Is There Any Need for Domain-Dependent Control Information?: A Reply

No. 1: Agents, AI in Art and Entertainment, Knowledge Representation, and Learning

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 13

Planning

In this paper, we consider the role that domain-dependent control knowledge plays in problem solving systems. Ginsberg and Geddis have claimed that domain-dependent control information has no place in declarative systems; instead, they say, such information should be derived from declarative facts about the domain plus domain-independent principles. We dispute their conclusion, arguing that it is impractical to generate control knowledge solely on the basis of logical derivations. We propose that simplifying abstractions are crucial for deriving control knowledge, and, as a result, empirical utility evaluation of the resulting rules will frequently be necessary to validate the utility of derived control knowledge. We ihustrate our arguments with examples from two implemented systems.