Nonmonotonic Probabilistic Logics between Model-Theoretic Probabilistic Logic and Probabilistic Logic under Coherence

Thomas Lukasiewicz

Abstract: Recently, it has been shown that probabilistic entailment under coherence is weaker than model-theoretic probabilistic entailment. Moreover, probabilistic entailment under coherence is a generalization of default entailment in System P. In this paper, we continue this line of research by presenting probabilistic generalizations of more sophisticated notions of classical default entailment that lie between model-theoretic probabilistic entailment and probabilistic entailment under coherence. That is, the new formalisms properly generalize their counterparts in classical default reasoning, they are weaker than model-theoretic probabilistic entailment, and they are stronger than probabilistic entailment under coherence. The new formalisms are useful especially for handling probabilistic inconsistencies related to conditioning on zero events. They can also be applied for probabilistic belief revision. More generally, in the same spirit as a similar previous paper, this paper sheds light on exciting new formalisms for probabilistic reasoning beyond the well-known standard ones.

Introduction

In the field of artificial intelligence, reasoning with "defaults"—statements that are generally true but allow for exceptions—is a critical challenge. Traditional model-theoretic probabilistic logic offers a powerful framework but is notoriously brittle. It often fails or produces trivial results when encountering probabilistic inconsistencies, such as conditioning on events with zero probability. Conversely, probabilistic logic under coherence, rooted in de Finetti’s principles, is mathematically robust but often too conservative. It frequently fails to infer properties that should intuitively be inherited, such as a specific "red bird" inheriting the "flying" property from the general class of birds.

This research aims to bridge the gap between these two extremes by introducing nonmonotonic probabilistic logics that can revise conclusions in the face of new, more specific information.

Methodology

The paper extends two prominent frameworks from classical default reasoning into the probabilistic domain:

  1. Probabilistic System Z: This method organizes a knowledge base into layers (a "z-partition") based on the degree to which rules are "tolerated" by others. It utilizes a probability ranking function ($\kappa$) to quantify the "surprisingness" of various probability distributions, focusing on "z-minimal" models—those representing the least surprising states of the world.
  2. Probabilistic Lexicographic Entailment: Addressing the "inheritance blocking" problem of System Z, this approach is more "adventurous." While System Z may cease property inheritance if a subclass is exceptional in any single way, lexicographic entailment considers the number of rules satisfied at each priority level, allowing for more intuitive property inheritance.

Both formalisms are unified under a semantic framework of admissible probability rankings, ensuring their relationship to the coherence principle is well-defined.

The Hierarchy of Probabilistic Logics

A central contribution of this work is the establishment of a clear hierarchy of logical strength among different entailment relations. As illustrated in Figure 1, the formalisms sit between the extremes of coherence-based logic and model-theoretic logic.

Figure 1: The hierarchy of logical strength, showing that System Z and Lexicographic entailment bridge the gap between coherence-based and model-theoretic logics.

This hierarchy demonstrates that System Z and Lexicographic entailment are stronger than coherence-based entailment, allowing for the inheritance of properties to non-exceptional subclasses (e.g., a "yellow bird" flying), while remaining weaker and more flexible than model-theoretic entailment.

Theoretical Properties and Generalization

The proposed logics satisfy the "Rationality Postulates" (System P), including Rational Monotonicity and Irrelevance. This ensures they can correctly ignore irrelevant data when making inferences. Furthermore, these systems are shown to be proper generalizations of their classical counterparts.

Figure 2: Commutative relationship between probabilistic (top) and classical (bottom) entailment relations, demonstrating how the new formalisms generalize classical default reasoning.

As shown in Figure 2, when probabilities are set to 1 (absolute certainty), the systems collapse into classical System Z and Lexicographic entailment. Additionally, in scenarios without zero-probability conflicts, these logics coincide with standard model-theoretic logic, preserving inferential power where appropriate.

Conclusion and Future Work

This paper provides a principled way to handle "conditioning on zero events" and probabilistic belief revision by focusing on the "least surprising" worlds. While theoretically sound, future work must address the computational complexity of these logics, as finding optimal probability rankings is likely NP-hard. Developing efficient solvers and exploring "weighted" defaults—where some typical rules carry more weight than others—remain key areas for further investigation.

Summary updated

Abstract

Recently, it has been shown that probabilistic entailment under coherence is weaker than model-theoretic probabilistic entailment. Moreover, probabilistic entailment under coherence is a generalization of default entailment in System P. In this paper, we continue this line of research by presenting probabilistic generalizations of more sophisticated notions of classical default entailment that lie between model-theoretic probabilistic entailment and probabilistic entailment under coherence. That is, the new formalisms properly generalize their counterparts in classical default reasoning, they are weaker than model-theoretic probabilistic entailment, and they are stronger than probabilistic entailment under coherence. The new formalisms are useful especially for handling probabilistic inconsistencies related to conditioning on zero events. They can also be applied for probabilistic belief revision. More generally, in the same spirit as a similar previous paper, this paper sheds light on exciting new formalisms for probabilistic reasoning beyond the well-known standard ones.
BibTeX
@inproceedings{lukasiewicz2002nonmonotonic,
  author = {Thomas Lukasiewicz},
  title = {{Nonmonotonic Probabilistic Logics between Model-Theoretic Probabilistic Logic and Probabilistic Logic under Coherence}},
  booktitle = {NMR},
  year = {2002},
  url = {https://openreview.net/forum?id=ON05yWDNQw0},
}
Figure