Google AI to play live Go match against world champion

[Gustaran]

Interesting, it's like the chess match between Kasparov and Deep Blue:

http://www.bbc.com/news/35501537

[Sian]

didn't their AI just recently become the first Computer to beat a relatively low-tier professional Go player?

[AdrienIer]

Yeah I think it was the french champion. The most surprising part was that the computer narrowly won the unlimited time games but crushed the guy in quick games.

[ipecac]

[Gustaran]

In a German article there is some interesting background information on how the Google software is combining features like Monte-Carlo tree search and neural networks.

Even for those not interested in Go, I wonder how cool it would be to have an AI in computer strategy games that can challenge a human player without any bonuses whatsoever. Civ 4 "Man against Machine" anyone?

[Dantski]

In a German article there is some interesting background information on how the Google software is combining features like Monte-Carlo tree search and neural networks.

Even for those not interested in Go, I wonder how cool it would be to have an AI in computer strategy games that can challenge a human player without any bonuses whatsoever. Civ 4 "Man against Machine" anyone?

I suspect the human player would still get upset over the AI cheating.

[Gustaran]

I suspect the human player would still get upset over the AI cheating.

[Boro]

well, this is certainly something to look forward to.

I suspect the human player would still get upset over the AI cheating.

certainly, even if we just look at the faster processing speeds computers work with

[Bacchus]

The most surprising part was that the computer narrowly won the unlimited time games but crushed the guy in quick games.

That's actually not surprising. The AI they are using is based on machine learning, it actively studies past professional games, it doesn't just try to pick a "best" move out of a tree based on some assessment of a deeply read resulting position. Obviously a computer can find a most commonly and successfully played move out of a position in the database much quicker than the human, who can only catch up in this sense under a correspondence format.

Because tree search in itself does so little in Go, the marginal gain for extra unit of time for the AI becomes miniscule quite quickly.

[Nystul]

The most surprising part was that the computer narrowly won the unlimited time games but crushed the guy in quick games.

That's actually not surprising. The AI they are using is based on machine learning, it actively studies past professional games, it doesn't just try to pick a "best" move out of a tree based on some assessment of a deeply read resulting position. Obviously a computer can find a most commonly and successfully played move out of a position in the database much quicker than the human, who can only catch up in this sense under a correspondence format.

Because tree search in itself does so little in Go, the marginal gain for extra unit of time for the AI becomes miniscule quite quickly.

From that basis I would suspect that the human merely played more solidly with extra time, but was equally hopeless. If the AI has a superior initial read of the position compared to the human ("instinctive" candidate move), I'm hard pressed to see how the human will catch up with extra time. Certainly not through calculation.