NetNews Usenet Archive 1992 #27/NN_1992_27.iso/spool/rec/games/programm/4791

Internet Message Format | 1992-11-21 | 1.2 KB

Path: sparky!uunet!spool.mu.edu!olivea!charnel!sifon!thunder.mcrcim.mcgill.edu!snorkelwacker.mit.edu!bloom-picayune.mit.edu!athena.mit.edu!rfermier
From: [e-mail] (Robert G Fermier)
Newsgroups: rec.games.programmer
Subject: Genetic Algorithms
Message-ID: <[e-mail]>
Date: 22 Nov 92 08:21:15 GMT
Sender: [e-mail] (News system)
Organization: Massachusetts Institute of Technology
Lines: 13
Nntp-Posting-Host: m1-142-15.mit.edu

What about having each mutant child play against it's parent some large number
of times, while the user is reading text or thinking or whatever, or run it
as a background process.  Any child which can consistently beat its parent should
be considered a better algorithm.  I suppose this might lead to inbred
algorithms, which are finely tuned for beating their predecessors but suck at
beating humans.  Perhaps having different strains, some of which were based off
of extrapolations from human moves (in position X, he did Y, so our new algorithm
will do Y also now that we are in position X) would solve this problem.

Just an idea, I really don't know much about genetic algorithms.

-- Rob Fermier
   [e-mail]