Probabilistic Strategy Selection for Flexible Cognition

While current artificial systems must be custom tailored to operate within a single domain of expertise, the human mind readily adapts to multiple domains. In moving toward artificial systems that display similar flexibility, we become interested in how efforts to learn about the mind can support the development of flexible artificial systems, and how the development of these systems can contribute to our understanding of how the mind works. When contrasting natural and artificial systems, we find that people learn and maintain multiple competing strategies for performing the same task, while current software lacks this ability. When adding numbers, asking questions, referring to objects in the environment, or discerning spatial relationships, people develop and flexibly use a variety of approaches. This paper describes a proposed research program that focuses on computationally modeling the ability to maintain and use multiple strategies for performing the same task. We first point to developmental evidence as motivation for this research program. We then describe a few computational efforts that seem to fall within this vein. Finally, we describe preliminary work within the field of lexical semantics based on a new proposed model of “word strategies”.