The ancient board game of Go is a marvel of abstract strategy, a two-player contest that stretches back over 2,500 years, originating in China and becoming the oldest game still played in its original form. A 2016 survey found that there are captivating a global people of over 46 million enthusiasts, and over 20 million current players.
But Go is not just a game of history and numbers; it’s a cosmos of complexity. The simplicity of its rules – placing black or white stones on a grid to capture territory – belies an astronomical depth of possible moves and counter-moves. In fact, Go’s complexity is often framed in celestial terms: the number of legal board positions is estimated at a staggering 2.1×10^170, dwarfing the number of atoms in the observable universe, which is a mere 10^80 by comparison. It’s a comparison that humbles even the grandeur of the cosmos and underscores the game’s intricate nature.
This vast complexity is not purely an intellectual curiosity – it poses a genuine challenge for artificial intelligence. Go, with its open-ended gameplay and emphasis on strategy and intuition, was long considered beyond the capabilities of computers, which had already mastered other strategy games like chess. Where chess programs calculate several moves in advance, Go’s branching factor of possible moves – often around 150-250 per turn – made such calculations mind-bogglingly vast.
Then came AlphaGo, an AI developed by DeepMind that stunned the world by defeating European Go champion Fan Hui in 2015, and a year later, 18-time world champion Lee Sedol. AlphaGo Zero, its successor, refined this capability further, learning from self-play and beat the previous version.