The ancient game that helped reveal the future of AI.
Go is valuable not because people should try to think like machines, but because it trains distinctly human qualities: judgement, imagination, patience and the courage to revise a plan.
A challenge that computers could not simply calculate
For decades, Go stood apart from other board games. Its rules are concise, but the number of possible positions is so immense that brute-force calculation is not enough. Strong play depends on recognising patterns, balancing local and whole-board priorities, judging uncertainty and choosing among many reasonable possibilities.
Those are exactly the qualities that made Go a great challenge for artificial intelligence. A machine could calculate quickly, yet still struggle with the kind of intuitive evaluation that experienced Go players develop over time.
2016: a landmark moment
In January 2016, Nature published DeepMind’s paper “Mastering the game of Go with deep neural networks and tree search”. AlphaGo had defeated European professional champion Fan Hui 5–0 — the first time a computer program had defeated a professional Go player in an even match.
The achievement combined deep neural networks with tree search. It was not merely a faster calculator. The system learned how to evaluate positions and focus its search on promising moves. Go had become a public demonstration that machine learning could handle problems involving vast possibilities and uncertain judgement.
AlphaGo
Defeated professional player Fan Hui; later defeated Lee Sedol in a celebrated five-game match.
AlphaGo Zero
Learned Go from the rules and self-play, without training on human game records.
AlphaZero
Generalised the self-learning approach across Go, chess and shogi.
AlphaFold
Used deep learning to achieve a major advance in predicting three-dimensional protein structures.
From Go to broader discovery
AlphaGo was followed by AlphaGo Zero and AlphaZero. AlphaZero showed that a single general approach could learn Go, chess and shogi through self-play, starting with the rules rather than human strategic knowledge.
DeepMind later turned to another grand challenge: protein structure prediction. AlphaFold dramatically improved the ability to predict a protein’s three-dimensional structure from its amino-acid sequence, helping researchers investigate biological mechanisms far more efficiently. The work on AlphaFold was recognised as part of the 2024 Nobel Prize in Chemistry.
The connection should be stated carefully: AlphaFold is not a direct version of AlphaGo, and protein biology is not a board game. The meaningful connection is that the same research organisation moved from mastering a difficult environment for learning and decision-making to applying advanced AI methods to a major scientific problem. Go was an important proving ground in that wider story.
Why should humans still learn Go?
The success of AI does not make Go less valuable for people. It makes the game more interesting. A student of Go practises how to:
- see the whole situation before reacting locally;
- compare several plausible choices rather than searching for one obvious answer;
- make decisions with incomplete information;
- learn from mistakes without being defined by them;
- balance ambition with restraint;
- recognise patterns while remaining open to surprising ideas; and
- respect an opponent whose thinking may be entirely different.
Simple rules, lifelong depth
A beginner can play a meaningful game on a small board after a short introduction. As ability grows, the same rules reveal deeper questions of timing, efficiency, sacrifice, influence and balance. Children can develop focus and confidence; adults can find intellectual challenge and calm concentration; players of different ages and languages can meet across one board.
Go connects ancient culture, modern science and everyday human learning. That is why it remains worth learning now.
Ready for your first game?
Begin with the basic rules, then learn by playing. Every game gives you a new problem to understand.