Michael Dale
Software Development
«
Utility Theory Based Multi Agent AI Analysis
The purpose of my thesis was to explore is under siege and the agents act as
analysis features for Artificial Intelligence variety of tasks to preserve themselves
optimization techniques and helpful
(AI) within a system where thousands
of agents operate individually. My
and their castle.
inspiration for this topic came from my I created several systems within my own
Online games with large numbers of a live function profiler, data exporter/
love of strategy and Massively Multiplayer
Non-Playable Characters, as well as Dave
Mark’s work on the power of Utility Theory
driven AI.
In order to have the ability to test
optimization methods and analyze them
to determine their effectiveness while
scaling, I first developed a simulation
that created a context and sufficient
complexity for my agents to operate
within, including dynamic events. I
developed a scenario where a castle
82
townspeople dedicated to performing a
SOFTWARE DEVELOPMENT
game engine to analyze my AI, including
importer, and statistical visualizations.
These systems acted as the basis for
developing and testing optimizations
using techniques such as improved
pathing, memorization, amortization and
data organization. Over the course of
the last two semesters, this project has
given better insight into the power and
implementation of a utility theory driven AI
system and general best practices for the
methods of optimizing and analyzing AI.