From news-rocq.inria.fr!jussieu.fr!math.ohio-state.edu!howland.erols.net!newsxfer2.itd.umich.edu!news.itd.umich.edu!ren.us.itd.umich.edu!not-for-mail Mon Aug 19 14:04:36 1996 Article: 5811 of rec.games.corewar Path: news-rocq.inria.fr!jussieu.fr!math.ohio-state.edu!howland.erols.net!newsxfer2.itd.umich.edu!news.itd.umich.edu!ren.us.itd.umich.edu!not-for-mail From: jklewis@ren.us.itd.umich.edu (John K. Lewis) Newsgroups: rec.games.corewar Subject: Re: Artificial Life, Redcode, & Tierra Date: 17 Aug 1996 20:30:46 GMT Organization: University of Michigan Lines: 51 Distribution: world Message-ID: <4v5a5m$5ml@lastactionhero.rs.itd.umich.edu> References: NNTP-Posting-Host: ren.us.itd.umich.edu X-Newsreader: TIN [UNIX 1.3 950824BETA PL0] Tim Chapman (lpyktpc@psyc.nott.ac.uk) wrote: : In real genetics there is an incredible amount of latency. Genetic : patterns can build up over time without actually being active until a : control gene is set. Genetic code can form small units, which can build : upon themselves to develop an advanced system. In genetic corewar this is : not possible as any latent changes in the warrior will crucially affect : it. I disagree. What you have described is very similar to what if happening in the randomly generated code I am working with. The code might carry in it some subroutine which is skipped for generations. After a specific mutation or crossover, that code is used (revealed?) and the basic strategy of the warrior changes. : Maybe better genetics could be simulated by encoding warriors at a deeper : level, where the genetic code is not directly related to the warrior code, : but rather codes for an algorithm to generate the warrior. The warrior : gene could be designed so as to allow sub-units of code to build up in : the gene, but which would not necessarily be present in the warrior unless : a particular control gene was active. This might give a way for more : advanced warriors to be evolved. One idea I was working with was creating "strategies" as genes. An example would be a bomber; Pointer,Bomb,Index,Loopback These are all required for produce a bomber, like this example: loopback mov.i bomb,@pointer add.b index,pointer jmp loopback bomb dat 0,0 pointer dat 0,0 index dat 10,10 Humans see this and say "hey, I can optimize this!" While the computer might do just crazy things like combine the bomb and the index, or make the pointer, index and bomb one thing. More interesting would be mixing two types of warriors/strategies. What if you mixed the bomber and the paper. Any, the point here is that to do this you need to thing up lots of strategies and thier basic needs (pointer,bomb...etc.). While I think this might produce very interesting designs, I want to find strategies overlooked by humans first. This will probably by the next stage I go to after looking ofr GA originated strategies. John- [ john k. lewis ] [ jklewis@umich.edu ] [ 7-3748 ] [ sig.virus v1.4 ]