Interactive Meta-Reasoning: Towards a CAD-Like Environment for Designing Game-Playing Agents

Abstract

CAD environments for conceptual design in architecture and engineering support design creativity by enabling rapid generation, testing, and revision of design concepts. By analogy, we present a CAD-like environment for conceptual design of game-playing agents that is intended to support design creativity by enabling rapid generation, testing and revision of agent designs. Our environment for designing game-playing agents, called GAIA, provides a high-level language, called TMKL2, for enabling interactive conceptual design of game-playing agents, automated tools for testing the agent's design through its execution in the game world, and access to the agent's reasoning, knowledge and behavior for affording interactive agent redesign. In addition, GAIA provides an automated meta-reasoner called REM that enables the game-playing agent to self-diagnose and self-repair of its reasoning strategies, thus providing critical support to the designer in the task of agent redesign. We illustrate GAIA's various abilities through examples from the multi-player turn-based strategy game called Freeciv. GAIA itself illustrates the use of interactive meta-reasoning in supporting collaborative human-computer creativity.