Operational representation - A unifying representation for activity learning and problem solving

Abstract

A typical AI system engages many levels of cognitive processing from learning to problem solving. The issue we would like to address in this paper is: Can a unified representational scheme be used in learning processes as well as the various levels of cognitive processing from concept representation to problem solving including the generation of action plans? In a previous paper we defined a set of representations called "atomic operational representations" that employs an explicit representation of the temporal dimension and that can be used to ground concepts in the physical world, such as concepts that involve various activities and interactions. In this paper we apply operational representations in a unified manner to the following cognitive processes: 1) the unsupervised learning and encoding of causal rules of actions and their consequences; and 2) the application of the learned causal rules to problem solving processes that produce desired action plans. The unique and explicit temporal characteristic of operational representations is the key feature that allows the encoded concepts to be used in a unified manner across the various levels of cognitive processing. Hence, abstractions in the form of operational representations have an important role to play in AI. Copyright © 2013, Association for the Advancement of Artificial Intelligence. All rights reserved.