Joerg Hiller
Mar 10, 2026 15:41
Demis Hassabis displays on AlphaGo’s decade-long influence, from Nobel Prize-winning protein folding to AGI improvement. This is what it means for AI’s future.
Google DeepMind CEO Demis Hassabis has printed a retrospective marking ten years since AlphaGo defeated world Go champion Lee Sedol, outlining how that 2016 victory spawned breakthroughs from Nobel Prize-winning protein prediction to what he calls a transparent path towards synthetic normal intelligence.
The March 2016 match in Seoul—the place AlphaGo received 4-1 towards the 18-time world champion—wasn’t presupposed to occur for an additional decade. Go’s complexity, with extra potential board positions than atoms within the universe, had lengthy been thought of AI’s Mount Everest. Transfer 37 in recreation two, a play so unconventional that commentators initially thought it was a mistake, grew to become the second many researchers level to as AI’s inventive awakening.
From Board Video games to Biology
The search and reinforcement studying strategies that powered AlphaGo’s victory have since been repurposed for issues that really matter. AlphaFold 2, which cracked the 50-year protein folding problem in 2020, used related architectural rules. The system has now predicted constructions for all 200 million recognized proteins, with over 3 million researchers accessing the free database for work starting from malaria vaccines to plastic-degrading enzymes.
That work earned Hassabis and colleague John Jumper the 2024 Nobel Prize in Chemistry—a uncommon occasion of AI analysis receiving science’s highest honor.
Mathematical Reasoning Hits New Heights
AlphaProof, described as AlphaGo’s “most direct descendant,” combines language fashions with the unique system’s reinforcement studying to show formal mathematical statements. Alongside AlphaGeometry 2, it achieved silver-medal efficiency on the Worldwide Mathematical Olympiad—the primary AI system to achieve that benchmark.
Gemini’s Deep Suppose mode pushed additional, hitting gold-medal normal on the 2025 IMO. That very same method now tackles open-ended scientific and engineering challenges.
AlphaEvolve, DeepMind’s coding agent, had what Hassabis calls its personal “Transfer 37 second” when it found a novel matrix multiplication technique—a elementary operation underlying nearly all trendy neural networks. The system is at present being examined on knowledge heart optimization and quantum computing issues.
The AGI Roadmap
Hassabis’s put up makes DeepMind’s AGI technique specific: mix Gemini’s multimodal world fashions, AlphaGo’s search and planning strategies, and specialised AI instruments like AlphaFold right into a unified system. The purpose is not simply an AI that may devise a profitable Go technique, however one able to inventing “a recreation as deep and chic, and as worthy of research as Go.”
An AI co-scientist system, at present in validation research at Imperial School London, already demonstrates this potential. By having AI brokers “debate” hypotheses, the system independently reproduced antimicrobial resistance findings that took human researchers years to develop.
Ten years in the past, AlphaGo proved machines may grasp a recreation people had performed for two,500 years. The strategies it pioneered are actually being utilized to fusion power, climate prediction, and genomics. Whether or not that path results in AGI stays unsure, however DeepMind is betting that Transfer 37 was simply the opening.
Picture supply: Shutterstock
