行动推理中循环因果关系的研究

近年来行动推理的研究成为人工智能领域的热门问题，而结果问题是目前行动推理研究的核心问题之一．该文针对许多行动推理系统不能处理循环因果关系的问题，提出了解决方法．基于适当修改后的McCain和Turner的因果理论，该文提出了一种能处理循环依赖的理论转化方法．转化后的因果理论消除了循环依赖，而且可以采用单调推理方法．基于因果闭包语义，证明了转化前后的因果理论具有相同的模型．当因果关系不存在循环依赖时，该文方法得到与McCain，Turner方法(1997)相同的结果．

Cyclic Causality in Reasoning about Action

Intelligent behaviors of humans require the ability to predict the result of actions. McCarthy holds the view that reasoning about action plays an essential role in commonsense reasoning. In recent years, reasoning about action is a popular issue in AI domain and becomes a discussion subject in IJCAI for several years. Ramification problem is currently one of the central problems is reasoning about action.This paper describes some background and the main task of reasoning about action, recapitulates the ramification problem and points out the dificulty in it. Most of the current methods of reasoning about action cannot handle cyclic causal relation. This paper presents a solution to this problem. Firstly, based on modified causal theory of McCain and Turner, it puts forward an automatable procedure that transforms a syntactically restricted causal theory into a monotonic one without cyclic causality. Next, it proves the correctness of the procedure under causal closure semantic by showing that an interpretation is the model of the transformed theory if and only if it is the model of original theory. This paper also proves that the method yields the same results as that of McCain and Turner (1997) when the theory is stratified.