A Model-Based Approach to Blame Assignment: Revising the Reasoning Steps of Problem Solvers

摘要

Blame assignment is a classical problem in learning and adaptation. Given a problem solver that fails to deliver the behaviors desired of it, the blame-assignment task has the goal of identifying the cause(s) of the failure. Broadly categorized, these causes can be knowledge faults (errors in the organization, content, and representation of the problem-solver's domain knowledge) or processing faults (errors in the content, and control of the problem-solving process). Much of AI research on blame assignment has focused on identifying knowledge and control-of-processing faults based on the trace of the failed problem-solving episode. In this paper, we describe a blame-assignment method for identifying content-of-processing faults, i.e., faults in the specification of the problem-solving operators. This method uses a structure-behavior-function (SBF) model of the problem-solving process, which captures the functional semantics of the overall task and the operators of the problem solver, the compositional semantics of its problem-solving methods that combine the operators' inferences into the outputs of the overal task, and the "causal" inter-dependencies between its tasks, methods and domain knowledge. We illustrate this model-based blame-assignment method with examples from AUTOGNOSTIC.