Last night, AlphaGo beat Lee Sedol again to finish the 5 game challenge match ahead 4 games to 1. It was an exceptionally tight game. One commentator suggested Lee Sedol was behind by 2.5 points at the end when he resigned. There was no potential left to steal the lead. Here is a very good look at the final game by Jonathan Hop, a strong amateur who has written some excellent books on Go.
AlphaGo has demonstrated that while it is not infallible, it is close to it. Predictions that we were still a decade away from a computer that could beat a top professional go player were way off track.
Lee Sedol played admirably. He played on the edge and came up with some remarkable moves, and I think he was an excellent test for the computer.
Thee must be all kinds of implications to this technology. Machine learning and nuro-networks that narrow down possibilities then assess the probability of success for any given position are nothing short of amazing. What will it mean for our lives in 10 years, or 20 years? Who knows.
My brother suggested to me that it could be a disaster if some Dr. Strangelove were able to unleash this technology in a diabolical and nasty manner and I suppose that is always a possibility.
In the shorter term, this challenge match has rocked the Go world. Go relies on some combination of assessment, intuition, calculation and creativity. Will AlphaGo introduce new ideas to the game us humans haven’t considered, or debunk ideas our top players have held dear?