Information Technology Reference
In-Depth Information
Chapter 17
Interactive Meta-Reasoning: Towards
a CAD-Like Environment for Designing
Game-Playing Agents
Ashok K. Goel and Spencer Rugaber
Abstract
We posit that experimentation is a central element of the creative process.
The question then becomes how can we support experimentation in creative tasks?
We take inspiration from the success of computer aided design (CAD) environments
that enable designers to construct, evaluate and revise models of engineering sys-
tems. Design of game-playing software agents is another creative task. By analogy,
we present a CAD-like environment that enables designers to construct, evaluate
and revise models of game-playing agents. However, unlike engineering systems,
intelligent agents may learn from experience. In particular, intelligent agents may
use meta-reasoning over their own models to redesign themselves. Thus, we envision
a CAD-like environment in which the human designer and the intelligent software
agent cooperate to perform interactive meta-reasoning to redesign the agent. In this
article, we describe three elements of this vision: (2) an agent modeling language
called TMKL2, (2) an interactive environment called GAIA for experimenting with
the models of game-playing software agents, and (3) GAIA's module called REM
that performs meta-reasoning for self-adaptation in game-playing software agents.
We illustrate these concepts for the task of design of software agents that play variants
of Freeciv, a turn-based strategy game.
17.1 Background, Motivations and Goals
Generate and Test
is a common general-purpose method in artificial intelligence
(AI) [
79
]. Given a problem, in the Generate and Test method, an intelligent agent
first generates a candidate solution and then evaluates the solution. If the solution is
acceptable to the agent, then the problem is solved; If not, then the agent may generate
another solution for evaluation. The agent may repeat this until an acceptable solution
Search WWH ::
Custom Search