i was impressed that the machine managed to outright win even one leg to tell you the truth. granted they were focusing specifically on heads up limit holdem, but i still feel that poker has way too many nuances of psychology involved for a machine to clearly run away with it. there are poker players who play specifically by the numbers, and they don't necessarily win consistently. allegedly, the machine was programmed to pick up routines in the player playing against it and then alter its game according to them. however, this again relies on specific data collected over a certain amount of time which might not necessarily help in a heads up situation.
laak and eslami got together after the first day and discussed that it was best to play a 'middle' game - not too aggressive, not to passive. that's how they ended up beating the machine. that kind of suggests to me that the way in which the machine was made is flawed/biased - ie. it can pick up if you're too aggressive/defensive, but if you're just middle there's no extreme for it to analyze and no opposite for it to work against you. therefore, it came down to man and machine playing an equal middle game and the machine just couldn't compete or adjust fast enough to the players actions, nor did it have an alternate approach.
i was curious though as to how often the machine bluffs for example. i feel like it is very possible to simply induce the machine into bluffing away chips. it will be very difficult for a machine to ever beat man at poker simply because its game is entirely dependent on the person its playing. the machine can't reason its next move outside of how its opponent is playing. given enough hands and enough time i think any decent poker player could figure out how to 'convince' the machine that he has a better or worse hand than he actually has. it's as simple as playing one pattern and then completely playing an opposite pattern. the machine requires multiple hands of data before it realizes you changed. then one can just change again and so on. a human doesn't necessarily require large amounts of data to realize that their opponent has flipped their strategy since psychology obviously plays some role in poker.
Comments
laak and eslami got together after the first day and discussed that it was best to play a 'middle' game - not too aggressive, not to passive. that's how they ended up beating the machine. that kind of suggests to me that the way in which the machine was made is flawed/biased - ie. it can pick up if you're too aggressive/defensive, but if you're just middle there's no extreme for it to analyze and no opposite for it to work against you. therefore, it came down to man and machine playing an equal middle game and the machine just couldn't compete or adjust fast enough to the players actions, nor did it have an alternate approach.
i was curious though as to how often the machine bluffs for example. i feel like it is very possible to simply induce the machine into bluffing away chips. it will be very difficult for a machine to ever beat man at poker simply because its game is entirely dependent on the person its playing. the machine can't reason its next move outside of how its opponent is playing. given enough hands and enough time i think any decent poker player could figure out how to 'convince' the machine that he has a better or worse hand than he actually has. it's as simple as playing one pattern and then completely playing an opposite pattern. the machine requires multiple hands of data before it realizes you changed. then one can just change again and so on. a human doesn't necessarily require large amounts of data to realize that their opponent has flipped their strategy since psychology obviously plays some role in poker.
http://www.cardplayer.com/poker-news/article/9324