Tuesday, November 29, 2005

I, Poker Bot

My friend Wade forwarded a link about poker bots (which chained to another article). I've been brewing a post about bots for a while since my background in AI should make my opinion worth something (maybe 1.5 BB on a 1c/2c table).

I've been hearing rumbling about bots for quite a while from a number of sources. Of course the media is notorious for exaggerating the frequency of use and danger of poker bots. There is merit for some of what they are saying because online poker sites aren't really going to be releasing a lot of stats on how many robots they believe are in use. But, media articles generally are not written by industry insiders.

For instance, in the articles above, they mention 'poker pros' being involved in the scheme -- most people probably think of Doyle B. or Daniel N. when people say poker pros, but I doubt anyone that good would bother working on something that could tarnish their reputation. Overall, though, I don't think the bots out there are good enough to really hurt anyone's bottom line if you are intelligent about table selection (I'm not going to deny that bots are currently playing). In fact, a table full of bots would just look like a bad game, so why get involved?

At a Wednesday game I host about 6 months ago, one of the guys (not a regular) said he and some of his friends were working on a poker bot. You know, remote desktop, the whole deal. I don't know where that went (I don't remember his name) but I've definitely heard from people trying to do it.

Based on my experience, I don't think that bots are really worth the time and effort involved. These are my reasons:

  1. Bots really only work at low limits. High limit players are much trickier and I would expect it would be difficult to get really non-rigid play from a bot. Loki and Poki are obviously the first real successes and, oddly enough, Daphne Koller, one of the first researchers in the field ten years ago, is just down the hall from me. If I remember things correctly, it takes a decent amount of hands (thousands) to pick up on patterns. This is a reality that cuts across AI, actually -- usually humans can pick up on patterns with an order of magnitude fewer samples than any machine learning algorithm. Actually, my dream is to change that, but that's another post.
  2. The cost vs reward is too low. Realistically, a bots BB/hr could probably be around 5 at low limits (equivalent to a good human). But that will drop off as the quality of opponents increase. On the other hand, to get a good bot going, it is going to take lots and lots of hours of customization (even if you base it off existing software) and very few people are qualified to pull it off. Furthermore, you might make better money just by playing poker instead of programming :)
  3. Bots are easily spotted. Fitting with #2, trust me, throw a bot into a game at 10/20 online and it is not going to do very well. Not only that, I expect that human players will pick up on a bot pretty quick -- probably more from patterns and timing than anything else. Not to say all that stuff can't be programmed in, but again, cost vs reward. Furthermore, there is a really easy test to check for a bot: ask it a question in chat such as "What's your favorite sports team?" or "What's your favorite hand?". Trust me -- no bot will pass that sort of test (which is actually a true Turing test). The only way around it is to have the bot operator signaled when relevant chat comes through, but that kind of defeats the purpose, doesn't it? And its dirt cheap to implement at online sites -- just have a support person dedicated to querying suspected bots.
  4. Risk vs reward is low. Obviously, getting caught will cost you money because your account will be frozen. Furthermore, expect a black book to exist made up of these people and more stringent screening in the future. Since you can't make big bucks anyway (I guess $100 an hour is reasonable, but difficult) tightening up in the online sites will really put pressure on these guys. The only good thing for the bot guys is they are pretty safe from governmental intervention since poker technically isn't legal anyway. That's one of the many reasons I would love to see a state legalize poker.
So, really, my point here is that, yes, bots are out there, and yes, more will be created and their AI will continue to improve, but don't expect them to fill up the $2/$4 tables at Party anytime soon. Furthermore, the online sites will want to protect their huge rake income and keep bots in check. Maybe I should get hired as a consultant to feret out bots -- I'm sure I could pull down a lot of money for that :)

Additionally, as players we can help prevent their infiltration of the online game just by getting active in chat and reporting any suspicions to support.

Another thing that will be an issue in the near future is the blend between bots and stat trackers like Poker Tracker. In fact, about 6 months ago I downloaded a program that gave me advice (raise, fold, etc.) as it was grabbing info from hand histories. I'm not sure if it is allowed anymore (and I can't remember the name of it) but this is definitely a gray area which will become more of an issue. As it is, sometimes when I'm two-tabling 1/2 running PT and GT+ I start to feel like a bot myself (and a bad one at that).

No comments: