A Statistical Solution to the Qualification Problem and How it Also Solves the Frame Problem

A close relative of the frame problem is the qualification problem. This problem concerns what preconditions an agent considers sufficient for an action to achieve an effect. In general, the consideration of an ideal list of sufficient pre-conditions will be impossible or impractical, and as such, an agent reasoning about action success will be obliged to do so from incomplete evidence. Standard approaches to this problem have been to use non-monotonic or consistency based logical methods that assume those sufficient preconditions which are usually true are true by default. However, these approaches all suffer from a classic problem of default logic called the lottery paradox, as a result of the coarse way that defaults capture statistical properties of the domain. In contrast, we present a novel method for solving the qualification problem using standard techniques for statistical inference. We take it that the agent acquires statistics about the proportion of success of its actions, conditioned upon the existence of certain preconditions which hold just prior to the action.