Path Analysis Models of an Autonomous Agent in a Complex Environment

We seek explanatory models of how and why AI systems work in particular environments. We are not satis ed to demonstrate performance, we want to understand it. In terms of data and models, this means we are not satis ed with descriptive summaries, nor even with predictive models. We want causal models. In this brief abstract we will present descriptive, predictive and causal models of the behavior of agents that ght simulated forest res. We will describe the shortcomings of descriptive and predictive models, and summarize path analysis, a common technique for inducing causal models. This research was supported by ARPA under contracts F30602-91-C-0076 and F30602-93-C-0100; and by AFOSR under the Intelligent Real-time Problem Solving Initiative, contract AFOSR-91-0067.