Demonstration

Abstract

In this section we describe the results of experiments on three tasks: the Eight Puzzle, Tic-Tac-Toe, and computer configuration (a part of the R1 expert-system implemented in Soar (Rosenbloom, Laird, McDermott, Newell, & Orciuch, 1984)). These tasks exhibit: (1) speed ups with practice; (2) within-trial transfer of learning; (3) across-task transfer of learning; (4) strategy acquisition (the learning of paths through search spaces); (5) knowledge acquisition in a knowledge-intensive system; and (6) learning of qualitatively different aspects of behavior. We conclude this section with a discussion of how chunking sometimes builds over-general productions.