Newsgroups: comp.ai.alife
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!gatech!swrinde!cs.utexas.edu!math.ohio-state.edu!jussieu.fr!oleane!pipex!warwick!bham!bhamcs!comlab.ox.ac.uk!sable.ox.ac.uk!trin0008
From: trin0008@sable.ox.ac.uk (Rick Heylen)
Subject: Re: Tierra Working Group Report
Message-ID: <1995May12.201406.7619@inca.comlab.ox.ac.uk>
Organization: Oxford University, England
References: <3ojf3i$235@gap.cco.caltech.edu> <3odk4d$sb2@gap.cco.caltech.edu> <3oe6s8$4u5@acmey.gatech.edu> <D89zE4.J85@freenet.carleton.ca>
Date: Fri, 12 May 95 20:14:05 BST
Lines: 55

In article <D89zE4.J85@freenet.carleton.ca>,
Tim Sallans <av574@FreeNet.Carleton.CA> wrote:
>However, I would be concerned about an individual on a particular
>platform redesigning their local Tierra environment to provide an interactive
>shell with the outside internet world.  What if someone modified
>the reaper to select for other than error count or complexity of
>operations.  What if they rewrote the Tierra op codes to allow 
>real world interfacing and modified the reaper to reward successful
>interaction.  Then the massive diversity of the connected system would
>serve to feed the creativity of an individual virus maker, feeding them
>to the system at large.

What you're basically saying is that you are worried about people carefully
engineering obscure security holes into their own system! Don't bother!
If people want to compromise the security of their accounts then they can 
easily do this (lets say) by writing a script that takes the headers off 
incoming mail and executes the messages as shell scripts and posts a reply of 
stdout. I think that a distributed version of Tierra is a nice idea and there 
are no security aspects introduced which are problematic.

Personally I think that Tierra is a computationally wastefull simulation
as the implementation of the environment is so different to the computer's own
structure that much computer time is 'wasted' in the translation.
The way that the creatures operate in Tierra with 'templates' is too convenient
to yield profound results. In other words evolution produces viable organisms
too easily in the Tierran environment. I have some ideas of my own on the 
optimal type of environment but I am too busy to explore the possibilities.
From my experience evolving RISC machine code I recommend that surprising and
thought-provoking results are obtained from large programs consisting of small
simple instructions living in a vast universe for a long number of generations.
To put some figures to these principles for my own work
instruction length=4 bytes
program length= ~4kB
universe size= 16MB
time for one instruction= <10e-8 s
It is a simple matter therefore to allow your 'creatures' to evolve for many
millions of generations as each generation takes somewhat less than a 
hundredth of a second to execute.
If the raw instructions for this distributed experiment resembled a fairly
general machine code then the simulations could run quickly due to the 
similarity between the creature code and the native architecture. If the native
architecture allowed some zany operations to execute quickly then these quirks
could be built in to the creature code on that machine. Since not all the 
machines would have these quirks, then creatures would evolve which exploit 
these quirks on one machine would not be portable to other machines.
This could be regarded as specialisation.
I am working on this topic at the moment (but slowly). My hypothesis is that
speciation in the natural world occurs due to adaption to local environments
(ie specialisation) and consequent loss of compatability with other creatures
with the same ancestors.
Any ideas how different species arrise?
I doubt it could be simulated in Tierra.

	Rick Heylen

