Tartanian 8 Creator Thinks Bots Will Beat Pros Within Three Years

Last month there was a lot of reporting on the results from this year's Annual Computer Poker Competition. With 11 bots in the field the ultimate winner was Baby Tartanian 8, and there has been quite a stir about the possibility that we are in sight of crossing the point where computers can beat the very best human players in a game of heads up No Limit Holdem.

Baby Tantanian 8 was created by Noam Brown,a Ph.D student in the Computer Science Department of Carnegie Mellon University, with help from his thesis advisor, Professor Tuomas Sandholm. They received their award from the Association For the Advancement of Artificial Intelligence for first prize in the 'Total Bankroll' category, and for third place in the more minor 'Instant Run Off' Category.

While there was a Tartanian 7, version 8 was a completely built from scratch. Brown and Sandholm gained access to the Comet Supercomputer at the San Diego Supercomputer Centre and decided to start all over again to optimize the speed at which the bot was able to process information. Ultimately, due to tournament restrictions, the resulting program had to be trimmed back somewhat and the end product was Baby Tartanian 8. Still a mean beast by all accounts.

Brown talked about what he called the process of “pruning” which was necessary to cut back on the number of possible choices the bot had available while still maintaining an efficient algorithm capable of arriving at the correct conclusion. In layman's terms, this meant cutting down the possible hand outcomes the bot "thinks" about so that it could play on time.

This all reminds me of ten years ago when chess players were going through the a similar situation where the world's best were no longer being able to compete against the might of machines. Slowly but surely, the programmers were able to trim back their code to a level where it was so efficient that even on a modest computer was enough to beat any human player. And today, the gap has widened to the point where it is no longer even worth attempting as a sporting contest.

The analogy of chess to NL Holdem is relevant because of the fact that there are more possible chess games than there are atoms in the entire universe (10^120 is a common estimate). This shows that brute force methods and plain memory are not the be all and end all of a solution to both games. It is extremely important to be able to cut down the number of options a program has.

As programmers use increasingly sophisticated methods, they should be able to find the same success with poker bots as chess programmers did in the early years of the new millennium.

The work involved is not just useful for making poker programs though. Brown explained:

This indicates that there is a lot more work going on for different everyday projects that can fuel progress with Tartanian and updated versions. There's is no need for a completely dedicated poker team working around the clock. Progress is likely to continue regardless.

How Much Progress Has Been Made?

Last year Tartanian 7, christened Claudico, went head to head with a four man team of top draw poker professionals. Doug Polk, Dong Kim, Jason Les, and Bjorn Li played 80,000 hands against the machine and came out as the winners. But only just!

Claudico design team member Sam Granzfried reviewed the results and declared that in his opinion, with what they had learned, within a year they could produce an updated version which would beat all humans.

Noam Brown thought this maybe a little optimistic and predicted that success would come in two to three years.

Even so, it would appear that the gauntlet has been laid down, and they seem confident in where they are heading.

How Bad is This For Poker?

Time for me to put my tin foil hat on!

All regular online players are aware of the existence of poker bots. Even if they're not capable of beating the best players, they can certainly play at a very high level. How high is difficult to be certain of, but better than average surely.

What is good right now, though, is that quality bots are hard to come by. This keeps the amount of infiltration down in games.

If, as predicted by Brown et al, there is an all conquering poker bot out there, it's only a matter of time before somebody releases one into the mainstream market. This could be seriously destructive to the integrity of the games online. Some online poker providers appear to do their utmost to prevent bot use, but with a mainstream program on the loose we could find ourselves swamped, and that's a worrying prospect.

We must hope that the IT boffins working for the main players in the industry are able to keep ahead of the curve, and prevent the game from being tainted any more than the other bot scandals have so far managed.

Is it possible that we could reach a point whereby somebody is able to play a single table whilst entering each action onto a second computer while receiving perfect advice for each hand? How do we police against this?

It's food for thought, and we're not there yet, but it would be a tragedy if we all walked into this in two to three years time. Confidence could become so low in the game's integrity that it starts to die off.

Maybe the answer is to change the game's rules slightly, such as we see in the new Six Plus Holdem. This would force the programmers to start again. We could possibly see a perpetual cycle of playing new games until they can be beaten by machines, and then moving on. It may sound like science fiction, but no one knows what the future will bring.