KR Conventicle 2019

The problem of belief change—the problem of how an intelligent agent should update their qualitative beliefs in response to new information—is a crucial issue in knowledge representation. It bears deep formal and conceptual connections to a number of AI subdisciplines, including nonmonotonic reasoning and belief merging, as well as social choice.

The past three decades have seen the emergence of a substantial and sophisticated body of work on this problem, engaging researchers from a variety of disciplines, including artificial intelligence, logic and philosophy.

The present workshop will focus on foundational issues pertaining to the axiomatic models that have been developed in this context. Possible topics include, for instance:

• Connections between qualitative and quantitative models of belief change
• Parallels and discrepancies between belief change and preference change
• Connections between belief change and social choice
• Learning-theoretic aspects of belief revision
• Weakenings or strengthenings of common postulates for belief change
• Connections between belief change and nonmonotonic logics

Talks will include both original contributions and brief topical surveys.

The problem of belief change–the problem of how an intelligent agent should update their belief state in response to new information–is a crucial issue in knowledge representation. It bears deep formal and conceptual connections to a number of AI subdisciplines, including nonmonotonic reasoning and belief merging, as well as social choice.

A substantial and sophisticated literature on this topic has emerged over the past two to three decades, spanning a number of disciplines. In spite of this, there remains widespread disagreement on one of the most fundamental aspects of the core model. Indeed, while the rationality constraints operating on single changes in view have long been well understood, the nature of the principles governing the outcome of a succession of such changes–so-called iterated change–remains surprisingly very much up for grabs. This tutorial provides an up-to-date and systematic survey of work on this vexed question, doubling up as an introduction to the study of belief change more generally.

Projected topics:

1. Single-shot Belief Change (AGM postulates; representation with total-preorders; conditional beliefs and rational consequence; Levi and Harper Identities)
2. Iterated revision I: background (Darwiche-Pearl postulates; Booth & Meyer’s principle (P); reductionist proposals; criticisms of Darwiche-Pearl and reductionism)
3. Iterated revision II: enriched states (revision with ordinal intervals and ranking functions)
4. Iterated contraction I: background (postulates of Chopra et al; reductionist proposals; iterated versions of the Levi and Harper Identities)
5. Iterated contraction II: enriched states (contraction with ordinal intervals and ranking functions)
6. Iterated change in conditional beliefs (Kern-Isberner postulates; other related work)

This tutorial will be a repeat of a tutorial given at IJCAI2019, two weeks prior, in Macau.

Systems and methods for Artificial Intelligence (AI) applicable in the real world require to represent and reason about uncertain knowledge. While this is a limitation of classical first-order logic, there is a large number of so-called non-monotonic logics, i.e., logics that aim to draw inferences only cautiously, allowing for revising them if new information becomes available. Cognitive analysis have shown that human inferential behavior can be better described employing such logics. In this tutorial we introduce the cognitive and formal foundations of cognitive logics, relevant benchmark problems, and challenges in modeling cognitive reasoning.

Projected topics:

1. What is the right logic for human reasoning in AI? — Classical fallacies
2. Formal models of commonsense reasoning (Reiter’s default logic; logic programming and weak completion semantics; conditionals and ranking functions: system Z and c-representations)
3. Cognitive aspects of Cognitive Logics (Cognitive theories and principles, psychological benchmark examples; Cognitive modeling of human reasoning, especially of conditional and syllogistic reasoning; Logical incorrectness of human reasoning and how to escape irrationality; the role of background knowledge in psychological experiments)
4. Formal and cognitive mechanisms of belief revision
5. Discussion