Cache flush: AI poker bot to compete against top players in tourney
Carnegie Mellon researchers go all-in
Poker is the next game for AI to beat. Researchers at Carnegie Mellon University have developed Libratus, a computer program that will go head to head with top poker players at Rivers Casino, Pittsburgh, beginning next week.
The hype around AI has been bubbling away for a while within the tech industry, and was brought to attention when DeepMind’s AlphaGo programme fought a highly publicised battle against Lee Sedol, a top Go player.
The evolution of AI can be traced through its various gaming battles with its creators, argues Tuomas Sandholm, professor of computer science at CMU.
"Since the earliest days of AI research, beating top human players has been a powerful measure of progress in the field,".
"That was achieved with chess in 1997, with Jeopardy! in 2009 and with the board game Go just last year. Poker poses a far more difficult challenge than these games, as it requires a machine to make extremely complicated decisions based on incomplete information while contending with bluffs, slow play and other ploys."
The event, titled "Brains Vs. Artificial Intelligence: Upping the Ante", will pit four professional poker players – Jason Les, Dong Kim, Daniel McAulay and Jimmy Chou – against Libratus. Over 120,000 hands of Heads-Up No-Limit Texas Hold’em will be played, with the winner receiving a prize of $200,000.
Poker is not the only game, where AI will be challenged. Baidu, China’s largest web service, has its own AI research lab led by Andrew Ng, will enter a robot in a popular game show Super Brain in China to play against human players in a series of facial and voice recognition tasks, according to the South China Morning Post.
The ability to recognise faces is simple for humans, but a difficult task for computers. It takes heaps of data to train a system to recognise humans.
Contestants on Super Brain in China will have to match participants to their baby photos, and to make it more difficult, the participants are all dressed similarly.
Games are good way to test AI. Simulated environments modelled on the real world may prove useful, and give researchers a way to observe the behaviour of AI in real time.
It’ll continue to be a hotbed for research, as experts tackle increasingly complex games. Last year, Facebook and Google announced its desire to build a superior bot that would be pitted against humans playing Starcraft, a difficult game for AI to master due to the advanced decision making and planning needed.
A win for AI may be exciting, but it’s important to remember that all systems are highly specialised and can only master one game at a time. It’s an ongoing problem experts are trying to tackle with the idea of transfer learning - a process where AI should be able to apply previously learnt knowledge to a solve a new, related task. ®