Poker Magazine

Robots: Round 2

Back in 2005 there was a competition to see who could program the strongest poker playing software. The rules were simple. Show up, have your bot compete against all the other bots, and the last bot standing (actually the programmer) would walk with $100,000.

As an added bonus, the guy who programmed the winning bot was invited to have a crack at beating “the humans” as well. I represented and fended off the reigning champion and the runner-up. The format was headsup Limit poker with escalating blinds until one or the other had all the chips.

That year (2005) eight teams competed. Most were individuals who had already built some sort of bot and fine-tuned what they had for a crack at the 100k. Hilton Givens won the competition, which is amazing considering that he was a solo competitor.

However in 2005 one team really shined. This team was committed to “solving” poker and nothing was going to stop them. Nothing. They are the guys who invented Poker Academy. If you don’t know already, Poker Academy (www.poker-academy.com) is software written by some of the geniuses at the University of Alberta. It is designed to help you with your poker game and it is good.

About 16 years ago, the University of Alberta decided it was going to solve poker. Solve it? Who knows? But they were going to give it the good ol’ college try (pun intended). They started as a team of one and over time their team has grown to twelve.

That’s right, twelve really smart guys, all programming / math wizards, and at this very moment still trying to solve poker. For the time being they are focused on Limit heads-up play. Very specific poker domain. Very complex problem. And very comforting to know that it is still a bit elusive to them — never mind me!

(Small footnote here … Some of the guys on the poker team were part of the team that recently solved checkers. That’s right, kids, it happened about two months ago. Checkers, solved. Chalk one up for the University of Alberta’s math and science guys!)

Anyway, let’s fast forward to now. Or more accurately, July 24th 2007. Two years later and Robot Wars 2 is upon us. New and improved software. New venue. And a new format. The biggest change format-wise is that it will be duplicate poker and the luck factor is going to be greatly reduced.

The Poker Academy guys invited me and Ali “Prince Ali” Eslami to Vancouver for the 22nd Conference on Artificial Intelligence (AAAI-07). We were to compete against their new and improved poker playing bots. And instead of a tournament format where luck admittedly plays a part as the blinds escalate, this time it would be duplicate poker, more hands, and in a money game format.

The duplicate format meant that Ali and I would be dealt the same series of cards in two parallel Man-Machine matches, with each of us playing the opposite hands in each game. At the end of the match, the total number of chips won or lost by us would be added to determine the winning team. The idea was to reduce the luck element of poker whilst collecting the data.

The weekend was illuminating and intense. Ali Eslami is a Limit player who can often be found playing the biggest limit games that the Commerce Casino in LA has to offer. He was the perfect partner for this adventure. He crushes the Limit games and had read all the papers that the University guys had published with regards to their poker-playing bots.

The blow-by-blow results of the competition were posted by the Poker Academy guys (http://www.cs.ualberta.ca/~games/poker/ man-machine/Results/). The short and sweet summary would be that we drew one, lost one, and won two. QED Humans win. But that does not give you the flavor of what happened.

What happened was that Ali and I were stunned by how sharp this thing is. In fact, after our first loss (in the second match), I felt as if I had been sucker punched. We got destroyed and it took a bit to get our footing again. It is a bit disconcerting when your opponent makes all its decisions in 1/10th of a second. The computer would be betting before the graphics had the chance to show the card!!!

It was not just one bot — it was many. They sent them in one after another. We played four matches in all, each one 500 hands each. For the third match they really went to town. They had our previous play analyzed. Yes, they had a program specifically designed to determine which bots would do better against our respective styles of play! They ended up sending a blend of different bots for each of us to deal with. It was meant to happen with the help of a coach program. These guys really spared no punches.

We may have been lucky in Round 3. It turns out the coach program designed to send in the “optimal” bot for the next hand froze. We ended up finishing Round 3 against whatever bot it was on when the coach program seized. They still had a strong bot against us, just not what the coach program may have determined to be optimal.

They are planning on a rematch. We held them off this time, but I have a prediction. Their next assault will be more finely tuned, their bots will be stronger, and they will not misstep. For me, the days of beating the bots might soon be over. It isn’t defeatism. It is realism. These guys are the real deal, and they are closing in.

So is that it? Do we just settle for the draw when that happens? Nope. How’s that? Well, when (yes, when) the day arrives when I can only get a draw … well, when that day arrives, I plan on getting them tougher competition. That’s right, the next round. And you know who you are. We all do. You are the heads-up Limit wizards who have already cranked out millions. Please be ready and willing as the torch may be passed to you sooner than you think if we are to keep the bots defeated.