For the first time ever, a computer has beaten a professional Go player, Fan Hui. This is a major breakthrough when it comes to artificial intelligence and neural networks, as beating top Go players has been the last symbolical challenge.
Google DeepMind’s AlphaGo is much more sophisticated than IBMs Deep Blue, the chess-playing computer developed by IBM. Arguably, we’re now living in a golden age of artificial intelligence.
Go is a board game for two players. Despite its relatively simple rules and the fact that Go has been around for 2,500 years, top players are still looking for the best strategies.
In particular, every time you play, there are hundreds of possible moves. Players are talking about finding the most elegant shape to build territories. In many ways, compared to chess, Go is an open-ended game with multiple ways of playing, reflecting your personality.
I first learned to play Go 15 years ago. At the time, I could beat the best bots — and believe me, I wasn’t a good player. Go bots have come a long way. That’s why today’s news is surprising and fascinating at the same time.
Many companies including Google and Facebook have been building teams of artificial intelligence researchers to work on the future. There are obvious applications, such as facial recognition, self-driving cars, Google Now, M for Facebook Messenger and more. But these teams are also working on more trivial issues, such as Go bots, in order to learn new techniques.
And it looks like DeepMind is one step ahead of the competition with AlphaGo. It combines the Monte-Carlo technique to evaluate the most efficient moves in the tree of potential moves with neural networks. DeepMind also uses external memory resources to boost the efficiency of neural networks.
In more pragmatical terms, AlphaGo is a bot that learns over time. The more games it plays, the better it gets. AlphaGo is even playing games against itself to get better.
A few days ago, DeepMind played against European Go champion Fan Hui. AlphaGo won by 5 games to 0. In March, AlphaGo will play against the best player in the world, Lee Sedol. Now, it’s not a matter of knowing “if” AlphaGo can beat Lee Sedol, but “when.”
As I wrote on Twitter, the difference between our good old monkey-derived brains and computer programs is getting smaller. At some point, “artificial intelligence” is going to become “intelligence,” like “new technologies” became “technology.”