home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: talk.origins
- Path: sparky!uunet!news.encore.com!bseymour
- From: bseymour@encore.com (Burch Seymour)
- Subject: Re: Laying a trap
- Organization: Encore Computer Corporation
- Date: Wed, 18 Nov 1992 15:56:12 GMT
- Message-ID: <Bxx5Lp.Kwr@encore.com>
- References: <1992Nov18.133247.8546@city.cs>
- Keywords: Computer program, random, mutation, chess
- Sender: news@encore.com (Usenet readnews user id)
- Nntp-Posting-Host: sysgem1.encore.com
- Lines: 30
-
- lionel@cs.city.ac.uk (Lionel Tun) writes:
-
- >Would it be possible to subject ChessMover to random
- >mutations, so that eventually you evolve ChessPlayer,
- >a chess program which plays very well, say at master
- >level?
-
- Not likely, as that isn't how evolution works.
- What (I think) you are describing is this:
-
- try->mutate->try->mutate->try->mutate (etc,etc)
-
- Evolution works as a massively parallel operation with cross sharing
- of information. If a mutation on one individual is *really bad*, that
- individual dies and the trait is gone. If one individual happens upon
- a *really good* mutation, then assuming that individual succeeds in
- reproduction, that trait now has a chance of entering the common
- gene pool.
-
- You program analogy fails to provide mechanisms for the following:
-
- 1) Environmental pressure to build better programs.
- 2) A mechanism for keeping advantageous mutations and discarding
- detrimental mutations.
- 3) A mechanism for producing various combinations of the known,
- working, bits of code.
- 4) A method of doing this experiment in parallel with large numbers
- of sites.
-
- -bs-
-