The Digester

AlphaGo's 2016 victory over Lee Sedol showed that neural networks and self-play can produce beyond-human intuition, sparking a decade of AI advances from game playing to biology while highlighting limits in explainability.

In March 2016 AlphaGo beat world champion Lee Sedol 4-1 in Seoul, and move 37 became an iconic example of unexpected AI strategy.
AlphaGo combined deep neural networks trained on human games with millions of self-play games to discover new winning strategies.
The match proved neural networks can learn pattern recognition and strategic intuition that surpasses top human players.
Researchers extended those techniques into general game engines like AlphaZero and scientific tools such as AlphaFold, which transformed protein structure prediction.
Modern large language models follow a similar pattern of broad pretraining followed by refinement through reinforcement learning or human feedback.
A major limitation exposed by AlphaGo is that these models are often black boxes that can produce unexplained or incorrect outputs.
AI progress is fastest in areas with abundant data and a clear, verifiable success signal, such as games, programming, math and protein folding.

How AlphaGo's 2016 win changed AI

Articles