Reasoning about action and change has long been of special interest to AI and issues of knowledge representation (see [Sandewall and Shoham, 19941). In particular, the issue of representing changes caused by actions in an efficient and economic way without the burden of explicitly specifying what is not affected by the actions involved and is left unchanged has been a major issue in this area, since typically this specification is huge and in some cases a priori not completely known. In a similar vein, one would also like to avoid explicitly stating all qualifications to actions and all secondary effects of actions. Most of the proposed solutions impose a so-called law of inertia on changes caused by actions which states that properties in the world tend to remain the same when actions occur unless this is known to be otherwise. Formally, the inertia assumption in AI has been treated as some kind of default reasoning which in turn has triggered a host of theories about this specific application and defeasible and nonmonotonic theories in general.