Libratus, computer poker champion

[Helios]

I've always liked news about computers that beat humans at games. I recall the milestones; checkers, backgammon, chess, jeopardy, go, and now Texas hold-em poker ( the modern and most popular variant of poker nowadays ). When a computer won at Go not too long ago, I told my friends who play poker, "Ya know, very soon a computer will master this game too." They replied, "No way, you see a human has the unique ability to do this or that. It's a hundred years away". Well...

"Libratus, an artificial intelligence developed by Carnegie Mellon University, made history by defeating four of the world’s best professional poker players in a marathon 20-day poker competition, called “Brains Vs. Artificial Intelligence: Upping the Ante” at Rivers Casino in Pittsburgh."

https://www.cmu.edu/news/stories/archives/2017/january/AI-beats-poker-pros.html

 

[Prideful]

There was an ai that beat the best in chess and most impressively, go. Poker by comparison is simply child's play. Card counting and reading the odds.

 

[mheslep]

...Poker by comparison is simply child's play. Card counting and reading the odds.

Odds calculation is part of the game, but the strategies involved in signaling to other players via bets as play progresses are complex.

Still, poker is not a fair human-machine comparison, as the machine really has nothing to lose. A machine can't bet the car keys or the mortage, nor grasp the meaning of such a bet by an opponent. That is, the nachine can't suffer existential consequences.

 

[russ_watters]

Odds calculation is part of the game, but the strategies involved in signaling to other players via bets as play progresses are complex.

Right. As the article points out, the key skill is being able to bluff (and sniff-out a bluff).

Still, poker is not a fair human-machine comparison, as the machine really has nothing to lose. A machine can't bet the car keys or the mortage, nor grasp the meaning of such a bet by an opponent. That is, the machine can't suffer existential consequences.

In tournament play, stakes aren't really at issue since everyone has a buy-in and it's money already spent before you start playing. The issue I see with this particular test was that it was a 20 day, 120,000 hand "marathon", which favors a machine's superior ability to track and learn. As the article says, the computer was able to adapt to the peoples' playing styles throughout the test. That is a lot harder if you are playing against 100 people in a week instead of 4 people for a month.

 

[mheslep]

In tournament play, stakes aren't really at issue since everyone has a buy-in and it's money already spent before you start playing.

Machines can play as if in a 'closed system' game, but I contend there is no such thing for people. We're finite, and we live interdependent with others. Our time, our resources, are finite, and our performance is observed by others. Playing in a tournament effects these, which factor into playing in the *next* tournament. We can't come up with endless tournament buy-in money with no winnings.