"Inside  every small problem is a large problem struggling to get out."

## Thread and Full Text View

Respond to the question: Probability with human factor hm?

 10/29/2006 02:04 AM by Tomas Jankus; Probability with human factor (hm?..)

Hi all :-)

i was programming an AI for a game and got stuck with one problem. I don't know if you will be willing to help me, but i'll give it a try :-)

it would be difficult enough to explain the concepts of the game i am working with, so let me give you a simplified example (stupid one :-)).

so, let's say the player rolls the die (1 to 6) and has two options - to stay with it (player will win (+1) if opponent rolls lower or same value, and he will lose (+0) if opponent rolls higher), or not to take a gamble and to go for a tie (player gets +0.5 immediately).

then, let's say we know that player goes for a gamble 50% of the time.

and now the problem. it would be logical to decide that the player stays to play if he rolled 6, 5, or 4 (50% best), but it is not! the one can stay with 1 some 5% of the time (just because he has a chance to win if opponent rolls the same 1), while not taking a chance with 4 again some 5% of the time. so, whenever the player stays to play i need to get a table like the probability he rolled 6=30%, 5=28%, 4=23%, 3=12% 2=5% 1=2% (a total of 1, or 100%).

the question is, is it possible to make some logical distribution from the given data, or should AI observe the percentage for each value seperately (wich is almost imposible having short runs in a game with much higher variablity)? or maybe the AI should gather some empyrical information, like "players with 50% play 1 no more than 5% of the time"?

any thoughts would be great,