Tuesday, 22 February 2011
-
Could IBM's "Watson" Become a Poker Machine?
This is a guest post from Just A Guy Thing.
With past demonstrations, IBM has shown that they can create machines that can beat humans at chess, and most recently, trivia. Which is fun to watch, but doesn't serve too much of a purpose for people that aren't interested in recreational chess or trivia. Of course, creating a machine that excels at chess or trivia isn't IBM's end game. Their goal is to create a machine that can mine through data to provide useful answers to most any sort of problem. So these applications in the media are simply a means to an end.As long as they're tooling around with different applications, let's find one for people who love money, which is just about everyone. Can IBM create a machine that can clean house among the best poker players in the world? If Watson can move through terabytes of data to answer cleverly-worded trivia questions, then surely it can analyze cards and betting patterns to render the human mind obsolete when it comes to sitting at a card table, no?
Not really. As this Slate article conveys, machines can kick ass at poker when there are only two participants, but what are the odds that two people will be playing head-to-head for an extended time? Not great. The technologoy isn't their to measure the behavior of the participants. However, that doesn't mean it can't be done, and it doesn't mean that great minds aren't working on that exact problem. It turns out that tendencies in a game of poker mirror closely those in financial markets and auctions. So the poker problem could be applied to a lot of high-dollar applications.
So the machines will probably end up taking us in poker as well. What does that mean for you? Not much, unless you're going to allow players to sit down with a laptop and let the computer do the work for them? And if you are, then you're probably the biggest sucker at the table.
Post a Comment
- Back to mancouch's Mancouch Site!
- Note: your comment will appear in mancouch's local time zone: GMT -05:00 (Eastern Standard - US, Canada)


Recommend


Comments (5)
I think getting a computer to consistently defeat great no limit holdem players (in non-heads-up situations, which will be most interesting) is going to be much more of a bitch than most people anticipate. Against the best Jeopardy players in the world, Watson won probably primarily because of its speed. That won't be a factor in poker.
Will metagame make much of a difference? Top players try randomizing their moves at times or do seemingly wild things to mess with their opponents. When a player makes a play that, in a vacuum, doesn't make any sense, how will a computer react to that new piece of incongruous information? Against the greatest players, who balance their ranges very well (equalizing the frequency of bluffs, monsters, and average hands as is fit to the situation), a computer's edge could be so minimal that it's debatable who is really better even after hundreds of thousands of hands played. Basically, it's incredibly tough to beat a player with next to zero leaks (exploitable tendencies), human or machine!
And what about tilt or switching gears? When a person suddenly starts playing differently, maybe because of their emotions, how is a machine going to read them? How will it know what level thinkers the other players are, so it can be one above them? (Level 1 is I know my hand, level 2 is I know yours, level 3 is I know you know mine, level 4 I know you know I know, and so on to increasing levels of insanity lol) The machine will have to adjust to the players adjustments, so it doesn't become predictable to them...
This is so freakin' complex, I'm wasting lots of time just pondering it! So many questions, and I want answers, damn it! (Well, not really, because I play poker for most of my income, teehee.) Good luck, game theory nerds: you've got a lot of work ahead of you.
Incredible !
IBM is really the winner
I like Watson. He's one funny bot.
Dear friend..share with you a very good website here
http://www.legendstrade.com/