10 Years Since AI Beat Go — From Move 37 to a Nobel Prize and the Road to AGI
In the decade since AlphaGo defeated Lee Sedol in 2016, AI has solved protein structures, won a Nobel Prize, and earned a gold medal at the Math Olympiad. Here's a summary of the memoir written by the CEO of Google DeepMind.
In March 2016, the AI Go program AlphaGo defeated world champion Lee Sedol 9-dan 4–1. More than 200 million people around the world watched the match, which marked the dawn of the modern AI era. Exactly 10 years later, Google DeepMind CEO Demis Hassabis has published a personal memoir looking back on it all.
Here's the story of how a technology born on the Go board went on to transform drug discovery, mathematical proofs, and weather forecasting — a decade-long journey.
Move 37: A One-in-Ten-Thousand Play
In Game 2 of the March 2016 match, AlphaGo played a move that professional commentators immediately dismissed as a mistake. Move 37 — a bold stone placement with only a 1-in-10,000 chance of appearing in human game records. A hundred moves later, it became clear that the stone was strategically decisive.
Lee Sedol 9-dan said: "The greatest lesson AlphaGo taught us is that the age of AI is not some distant future — it's already here."
Lee Sedol's counter, Move 78, was itself a 1-in-10,000 move. Dubbed the "divine move" by the professional Go community, it was a moment where human and AI pushed each other to new heights.
From the Go Board to the Lab — AlphaFold and the Nobel Prize in Chemistry
AlphaGo's core technologies — deep neural networks combined with reinforcement learning (a method where AI teaches itself by playing millions of games and learning how to win) — didn't stop at Go.
Key Milestones Over the Past Decade
2017 — AlphaGo Zero — With no human game data at all, it learned Go entirely from scratch through self-play and became the strongest Go player in history.
2018 — AlphaZero — Mastered not just Go but also chess and shogi (Japanese chess). It defeated Stockfish, the world's top dedicated chess engine, after just a few hours of training.
2020 — AlphaFold 2 — Solved the 50-year-old protein folding problem (predicting what 3D shape a protein will form based on its amino acid sequence). It predicted the structures of more than 200 million known proteins and made the entire database freely available to over 3 million researchers worldwide.
2024 — Nobel Prize in Chemistry — Demis Hassabis and John Jumper were awarded the Nobel Prize in Chemistry for AlphaFold's achievements. It was the first time a Nobel Prize was given for AI-driven scientific breakthroughs.
Winning Gold at the Math Olympiad
AlphaGo's technology also proved its power in mathematics.
AlphaProof and AlphaGeometry 2 achieved silver-medal-level performance at the International Mathematical Olympiad (IMO), combining AlphaGo's reinforcement learning and search algorithms to formally prove mathematical theorems.
But it didn't stop there. Gemini's Deep Think mode reached gold-medal level at the 2025 IMO. Technology that started on the Go board had reached the level of the world's best mathematicians.
AI Scientists Have Entered the Lab
DeepMind's technology is now spreading across multiple scientific fields.
AlphaEvolve — An AI coding agent that discovers more efficient algorithms on its own. It found new methods for matrix multiplication (the core mathematical operation powering modern AI). It's also being tested for data center optimization and quantum computing.
AI Co-Scientist — A system where multiple AI agents "debate" scientific ideas with one another. In a validation study at Imperial College London, the AI independently arrived at the same antimicrobial resistance hypothesis that a research team had spent years developing.
AlphaGenome — From understanding the genome to accelerating nuclear fusion energy research and improving weather forecasting accuracy, the applications of AlphaGo's underlying technology continue to expand.
Hassabis's Roadmap to AGI
In his memoir, Hassabis lays out a concrete path toward AGI (Artificial General Intelligence) — AI that isn't just exceptional in one narrow domain but can tackle fundamental challenges like curing diseases or achieving limitless clean energy. He says three things are needed:
1. The ability to understand the world — Multimodal AI that processes text, speech, video, images, and code all at once (Gemini fills this role)
2. AlphaGo's search and planning capabilities — The ability to explore many possibilities and formulate optimal strategies, just like reading moves ahead in Go
3. Specialized AI tools — An architecture where a general-purpose AI can call on specialized AIs like AlphaFold as "tools" whenever needed
Hassabis concludes: "The breakthroughs sparked by that first creative flash we saw in Move 37 are now converging. The path to AGI is opening up — and a new golden age of scientific discovery is about to begin."
Why This Matters Even If You Don't Know Go
The reason this story matters is simple: technology that started on a Go board 10 years ago is now developing new drugs, predicting weather, and solving math problems.
The AlphaFold database is freely accessible to anyone. It's already producing real-world results — from malaria vaccine development to research on plastic-degrading enzymes.
The number of possible positions in Go is 10 to the power of 170 — more than the number of atoms in the observable universe. The technology that conquered that complexity is now tackling real-world problems just as vast.
Related Content — Get Started with AI | Free Learning Guide | More AI News
Sources
Stay updated on AI news
Simple explanations of the latest AI developments