Abstract

Abstract. There is a critical and urgent need for automated analytical agents operating in complex domains to provide meta-level explanations of their reasoning and conclusions. In this paper, we identify the principles for designing analytical agents that can explain their reasoning and justify their conclusions at different levels of abstractions to potential human customers with varying goals. We also analyze the goals of users of an automated agent for investigative analysis along the dimensions of why, what, how and when, and develop a taxonomy of human goals that will leverage the explanations generated by the agent. 1

Citations