CAJ | 학술논문

人工智能研究中，行动这一概念通常在理论框架中有完全的定义。然而，现实中的行动有时难以完全刻画。智能体需要从过去的经验观察中习得行动的后果。本文提出一种基于时态结构的行动一结果学习理论。在自然数时间结构中，智能体通过观察过去的恒常联系建立因果关系。智能体依据己建立的因果关系指导将来的行动。同时，我们给出关于该理论的一个完全的逻辑演绎系统，并给出基于该逻辑的智能体行动的有效算法。
인공지능연구중，행동저일개념통상재이론광가중유완전적정의。연이，현실중적행동유시난이완전각화。지능체수요종과거적경험관찰중습득행동적후과。본문제출일충기우시태결구적행동일결과학습이론。재자연수시간결구중，지능체통과관찰과거적항상련계건립인과관계。지능체의거기건립적인과관계지도장래적행동。동시，아문급출관우해이론적일개완전적라집연역계통，병급출기우해라집적지능체행동적유효산법。
Actions in Artificial Intelligence, such as in planning and situation calculus, are well defined. Effects of actions are given in the design. In reality, however, not all actions＇ effects are known by the agent. She must learn and attribute the causation between actions and effects through her observations. This paper presents a logic for learning effects of actions based on a simple temporal model. Time points are modeled by natural numbers, at which the agent observes constant conjunctions in the past. By simple induction on these data, the agent attributes and revises the causation between actions and effects. Then she uses it to arrange her future actions for some given goal. A complete deductive system and a tractable algorithm of decision making by the logic are given in the paper.