|
In the real world, there are so many uncertainties that we go on with our days, living and making decisions with only partial information. It would seem that computers aren't ready for the real world yet, because they can only go so far without knowing all the outcomes.
According to Jonathan Schaeffer, head of the University of Alberta's Computer Science Department and a research chair in the field of Artificial Intelligence, that is the problem facing AI development today. Chess has already been used as an important ground for researching AI, but the step up from chess - poker - isn't coming easy.
Why poker?
Well, whereas chess allows the artificial intelligence and its opponent to know everything there is to know about the situation because the game is wholly visible, poker forces players to deal with partial information.
Currently, Artificial Intelligence programmers are finding it difficult to move to
poker games as a testing ground for AI because they haven't figured out a way to write programs that can make decisions from incomplete information. In the real world, Schaeffer says, "knowing everything is so rare. Everything we do all day long is all about partial information. So
poker’s much more representative of what the real world’s like, and in
that sense it becomes a much harder problem."
Schaeffer is also known as one of the designers of Hyperborean, a poker-playing AI that recently won the top prize against other poker-playing AI programs. While being the best is a step up, however, poker AI is one thing and an unpredictable player is another. All in all, the leap from knowledge to guesswork is a big one, and one that won't easily be done.
When they've bridged this gap between the known and the unknown for AI, however, it will be interesting to see if the AI still bases its playing on logic, or on adopting a particular style. Hopefully, that sort of advance in AI research will come sooner than expected.
Besides, wouldn't it be interesting to see an AI actually bluff?
|
|
