Joint revision of belief and intention

Proceedings of the Eleventh International Joint …, 1989

The study of belief revision and reasoning about actions have been two of the most active areas of research in AI. Both these areas involve reasoning about change. However very little work has been done in analyzing the principles common to both these areas. This paper presents a formal characterization of belief revision, based on the principles of minimal change and maximal coherence. This formal theory is then used to reason about actions. The resulting theory provides an elegant solution to the conceptual frame and ramification problems. It also facilitates reasoning in dynamic situations where the world changes during the execution of an action. The principles of minimal change and maximal coherence seem to unify belief revision and reasoning about actions and may form a fundamental core for reasoning about other dynamic processes that involve change.

2008

We present a logical setting that incorporates a belief-revision mechanism within Dynamic-Epistemic logic. As the " static " basis for belief revision, we use epistemic plausibility models, together with a modal language based on two epistemic operators: a " knowledge " modality K (the standard S5, fully introspective, notion), and a " safe belief " modality 2 (" weak " , non-negatively-introspective, notion, capturing a version of Lehrer's " indefeasible knowledge "). To deal with " dynamic " belief revision, we introduce action plausibility models, representing various types of " doxastic events ". Action models " act " on state models via a modified update product operation: the " Action-Priority " Update. This is the natural dynamic generalization of AGM revision, giving priority to the incoming information (i.e., to " actions ") over prior beliefs. We completely axiomatize this logic, and show how our update mechanism can " simulate " , in a uniform manner , many different belief-revision policies.

The SharedPlans theory provides an axiomatic framework of collaborative plans based on four types of intentional attitudes. However, there still lacks an adequate semantics for the 'potential intention' operators. In this paper, we give a formal semantics to potential intentions, and examine models that can validate various relations between beliefs, intentions, and potential intentions.

STAIRS 2004: Proceedings of the Second Starting …, 2004

Artificial Intelligence, 2003

This paper presents a general, consistency-based framework for expressing belief change. The framework has good formal properties while being well-suited for implementation. For belief revision, informally, in revising a knowledge base K by a sentence α, we begin with α and include as much of K as consistently possible. This is done by expressing K and α in disjoint languages, asserting that the languages agree on the truth values of corresponding atoms wherever consistently possible, and then re-expressing the result in the original language of K. There may be more than one way in which the languages of K and α can be so correlated: in choice revision, one such "extension" represents the revised state; alternately (skeptical) revision consists of the intersection of all such extensions. Contraction is similarly defined although, interestingly, it is not interdefinable with revision.

1993

Autonomous Agents and Multi-Agent Systems, 2008

This article introduces a new theory of intention representation which is based 1 on a structure called a Dynamic Intention Structure (DIS). The theory of DISs was motivated 2 by the problem of how to properly represent incompletely specified intentions and their 3 evolution. Since the plans and intentions of collaborating agents are most often elaborated 4 incrementally and jointly, elaboration processes naturally involve agreements among agents 5 on the identity of appropriate agents, objects and properties that figure into their joint plans. 6 The paper builds on ideas from dynamic logic to present a solution to the representation 7 and evolution of agent intentions involving reference to incompletely specified and, possibly, 8 mutually dependent intentions, as well as the objects referenced within those intentions. It 9 provides a first order semantics for the resulting logic. A companion paper extends further the 10 logical form of DISs and explores the problem of logical consequence and intention revision.

Lecture Notes in Computer Science, 1992

This paper contains a brief survey of the area of belief revision and its relation to updating of logical databases as it has developed during the last years. A set of rationality postulates for revisions are presented. A couple of semi-algorithmic models of revision and updating are introduced. The models are connected to the postulates via some representation theorems. Finally, the relation between belief revision and nonmonotonic logic is explored.

In this paper we present a the structuralist framework of belief revision.This is considered to be an extension of AGM framework of belief revision with some semantic properties which constitute the structure. The properties are causal properties, causal mechanism, causal process, intervention. In the theories of belief revision based on classical AGM [1], beliefs are being represented as a set of sentences that are closed under logical implication. Operations for belief change such as expansion, contraction and revision, are defined only with respect to belief sets. As a result , structural information given by an epistemic entrenchment ordering disappears after operations of belief change. Although the AGM model is mathematically elegant, simple, it only merely serve as an idealized model for belief change. Also, some postulates such as success(in case of revision), recovery (in case of contraction) are found to be unsatisfactory in case of rational belief revision. Success Postulate states that new information needs to be reflected in the revised belief state of an agent, no matter whether it is relevant or not. An agent has no choice of rejecting the undesirable irrelevant information. Some other postulates are also found to be controversial, counter-intuitive. Belief revision models based on AGM are geared up for one step revisions of simple plain beliefs and fails to handle cases where the new information comes in iterations and in the form of conditional beliefs. Keeping these issues in mind the problem seems to be as follows: how does an agent after discovering causal relations from the environment revises her beliefs such that beliefs closer to the core are not lost? The problem lies in finding a mechanism with the help of which one can give up conditional beliefs in the process of Belief revision while preserving the core conditional beliefs such as laws or causal statements. We present a theory of belief revision which preserves the structural information after iterated revisions. Following the constructive approach to belief revision, we propose a new entrenchment ordering called " causal epistemic en-trenchment ordering " , which is constrained by the structural and semantic properties such as structure, intervention, causal properties, causal mechanisms. We show how the ordering of beliefs are influenced by these semantic factors. The resulting entrenchment ordering of beliefs (CEE) would help us in ordering conditional beliefs (1) 2. preserve the structure in iterated revisions 3. It can be applied to the scientific theory change.

1994

In this paper we report preliminary results on how people revise or update a previously held set of beliefs. When intelligent agents learn new things which conflict with their current belief set, they must revise their belief set. When the new information does not conflict, they merely must update their belief set. Various AI theories have been proposed to achieve