Generate, Test and Debug: A Paradigm for Combining Associational and Causal Reasoning

Abstract

Efficiency and robustness are two desirable, but often conflicting, goals of problem solvers. This paper examines how a combination of associational and causal reasoning can be used to achieve both goals. We describe the Generate, Test and Debug (GTD) paradigm, which uses associational reasoning to solve most problems efficiently, while relying on causal reasoning to maintain overall robustness. The problem-solving characteristics of associational and causal reasoning are presented, based on an analysis of the types of knowledge and reasoning used in GTD. In particular, we argue that the characteristics depend largely on the extent to which interactions between events are represented and reasoned about — associational reasoning is efficient because it uses rules that (nearly) encapsulate interactions, while causal reasoning is robust because it analyzes the effects of events and their interactions.