Artificial Evolution of Intelligence:  Lessons from natural evolution
	An illustrative approach using Genetic Programming 
David Andre

A Senior Honors Thesis submitted to 
Symbolic Systems Program
Stanford University
in partial fulfillment of requirements for the Bachelor of Science 
Spring, 1994

Abstract 

	Given that natural evolution has produced the only examples of
intelligent agents, perhaps Artificial Intelligence can learn a great
deal from investigating the processes of evolution that created Human
Intelligence.  This thesis presents an approach in which several
principles important in natural evolution are analyzed and used to
motivate an autonomous agent architecture that incorporates modules
from many different approaches to AI.  The modules, consisting of many
partial solutions to the various tasks required of an autonomous
agent, interact with one another to solve problems.  This forced
interaction leads to an internal representation that is developed
on-line, as the agent's modules learn to interact.  Although
full-scale systems of this type have not yet been implemented,
empirical investigations into simple models containing the core
aspects have been successful.  Utilizing a genetic programming
paradigm, the research to date indicates that multi-module individuals
capable of learning, storing, and utilizing information about their
simple environments to produce plans of action can be evolved using
this approach.  In addition, this thesis presents a brief history of
genetic programming and its significance for the field of artificial
intelligence, and presents a public domain tool for genetic
programming.