In conventional capitalist economics one assumes value   
   monism and a scalar utility (AIMA, Russell and Norvig,   
   First edition, pgs 473-484, Prentice Hall, 1995). For an   
   AI this might be something like U = (N-1)/L where N are   
   the number of offspring (diskcopes) the AI has and L is   
   its lifespan. (Perhaps a virus-AI living on the web.)   
   It may be, however, that value monism is wrong and we   
   can not reduce all rewards to a single scalar (Decisions   
   with multiple objectives, Keeney and Raiffa, pg 569,   
   John Wiley, 1976).  In this case a vector utility may be   
   required, something like U vector = ( L, N) where L and   
   N are now vector components.  Performance evaluation   
   using a vector utility has been employed in my Asa H   
   experiment (R. Jones, Trans. Kansas Acad. Sci., vol.   
   109, no. 3/4, pg164, 2006) and (Multiobjective Heuristic   
   Search, Dasgupta, Chakrabarti, and DeSarkar, Vieweg   
   1999). Initial results suggest that use of a vector utility   
   may make the system less likely of get stuck in local   
   maxima.  If the system can not improve L, for example,   
   it may be able to increase N.  After evolving for a while   
   one may then find L and N can both improve.   
      
   [ comp.ai is moderated ... your article may take a while to appear. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)