From games to biology and beyond: 10 years of AlphaGo’s impact
Google DeepMind
DeepMind is marking ten years since AlphaGo beat Lee Sedol at Go with 'Move 37.' Demis Hassabis says that win kicked off everything from AlphaFold's protein maps to today's AGI push.
Based on reporting by Google DeepMind — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
A decade on, Demis Hassabis is still talking about a single stone placed on a Go board in Seoul. Move 37, the play that made professional commentators think AlphaGo had glitched, turned out to be the moment DeepMind's leadership realized their reinforcement-learning approach could do more than win board games. It could tackle problems nobody had cracked in fifty years.
That's not hyperbole dressed up for an anniversary post. AlphaGo's descendants have genuinely reshaped how science gets done. AlphaFold 2 solved protein structure prediction in 2020, and DeepMind then published folded structures for all 200 million known proteins in an open database that over 3 million researchers now use — work that earned Hassabis and John Jumper the 2024 Nobel Prize in Chemistry. The lineage from a Go-playing program to a Nobel-winning biology tool is the kind of thing that sounds made up until you check the dates.
The pattern keeps repeating. AlphaProof and AlphaGeometry 2 combined AlphaZero's search techniques with language models to hit silver-medal performance at the International Mathematical Olympiad, and Gemini's Deep Think mode later reached gold-medal level using ideas traced directly back to AlphaGo. AlphaEvolve, DeepMind's code-discovery agent, found a new way to multiply matrices — arguably its own Move 37 — and is now being pointed at data center efficiency and quantum computing problems. Even an AI co-scientist tool, tested at Imperial College London, independently reproduced years of human research on antimicrobial resistance just by debating hypotheses against itself.
What Hassabis is really arguing, underneath the victory-lap framing, is that none of this was ever really about Go. The 10^170 possible board positions were just a big enough haystack to prove that neural networks plus search plus self-play could find needles humans couldn't. Now DeepMind wants to fuse that search-and-planning machinery with Gemini's multimodal world model and specialized tools like AlphaFold, betting that's the combination that gets to AGI. Whether that bet pays off is the open question the next decade actually has to answer, not the one being celebrated this week.
My take — AI-written commentary, not fact-checked reporting
I'll give DeepMind this: AlphaFold's open database is one of the few AI achievements the whole scientific community actually uses without complaint, and it's a genuinely good advertisement for the difference between hype and delivered utility. But stringing a straight line from a Go move to AGI is the kind of retrospective mythmaking every big lab does once it's raised enough money to need a founding legend — the real test isn't Move 37, it's whether the next Gemini can do something as unglamorous and checkable as AlphaFold did, instead of just acing another benchmark.
Read more about this at: Google DeepMind