Belief change in rational agents

在人工智能的知识表示(KR)中，智能体处理的问题域可以是特定的问题域(如诊断)、物理域(如机器人遇到的问题域)或虚拟域(如由Web提供的域)。在所有这些情况下，代理人的知识通常都是不完整的，在某种程度上不准确，并且不断变化。KR中的一个重要问题是如何管理这种信息的变化，无论主体是被告知了一个新的事实，还是采取了一些行动，或者被给予了一组来自不同来源的观察，本研究的主要焦点是信念变化，即主体应该如何根据新的信息来修改其信念。虽然已经开展了大量工作，但仍有一些关键问题尚未得到充分解决。例如，我们所拥有的正式模型是否完全描述了信念变化的所有方面，这一点并不清楚；此外，对于代理人应该如何纳入关于固定环境的一系列观察，也没有达成一致。这项研究将解决信念改变的问题，以及在一个可以执行行动的动态代理世界中进行推理的更广泛的问题。首先，将开发描述信念变化各方面的数学模型，其次，将根据计算模型在不同领域的适用性来开发和研究计算模型。第二，研究主题涉及与偏好的推理。通常，代理不仅有(严格的)目标要实现，而且还有需要建立的可取但不必要的东西。此外，还将讨论偏好在规划中的使用，以及偏好在信念改变中的作用。

In Knowledge Representation (KR) in Artificial Intelligence, an agent deals with a problem domain which could be a specific problem area (such as diagnosis), a physical domain (such as encountered by a robot) or a "virtual" domain (such as provided by the web). In all such cases, the agent's knowledge will generally be incomplete, inaccurate to some extent, and constantly changing. An important problem in KR is to manage this change of information, whether the agent has been told a new fact, or has carried out some actions, or has been given a set of observations from different sources.The main focus of this research is on belief change, which deals with how an agent should modify its beliefs on the basis of new information. While much work has been carried out, there are crucial issues that have yet to be fully addressed. For example, it is not clear that the formal models that we have fully characterize all aspects of belief change; as well, there is no agreement as to how an agent should incorporate a sequence of observations about a fixed environment. This research will address the problem of belief change, and the wider problem of reasoning in a dynamic world of agents that can carry out actions. First, mathemetical models characterizing aspects of belief change will be developed, and second, complementary to this, computational models will be developed and studied in the context of their applicability in different areas.A secondary, but strongly related, topic of research concerns reasoning with preferences. Typically an agent will have not just (strict) goals to attain, but also things that are desirable, but not necessary, to establish. The use of preferences in planning, as well as the role of preferences in belief change, will also be addressed.