The Cognitive Architecture Approach to AI Research

Soar is a cognitive architecture that has been under development for close to 25 years. Recently, we have begun a set of major architectural extensions to Soar including reinforcement learning, episodic memory, semantic memory,
emotion, and visualization/imagery. The goal of these extensions is to greatly enhance Soar's ability to support human-level behavior. In this talk I will describe our plans for the integration of these components with special emphasis on the methodology we have adopted. The goal of our methodology is to avoid depending on great moments of insight to move the work along, but instead define a process that deliberately (but not necessarily slowly) moves the research forward. Our methodology includes an analysis phase where we attempt to determine functional, behavioral, structural, computational, and integrative constraints for each architectural extension. We then attempt to develop a conceptual framework that defines the space of possible designs for each extension. Each framework is used as a generator for possible designs, which are then evaluated based on our requirements, followed by prototype implementations, further evaluation, new designs, and so on.
I received my B.S. from the University of Michigan in 1975 and my Ph.D. in Computer Science from Carnegie Mellon University in 1983. My thesis advisor was Allen Newell. I was a member of research staff at Xerox Palo Alto Research Center from 1984 to 1986. Since 1986, I have been on the faculty of the Computer Science and Engineering Division of the Electrical Engineering and Computer Science Department of the University of Michigan where I am a Professor. I am a founder of Soar Technology, an Ann Arbor company specializing in creating autonomous AI entities based on Soar.