Reasoning with Incomplete Knowledge in a Resource-Limited Environment: Integrating Reasoning and Knowledge Acquisition
CARNEGIE-MELLON UNIV PITTSBURGH PA ROBOTICS INST
Pagination or Media Count:
This paper describes an approach to reasoning with incomplete information in a resource-limited environment. Approaches to date either assume infinite resources and proceed to enumerate a large inference space, or assume few resources and ignore the missing information. They do not reason about resource constraints and the inference methods admissable under them. A Hearsay- II-like system is described where each knowledge source is a separate production system. During rule evaluation, a rule antecedant is evaluated using minimal- resource methods. A rule antecedant is evaluated to true, false, or an expected resource cost to acquire the information necessary to complete its evaluation. If conflict resolution chooses a partially evaluated rule, it posts a goal asking other knowledge sources to provide the missing information, suspends the knowledge source, and informs the knowledge sources manager about the suspension and accompanying goal. The manager decides whether the goal is worth pursuing now, that amount of resources to apply to the task, what knowledge sources to apply, and when to give up. The knowledge sources that attempt the goal can implement a variety of inferential and knowledge acquisition techniques.