An Architectural Account of Variation in Problem Solving and Execution

Background and Motivation Theories of the cognitive architecture (Langley et al., 2009) aim to specify the constant features of the human mind. They make assumptions about the representation and organization of memory, as well as about performance and learning mechanisms that operate over its structures. They specifically do not include content that changes, whether rapidly through reasoning and problem solving or gradually through learning. However, we also know that humans are highly adaptive and they can work on a given task in many different ways. This creates a tension between the desire to identify universals in human cognition and to explain observed variation. Some architectures, like Soar (Laird et al., 1987) and Epic (Kieras & Meyer, 1997), make few commitments except for knowledge representation and the basic cognitive cycle. In contrast, Icarus (Langley et al., 2009) makes more specific claims about structures and mechanisms, committing to means-ends analysis for problem solving and to reactive control for execution. In recent research, we have sought a middle ground, devising a new cognitive architecture that incorporates stronger assumptions than Soar and Epic, but weaker ones than Icarus. In this paper, we present the phenomena it attempts to cover, its theoretical postulates, and planned extensions to the framework. We will see that the theory builds on classic ideas but combines them in novel ways that move beyond its predecessors.