Types of Planning: Can Artificial Intelligence Yield Insights into Prefrontal Function?a

摘要

In this paper, some of the features of models of planning emerging from the area of artificial intelligence (AI) are explored. The goal of this exposition is to explain how researchers are getting machines to attack problems that appear to be similar to those handled in the human prefrontal cortex. In particular, I tried to explain some of the features of AI models that might help explain how planned behavior can occur, with an eye toward examining the specific information-processing constraints necessary for computational models of planning. I described how some AI researchers are converging on a model in which (1) a memory of complex planning information is used in guiding long-term behavior, (2) hierarchically ordered schemata are used to represent this information, (3) activation spreading-like effects occur both in the choice of the memory schemata to use and in monitoring the processing during the execution of those schemata, (4) schemata processing is activated and/or affected by environmental stimuli, and (5) multiple schemata with differing temporal extent are active in parallel. A specific AI planning model, developed in conjunction with Dr. Lee Spector of Hampshire College, was also presented; it was shown how it uses the features above to give rise to interesting planning behaviors for robotic systems.