Ten Years After Move 37, DeepMind Says AlphaGo Set the AGI Agenda

Ten years ago, AlphaGo became the first program to defeat a world champion at Go — a milestone DeepMind says arrived a decade before many experts thought possible.

In 2016, more than 200 million people watched the system face Lee Sae Dol in Seoul. The match turned on "Move 37" in Game 2, a play so unconventional that professional commentators initially dismissed it as a mistake. Roughly one hundred moves later, the stone sat in exactly the right position for AlphaGo to win. DeepMind describes the moment as proof that AI could go beyond mimicking human experts and find entirely new strategies.

Go has long served as a proving ground for AI research because of its scale: there are 10^170 possible board positions, far more than atoms in the observable universe. AlphaGo made the game tractable by combining deep neural networks with advanced search and reinforcement learning — an approach DeepMind pioneered. It first learned from games played by human experts, then played hundreds of thousands of games against itself, reinforcing the strongest winning strategies.

The successors came fast. AlphaGo Zero learned entirely from random play and became, in DeepMind's words, arguably the strongest player in history. AlphaZero generalized the system to any two-player perfect information game, mastering chess from nothing but the rules in a matter of hours and beating the best specialized programs of the time, including Stockfish. Even in a game as heavily analyzed as chess, AlphaZero produced new strategies.

From the board to the lab

DeepMind frames the Seoul victory as the moment its technology became ready for its real goal: accelerating scientific breakthroughs. The first target was protein folding, a 50-year grand challenge. In 2020, AlphaFold 2 cracked it. The company then folded all 200 million proteins known to science and released them in an open database. Today, over 3 million researchers use that database, working on everything from malaria vaccines to plastic-eating enzymes. In 2024, Demis Hassabis and John Jumper received the Nobel Prize in Chemistry for leading the project.

The AlphaGo lineage now runs through several systems:

• Mathematical reasoning. AlphaProof, the most direct architectural descendant, combines language models with AlphaZero's reinforcement learning and search. Together with AlphaGeometry 2, it became the first system to earn a silver medal at the International Mathematical Olympiad. An advanced version of Gemini's Deep Think mode went further, achieving gold-medal performance at the 2025 IMO using an approach inspired by AlphaGo.
• Algorithm discovery. The coding agent AlphaEvolve explores the space of computer code the way AlphaGo searched for moves. It found a novel way to multiply matrices — a fundamental operation powering nearly all modern neural networks — and is now being tested on problems from data center optimization to quantum computing.
• Scientific collaboration. An AI co-scientist built on AlphaGo's search and reasoning principles has agents "debate" hypotheses. In validation studies at Imperial College London, it analyzed decades of literature and independently arrived at the same hypothesis on antimicrobial resistance that researchers had spent years developing experimentally.

Why it matters

DeepMind argues these scientific systems, however capable, remain highly specialized. Fundamental breakthroughs — limitless clean energy, understanding diseases that defy us today — will require general systems that find structure across subject areas and generate new hypotheses the way the best scientists do.

The company's stated recipe for AGI combines Gemini's multimodal world models, AlphaGo's search and planning techniques, and the ability to call specialized tools, such as AlphaFold for protein structures. The latest Gemini models already use techniques pioneered with AlphaGo and AlphaZero.

DeepMind sets a high bar for what counts as true creativity. Move 37 showed AI could think outside the box, the company writes, but genuine invention would mean not just finding a novel Go strategy but "actually invent a game as deep and elegant, and as worthy of study as Go."

Ten years on, the company says the breakthroughs catalyzed by Move 37 are converging toward AGI — and what it calls a new golden age of scientific discovery.