XPost: comp.ai.games   
   From: wilson.max@gmail.com   
      
   >3) The thing is I'm praticing AI design aswell, althougth this is of   
   >course a rather simple problem - the way I'm tackling it at least. The   
   >problem is I'm not sure which class should take care of the methods of   
   >getting the next possible states. Should the puzzle itself provide the   
   >next possible moves when asked or should the agent have to take care   
   of   
   >finding the next state?   
      
   Ah, I see. You're not asking about class design.   
      
   It depends upon the flavor of AI that you're making. If it's supposed   
   to have complete knowledge of its environment (like in chess), then it   
   doesn't matter which _class_ does the fringe expansion so long as the   
   AI can request it at will. Effectively it has to be part of the AI's   
   mind. I forget the terminology for this, but in Markovian terms it   
   means that the agent knows the state transition function f(state,   
   action) = {next-state, reward}. If on the other hand, you want the   
   agent to be ignorant of the state transition function (say you're doing   
   Q-learning or something) then you put the state transition function in   
   the environment (the puzzle), and the agent has to learn through trial   
   and error what the rules are for moving. Note: you still need to make   
   the environment give a list of next possible _actions_, but that's not   
   the same thing as letting it know what next-state results from each   
   action.   
      
   Keywords: Markov Decision Process (MDP), Q-learning.   
      
   Max Wilson   
      
   [ comp.ai is moderated.  To submit, just post and be patient, or if ]   
   [ that fails mail your article to , and ]   
   [ ask your news administrator to fix the problems with your system. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)