Planning algorithms and planning problems

摘要

A body of recent work in plan generation can be characterized as an attempt to under-stand the "fit" between design decisions that can be made in developing planning algorithms, and characteristics of the problems those algorithms are intended to solve. In this talk, I will survey this work, and relate it to more general questions about the role of experimentation and evaluation in the AI enterprise. In particular, I will argue that the doing of AI and the evaluating of an AI system can often be one and the same activity: both amount to understanding whether and why a system is doing the right thing.