoul against one of the strongest Go players in the world, a result many had expected to be a decade away.
Transformers and language models · 2017–2024
The transformer
Eight researchers at Google publish “Attention Is All You Need”. Their network design, the transformer, becomes the basis of nearly every large language model.
GPT-3 and AlphaFold 2
OpenAI’s GPT-3 writes fluent text from a short prompt. DeepMind’s AlphaFold 2 predicts the shapes of proteins with accuracy close to laboratory methods, a problem that had been open for fifty years.
ChatGPT
OpenAI releases ChatGPT on 30 November. It reaches an estimated hundred million users in two months, and AI becomes something almost everyone has tried.
GPT-4 and a summit at Bletchley Park
GPT-4 arrives in March. In November, governments meet for the first international AI safety summit at Bletchley Park, where Turing and Good once broke codes.
Nobel Prizes for AI
The physics prize goes to John Hopfield and Geoffrey Hinton for foundational work on neural networks. Half of the chemistry prize goes to Demis Hassabis and John Jumper for AlphaFold.
Next
Superintelligence
A machine that exceeds the best people at practically everything. Not yet.
Keep reading
What is superintelligence?
An intellect that outperforms the best human minds in practically every field, and how it differs from AI and AGI.
Read the explainer →The intelligence explosion
The 1965 argument that a smart enough machine would improve itself, the case for it, and the reasons to doubt it.
Read the argument →