Google’s Deepmind group has announced that AlphaGo will retire undefeated after sweeping the competition at the Future of Go summit in China. The AI crushed the world number one Ke Jie 3 – 0 over the weekend, while also finding the time to defeat a team of five champion players.
Deepmind is satisfied that it has managed to meet a challenge that was once thought to be impossible. From here, it is looking to apply the knowledge gained from working with AlphaGo to build a more general algorithm that can be applied to assist in research.
Of course, the work with Go is not yet done. Deepmind is set to release one last paper on AlphaGo’s algorithm. This will add to an earlier paper and detail how the team improved efficiency. It will also expand on how this very algorithm can be adapted to meet a broader set of problems.
There is also one last gift to the Go community from Deepmind. The group will release 50 replays of AlphaGo’s learning games; where it played against itself. These will be full length videos with time controls, allowing enthusiasts to fully examine the AI thought process that goes into deciding on what plays to make.
Computer scientists have been working to create a computer like AlphaGo since Deep Blue beat Gary Kasparov at chess. The idea was that there were too many possible moves in Go for a machine to be able to calculate all the possible moves in a reasonable amount of time. Which is what makes the success of AlphaGo over the weekend extremely special for technology.
The fact that AlphaGo was not programmed with specific moves in mind, but rather learned them on its own only makes this a massive milestone for AI research. The next challenge is getting computers to think in real time, instead of being allowed to sit and deliberate over a turn based game.
Deepmind is working with Blizzard to teach a version of AlphaGo how to play Starcraft; which is probably where the next AI milestone will come. But for now, the Go playing AI has earned its retirement and has ushered in a new age of self teaching computers.
[Source: Deepmind]
