Reasoning about actions: steady versus stabilizing state constraints

摘要

In formal approaches to commonsense reasoning about actions, the Ramification Problem denotes the problem of handling indirect effects which implicitly derive from so-called state constraints. We pursue a new distinction between two kinds of state constraints which will be proved crucially important for solving the general Ramification Problem. Steady constraints never, not even for an instant, cease being in force. As such they give rise to truly instantaneous indirect effects of actions. Stabilizing state constraints, on the other hand, may be suspended for a short period of time after an action has occurred. Indirect effects deriving from these constraints materialize with a short lag. This hitherto neglected distinction is shown to have essential impact on the Ramification Problem: if stabilizing state constraints interact, then approaches not based on so-called causal propagation prove defective. But causal propagation, too, is shown to risk producing anomalous models, in case steady and stabilizing indirect effects are propagated indiscriminately. Motivated by these two observations, we improve the theory of causal relationships and its Fluent Calculus axiomatization, which both are methods of causal propagation, so as to properly handle the distinction between steady and stabilizing constraints.