Greg Barton - 2005-08-24

The title of this post was to be "Components of a disgustingly overcomplex evolving system" but that was too long. :)

I know I'm leaping ahead a bit here.  I don't even have a working RETE implementation yet. :)  But here are some ideas on what kind of systems I'd like an evolvable RETE to power.

Basically, my basic idea is this: real, pie in the sky AI, will only be accomplished by a multilevel grossly complex system of evolving systems.  THis is the only way I can see to simulate the circumstances that produced the only form of intelligence we currently recognise: us.  So, at the moment, these are the evolving components as I see them, each of which could be powered by an evolving RETE, or a collection of them:

1. The Agent:
  This would be the prime actor in the system.  Agents attempt to solve the objective function.  To do this, they use their own evolved abilities, Tools, and Memes.

2. The Tool
  A Tool is a RETE that is not directly evolved by the system, but is evolved by the Agents themselves.  Agents could start out with varying abilities to create and use tools.  The Tool creation/use/trade process could lead to a form of evolution, emerged by the actions of the Agents.

3. The Meme
  A Meme is a fact or set of facts (working memory objects) that is traded between Agents.  The trading (copying) process is imperfect, so mutation based evolution can occur.

4. The evolutionary framework itself
  This opens the door to infinite self reference, but since the process by which entities interact, the evolutionary rules themselves, could change, why not make that change evolutionary in nature?  There would have to be an immutable base set of rules. (unless I can think of some way for the snake to eat it's own tail)  Also, this change would have to be "slow" for the overall system to be "stable."