No-limit artificial intelligence

Checkers (or "draughts") has just been cracked. Chess hasn’t yet, but it might as well have been. Inevitably, and for some time, poker has been on the cards (see what I did there?).



Now even "no-limit" poker is being put to the test, as AI gurus pit their program (Polaris) against two human pros - Phil “The Unabomber” Laak and Ali Eslami - in a 500-hand competition being held at the AAAI-07 conference in Vancouver, Canada, tonight.



"Limit" Texas hold 'em, in which the bets in each round are of fixed amounts, has long been well-handled by computers. The parameters and probabilities can easily be quantified. In fact, if you log into any online poker site, the chances are that several players at the table are being told what to do by poker analysis software running alongside their game. "Pot limit", in which you can bet anything up to the amount already in the pot, adds a level of complexity.



But only "no limit" poker elevates the game to something of an art form. Sure, there are still probabilities to calculate, risk versus reward, speculation and so forth, but the sheer fact that, in any betting round, you can decide to push all your chips into the middle takes the game to another level altogether. This is the challenge facing Polaris in Vancouver.



However, anyone who knows poker – and I’ve been playing and studying the game for 15 years – will tell you that, over the course of one evening, even an average player may beat a top-ranking professional. This is because there is such a significant element of luck in the short-term. You can make the right decisions all night only to come out behind. Over time, however, the luck evens out and the good players rise to the top.



The team behind Polaris hope to eliminate "luck" in a clever way. The system will play both pros separately but simultaneously: the cards received by Laak will automatically be the same as those dealt to the computer playing against Ali Eslami. Similarly, the cards played by Eslami will also be the same as those dealt to the version playing Laak. It’s a nice idea and I am very curious to see how it turns out. The art of the bluff, much overplayed in Hollywood, is a very tough skill to develop. I wonder if the pros will sense a computer bluff, having no human opponent to scrutinise...



But I also have a worry. It may be cutting edge poker AI, and they may have evened out the luck to some extent, but poker is nevertheless a game where one session, even a 500-hand session, is a very brief experimental period.



If the pros win tonight, and I hope they do, it won't prove once and for all that humans still have mastery over poker machines. Likewise, if the machine wins, it certainly won't show that humans no longer rule no-limit hold ‘em poker. It is an interesting experiment, but we must keep it in perspective, whatever the outcome.



Sean O’Neill, New Scientist magazine sub-editor

robots' emotions

What if robots not only seemed emotional, but acted on their emotions too? This is the idea behind a project to give a robot called the iCAT, (one of Time magazine's best invention of 2005) "emotional logic", as outlined in this Technology Review article.



The robot itself is made by a team at Philips Research as a tool for experimenting with human-robot interactions. It features speech recognition and servomotors that generate a wide variety of facial expressions to simulate different emotions. See videos of iCAT in action here and here.



And now, Mehdi Dastani and colleagues at Utrech University in the Netherlands are using the robot to test out 22 artificial emotions - including anger, hope, fear and joy - that determine its behaviour.



The Dutch scientists believe that assuming these "emotional" states could help robots perform complicated tasks without getting too bogged-down in planning and analysis. During navigation, for example, many sophisticated robots repeatedly analyse their position and strategy, requires plenty of computer power.



A robot with artificial emotions could perform much less analysis, relying instead on its "feelings". Once it sees that its current plan is going wrong it would become "fearful" and this would modify its behaviour from that point onwards.



Ultimately, Dastani thinks robot "emotions" could help machines interact with people in more sophisticated ways.



Personally, I think it would be interesting to see how such emotionally controlled robots could help with therapy for autism. Over the past few years, several studies have shown that autistic children benefit from interacting with robots. It encourages the development of social skills like making eye contact and paying attention to others. Perhaps an emotioPublish Postnal iCAT could provide a whole new perspective on emotions and social cues.



Magdalena Kogutowska, New Scientist contributor