Friday, 17 February 2017

Another human activity taken down

While I was overseas I kept up with events outside the chess world, including the Human v AI Poker Match. The Libratus Poker bot, developed at Carnegie-Mellon played 4 poker pros in heads up no limit holdem, and beat all 4 of them. As with most AI v Human matches, this was a rematch after an earlier version of the program lost to human players a few years back.
Reading about the match, and the approach the bot used, there was one significant factor which I recognised from earlier Chess/Checkers/Go matches. While there was plenty of smarts built into each of these engines, it was the lack of emotional attachment to decisions that seems to be a big difference. Humans may decide that a certain move or choice is the 'smart play', but then decide not to play it due to other factors (tiredness, intuition, emotion). Bot's on the other hand will make a decision and then execute it. In chess this is often manifested by a change of plan depending on the previous move, which humans are slower to try. According to the authors of Libratus, in certain situations Libratus was 'fearless', which gave it an advantage.
While the win for the bot is significant I'll withhold judgement on whether it will change poker in the same way that engines changed chess. There may be a reassessment of how certain hands should be played, but overall I think the 'game space' is small enough that humans already have most of it covered.

No comments: