The Future of Go Summit starts today, where “legendary players and DeepMind's AlphaGo explore the mysteries of Go together”, or so says the website. The highlight of this event, organised by Google, is a showdown between Ke Jie, the top ranked Go Player in the world, and AlphaGo, the AI that beat Lee Se-edol, another top Go player who was humbled by the AI.
It all started when during last year’s Go matches between Seedol and AlphaGo, Ke Jie said he thought he could beat the AI. Well, AlphaGo’s creators seem to have accepted the challenge. Or perhaps the AI directed its unwitting creators to accept the challenge. Either way, a match is happening, and it’ll be streamed live on this link.
The Future of Go Summit will also have a bunch of panels from Google’s AI and machine learning experts, but the matches are in focus here. Before the showdown between 19-year-old Jie and AlphaGo, the AI will appear in two exhibition matches. Both of these are designed to present challenges AlphaGo has never faced before, at least in the public domain.
In the first challenge, two human players (presumably “legendary Go players) will play each other, with a copy of the AlphaGo program assisting each of them. It’s being called Pair Go, and essentially mixes strategies from the human pros with AlphaGo’s proven unorthodox moves. The AI is in essence facing off against itself as well here, which could make for very interesting gameplay.
The second exhibition match will have five Chinese pros take on poor old AlphaGo. So, the AI will have to contend against strategies coming from five human minds instead of just one. All this culminates with AlphaGo’s three-match contest against Ki Jie, in a one-on-one showdown. Five people going up against it shouldn't be any different than taking on one person, from AlphaGo's point of view. However, it gives the humans a better shot, since there are five minds trying to decipher the AI's unorthodox moves.
What’s changed since last year?
On the fact of it, nothing much has changed for AlphaGo yet. That is, till you consider the fact that it has been training for a full year. AI algorithms improve themselves by repeating a task, which means that the seemingly unbeatable Go playing AI has had a full year to “improve itself” since beating Lee Se-dol.
In its matches last year, AlphaGo’s gameplay was repeatedly described as aggressive, and the AI was said to be making moves that human players usually wouldn’t make. The same would be expected this time, but perhaps with even more creativity, that even five human minds working side-by-side cannot comprehend.