A Google computer has handily beaten a leading champion at the board game Go. It’s arguably a far more impressive AI achievement than computers prevailing at chess.
The game of Go appears on the surface to be simple, involving placing black or white pieces on a grid in an attempt to capture opposing pieces by surrounding them. The problem is that at any one moment, the number of legal moves available will usually far outnumber those in any given chess position.
This means assessing numerous options (and in turn the most likely exchange) of moves that will follow, something that is arguably more suited to the multitasking human brain than a computer processor that works in series rather than parallel. It’s also a particular challenge just to assess whether a player is in a strong position thanks to the ease with which pieces can go from being offensive weapons to being captured.
Google notes that the number of possible layouts of stones is more than the number of atoms in the universe and “more than a googol times larger than chess.”
It’s developed a system called Alpha Go, which tries to improve on standard artificial intelligence in a couple of ways. Firstly, it uses a neural network that simulates the way the brain can use connections between neurons to effectively create a “tree” of possible combinations of moves and outcomes and identify the best move to make next.
Secondly, it learns from games and develops strategies: in effect, shortcuts and “rules of thumb” that allow it to better target the range of possible moves to consider.
Eventually Google put the system to the test against some rival AI systems, which proved successful when it won 499 out of 500 games. Google then challenged European champion human player Fan Hui, with Alpha Go sweeping a five game series. According to Google, that’s the first time a computer has beaten a professional-level human player.