From: crabbe@usna.edu   
      
   John  writes:   
      
   >     I try to find help here with an exercise from the book Artificial   
   > Intelligence:Modern Approach(2nd edition). I try prepare exercise 21.1 for   
   > my students. The book seems to be good but there are a couple of a simple   
   > confusions between different sources:   
   >   
   >     In passive reinforcement learning:   
   >   
   >  Are the moves of these agents a) fully random b) towards best utility   
   > stochasticly (i.e. prob. of 80%), or c) towards best utility   
   > deterministically?   
   > I.e. What does this fixed policy practically means??   
   >   
   > Either I am stupid or the book should tell this essential assumption more   
   > clearly.   
      
   Well, you're right in that it isn't precisely clear, but wrong in that   
   it is essential.  The main criteria is that the policy is fixed- the   
   learner can't change it at all, but they make no other restrictions.   
   I would interpret that as:   
      
   The policy does not use the utility as calculated by the learner at   
   all (otherwise then the learner would be affecting policy), and they   
   probably mean that the policy is not stochastic, but completely   
   deterministic (but that's just me reading between the lines).  My   
   intuition with no thought whatsoever is that it doesn't matter if its   
   stochastic or not.   
      
   What you really should be doing is asking this on the instructors   
   email list.  Go to   
   http://www.cs.berkeley.edu/~russell/instructors.html   
      
   Stuart Russell often answers questions directly.   
      
   share and enjoy,   
   ric crabbe   
      
   [ 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)