🚀 go-pugleaf

Tom Fawcett writes....

I'm surprised by this discussion of how to represent blackjack so a
genetic algorithm could solve it.  If you have to put that much effort into
getting the representation just right, it seems like you -- the humans -- have
already done the interesting learning.

Richard Freytag adds....

I agree completely.  It seems that much of the effort involved in GAs is 
in how to map the problem representation onto the GA bit string so that
the search under the selected recombination operator(s) has meaning 
beyond a random search.  The generally, emphasize generally, complete
networks used in neural nets avoid this problem with the representation
topology at the cost of vastly expanded search space from any given state
vector.  I regularly see efforts that do not take significant advantage
of the linkage pattern imposed by the recombination operator with the 
string bit pattern [Goldberg, GAs & Simulated Annealing].  
The result is nearly random search that suceeds 
because there are many satisficing state vectors in the problem space and
no clear idea of what the optimum is.  Best examples can be seen in many
efforts that simulate artificial life.  These papers need to show that
consideration has been given to what the linked alleles would be and map
them as close as possible onto the bit string.  I saw an article in
AI Expert 6 monthes ago that did just that with the TSP.  Results were
presented at GA Conference IV.  Reference escapes me at the moment but
email me and I will send it too you.  

-Richard Freytag