Causal Theories of Action and Change

Abstract

For many commonsense reasoning tasks associated with action domains, only a relatively simple kind of causal knowledge (previously studied by Geffner and Lin) is required. We define a mathematically simple language for expressing knowledge of this kind and describe a general approach to formalizing action domains in it. The language can be used to express ramification and qualification constraints, explicit definitions, concurrency, nondeterminism, and dynamic domains in which things change by themselves. Introduction It has always been clear that causal knowledge plays a central role in commonsense reasoning about actions. However, it has not always been clear what this role is, or that it cannot be played by non-causal knowledge as well. In the AI literature, this is evident in the use of state constraints for deriving the indirect effects of actions. Intuitively, a state constraint is a proposition that rules out certain states of the world as impossible but says nothing about cau...