Reasoning about action using a possible models approach

Abstract

Ginsberg and Smith [6, 7] propose a new method for reasoning about action, which they term a possible worlds approach (PWA). The PWA is an elegant, simple, and potentially very powerful domain-independent technique that has proven fruitful in other areas of AI [13, 5]. In the domain of reasoning about action, Ginsberg and Smith offer the PWA as a solution to the frame problem (What facts about the world remain true when an action is performed?) and its dual, the ramification problem [3] (What facts about the world must change when an action is performed?). In addition, Ginsberg and Smith offer the PWA as a solution to the qualification problem (When is it reasonable to assume that an action will succeed?), and claim for the PWA computational advantages over other approaches such as situation calculus. Here and in [16] I show that the PWA fails to solve the frame, ramification, and qualification problems, even with additional simplifying restrictions not imposed by Ginsberg and Smith. The cause of the failure seems to be a lack of distinction in the PWA between the state of the world and the description of the state of the world. I introduce a new approach to reasoning about action, called the possible models approach, and show that the possible models approach works as well as the PWA on the examples of [6, 7] but does not suffer from its deficiencies.