Re: Creatures game (was re: AI)

From: Menno Rubingh (rubingh@delftnet.nl)
Date: Sun Dec 19 1999 - 14:13:37 MST


Curt Adams and Alejandro Dubrowski wrote, on Sun Dec 19:

> a simplistic
> neural-net-like brain structure. They do demonstrate adaptive
> behavior without explicit programming,

A group of a few people from this mailing list is right now working,
non-coordinatedly, and in their own spare time, on things like this.
We hope eventually to create a ''self-aware'' AI, that is, one that would
pass a Turing test like a human. We have no working, even simple, program,
*yet*, ... but Rome wasn't built in a day. :-) The things I'm right now,
admittedly perhaps not very ''professionally'', doodling around with,
fit the description you give of the Norns and the kind of environment they
have to learn to ''live'' in, and my ''creatures'' indeed DO use a
neural-network-like brain.

However, unlike the Norns, from what I understand from how you describe
the Norns, our approach is not to code the brain contents genetically
and then have a large population of creatures do genetic evolution.
Instead, in our approach, the individual neuron ''cells'' within one brain
are the things that evolve (multiply, mutate, and so on). Each of
these 'neurons' represents (is the coding, or ''realization'', of)
a behaviour pattern of the creature. Basically, what's coded in one
neuron is the behaviour pattern ''If you see inputs A, B, C then execute
action X''. The creature starts out with a completely blank brain,
and as it receives input and as the output channels of the brain
clamor for attention, the brain itself generates the neurons, as
random combinations of these input and output ''symbols''. As the
creature moves about in its simulated environment, only the successful
behaviour patterns (= neurons) survive. (A neuron causing the creature
to bump into a wall is ''punished'', a neuron causing the creature
to step on a food particle in front of it is ''rewarded'', and
so on.)

( Our (small) group has its own open mailing list, see
        http://www.rubinghscience.org/aiclub/ )

Regards, Menno (rubingh@delftnet.nl)



This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:06:08 MST