May the Best Robot Win

01:07
07 Mar

With this past week’s poker news seeing bots move to the forefront again, with a Swedish Court of Appeals surprisingly finding their use to win money from unsuspecting players as not being fraud, it is no surprise that here at Pokertubewe have been investigating all things bots. What might come as a surprise to many is that there is actually a competition where poker robots and their programmers slug it out to determine the best artificial poker player.

In fact, there are many versions of such competitions out there, from the Annual Computer Poker Competition, which see tens of millions of hands being played, to the MIT Pokerbots event which sees students given one month to design and program a bot capable of beating players chosen by the institute.

---

On top of this, last year saw a match-up which many saw as critical if humans were to still compete on level terms with machines. Four pros took on the Carnegie Mellon University Sandholm-designed bot Claudico.

Sponsored by Rivers casino, CMU and Microsoft, a remarkably wide margin of victory for the humans ensued in this heads-up poker contest. The designers later claimed that it was ‘statistically insignificant’ when the month-long series of games saw the human players come out $732,713 ahead.

So, still some hope left for the flesh-and-blood variety of poker player you might think, but as other mind-games have realised, such victories ought to be savoured as they won’t last for long!

---

Checkers solved, what’s next..?

Checkers, or draughts as it is known in some parts of the world, was one of the first to see humans superseded by their artificial rivals, when the program Chinook defeated the world champion back in 1994.

Well, technically a tie. Dr. Marion Tinsley had to retire on the grounds of ill-health after six draws. For those who don't know the good doctor, he is a checkers expert so far-beyond the human challengers of his day that he lost only seven games out of tens of thousands played, and came outright first in every event he ever played in against humans.

Not content with this Pyrrhic victory, Chinoook’s main developer Dr Jonathan Schaeffer decided he would solve the game entirely – and he managed exactly that. As reported on Chessbase.com:

---

Chess under pressure…

What Schaeffer and his colleagues had done to checkers, others have attempted to do in ches. Although the vastness of chess means it is unlikely to be mathematically solved in our lifetime, software is now available, even for your average phone, which would be capable of defeating the World Chess Champion . These chess-playing engines are so strong nowadays that the only human vs computer match-ups involve the machines giving their carbon-based opponents ‘odds’ of several pawns to make the games remotely interesting as a contest.

And so, back to poker, which with strategy and mathematical calculations not too dissimilar you might think would make it easy for machines to replicate and then improve upon our own faulty play? Not quite so easily, as it turns out…

---

Poker is a different beast...

Cards are not always visible, and it is this factor which makes changes the challenge for poker bots. Programmers find it very difficult to design and develop AI bots which can counter such hidden information . According to Schaeffer:

Obviously something which bots struggle to do.

Where they do excel is when they use brute computing force to calculate (which is why checkers and chess have both ‘suffered’). They can also run millions of simulations which then gives them ‘experience’ in their battle against human players.

Naturally bots, despite their limitations, have managed to conquer the majority of human opponents online. The recent and ongoing Swedish case has seen a bot ring responsible for close to $500k of winnings against some 25,000+ players, and last year’s PokerStars PLO bot ring saw losses incurred against the cheats run into the millions.

---

No limits?

In addition, heads-up Limit Hold’em has already been conquered by poker robots, with ‘Cepheus’ being the end result of a ten-year project by researchers at the University of Alberta. The machine can never lose to a human, which sounds very scary – ominous, in fact, for the future of poker – but then who even plays heads-up Limit Hold’em? And if you add even one player to the table, the program can no longer be sure of winning.

Still, it could be a sign of things to come. Chess engines started off by ‘solving’ three-piece endgames, then four-piece and are now up to working on seven- and eight-piece endings. As one witty individual put it, we need to worry when they can solve 32-piece positions!

---

Competition is the way forward….

Let's move back to the arena of ‘bot competitions’ where programmers try to outwit their opponents in a variety of different formats.

Knockout events see the bots go head-to-head – the loser being the one with least chips naturally, but this does have an effect on the best playing style for such a format.

Adam Kucharski, from Atlantic.com, says:

On the other hand, there are also competitions which reward the most chips gained. Kucharski states:

to do so, they need to go on the offensive and find ways to take advantage of the others.”;

---

So, the bot programmers make adjustments and run millions of scenarios in the months leading up to the events in order for creations to adapt to the particular needs of each tournament.

This, of course, is more akin to how humans look at poker, and eventually – be it next year, 10 years from now, or in some distant poker future – the strongest humans will meet their match against the ‘ Terminator SkyNet’ version of poker robots.