From chess to chatbots: AI history and why a machine could win

Milestones: what each one showed

1950Alan Turing asks whether machines can think and proposes the imitation game: can a person tell, in a written conversation, that they are talking to a machine? (original paper)
1956A summer project at Dartmouth establishes AI as a field. The proposal promised “significant advance” on selected problems in one summer — we now know it was work for decades.
1965I. J. Good describes a possible intelligence explosion: machines could speed up the development of further intelligent machines.
1997IBM's Deep Blue beats Garry Kasparov in a six-game match — a breakthrough in a specialised task. (IBM)
2012The neural network AlexNet wins an image-recognition contest by a landslide and starts the deep-learning wave. (paper)
2013The Czech computer scientist Tomáš Mikolov and colleagues at Google introduce word2vec — a method that learns from large amounts of text to express words as numbers, so that their meanings can be calculated with: KING − MAN + WOMAN comes out as QUEEN. According to CIIRC CTU, the work laid foundations for today's language models; in 2023 the authors received the NeurIPS Test of Time Award for it. (paper, CIIRC CTU, in Czech)
2016AlphaGo beats Lee Sedol 4–1. Go has more possible positions than there are atoms in the observable universe. (DeepMind)
2017The transformer paper is published — the architecture today's language models are built on (see How AI works).
2020–2021AI enters science: DeepMind's AlphaFold 2 model predicts the 3D shape of proteins with accuracy close to laboratory measurements (CASP14 assessment). (Nature, what exactly it does)
2022On 30 November ChatGPT launches and generative AI reaches the general public. (announcement)
2024The Nobel Prize in Chemistry goes to David Baker for computational protein design and to Demis Hassabis and John Jumper for protein structure prediction. (Nobel Prize) The Nobel Prize in Physics goes to John Hopfield and Geoffrey Hinton for foundations of machine learning with neural networks.

Move 37

In the second game against Lee Sedol, AlphaGo played a move that commentators first took for a mistake. By the program's own estimate, a human would have played it with a probability of roughly one in ten thousand. The move turned out to be excellent and decided the game. For the first time the world saw that a machine need not just imitate people — it can find approaches beyond ordinary human intuition.

This history is not a straight line that inevitably ends in superintelligence. It shows that different methods can beat humans in important domains — but transfer to a different setting always needs further evidence.

Why could a machine outperform a human?

The human brain is remarkable: on about 20 watts, the power of a dim light bulb, it does things no computer can. But it is not a known physical maximum of intelligence. That does not mean we know how to build a far more capable system. Digital systems do have several possible advantages:

CopyingTraining a doctor takes ten years. A finished model can be deployed many times without repeating training. Copies still need hardware and energy — and repeat the same mistakes.

Parallel workMany systems can explore different hypotheses at once. But a thousand copies of the same wrong assumption do not make a thousand independent opinions.

Speed and scaleComputers do some operations incredibly fast and can be combined into huge systems. A brain has to fit in a skull.

Memory and operationDigital storage holds enormous data and machines don't tire. But a model has no flawless access to everything it saw, and over long work it can pile up errors.

A machine’s parts are orders of magnitude faster. That is not a measure of intelligence, though: brains and chips compute completely differently, and the brain makes up for it with how cleverly it is wired — on about 20 watts.

So the practical question is not “how many times faster is a chip than a brain” but “what work can the system reliably finish, and at what cost”. No known law says human intelligence is the ceiling — we look more like the first species smart enough to build a civilisation, not the smartest possible one.