Possibility as similarity; the semantics of fuzzy logic

Enrique H. RuspiniUAI 1990

TL;DR

This paper proposes a novel semantic model for fuzzy logic, grounding it in the concept of "similarity" between possible worlds. This approach offers a more intuitive and conceptually unified understanding of fuzzy logic, distinguishing it from probabilistic reasoning and providing a natural extension of classical modal logic.

What Problem Does This Paper Address?

Fuzzy logic, a powerful tool for handling imprecise and uncertain information, has historically been understood through various formalisms. This paper aims to clarify the fundamental conceptual and functional differences between fuzzy (possibilistic) logic and probabilistic reasoning. While both deal with uncertainty, they do so in fundamentally different ways. Probabilistic reasoning quantifies uncertainty based on the frequency or likelihood of events (set extension), whereas fuzzy logic deals with degrees of truth and vagueness. The authors seek to provide a unified semantic framework for fuzzy logic that is conceptually distinct from probability theory and can be understood as a natural extension of modal logic, focusing on the notion of similarity between states of affairs.

Methods

The core of the paper's contribution is a semantic model based on "possible worlds." These are defined as conceivable states of affairs consistent with the laws of logic. The key innovation is the introduction of a generalized similarity relation, a function that quantifies the resemblance between any two possible worlds. This similarity relation, ranging from 0 (no resemblance) to 1 (identity), forms the basis for defining fuzzy logic's core concepts.

Specifically, the authors:

  • Generalize the accessibility relation from modal logic: Instead of a binary relation indicating accessibility between worlds, they introduce a family of relations $R_\alpha$, where $w R_\alpha w'$ if the similarity $S(w, w') \ge \alpha$. This allows for graded degrees of accessibility.
  • Define similarity functions: These functions $S(w, w')$ are required to satisfy properties like reflexivity, symmetry, and a relaxed form of transitivity, aligning them with the concept of a metric. They are shown to be related to triangular norms (e.g., Zadeh, Lukasiewicz, product norms).
  • Formalize degrees of implication and consistency: Using the similarity relation, they define graded notions of implication ($I(p|q)$) and consistency ($C(p|q)$), which quantify how much one set of possible worlds needs to be "stretched" to encompass or intersect another.
  • Interpret possibilistic distributions: Necessity and possibility distributions are framed as measures of proximity between the evidential set (worlds consistent with evidence) and other sets of possible worlds.
  • Generalize modus ponens: The paper shows how Zadeh's generalized modus ponens can be understood as a form of logical extrapolation, leveraging similarity to infer conclusions even when there isn't a perfect match between evidence and rules.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Similarity as the foundation of fuzzy logic: The paper argues that fuzzy logic can be elegantly understood as a logic of similarity between possible worlds, offering a more intuitive semantic grounding than previous approaches.
  • Fuzzy logic as an extension of modal logic: By generalizing the accessibility relation with a graded similarity function, the model shows how fuzzy logic concepts naturally extend classical modal logic.
  • Conceptual distinction from probability: The model clearly delineates fuzzy logic from probabilistic reasoning by emphasizing that similarity between states doesn't imply increased likelihood, unlike probabilistic measures of set extension.
  • Graded implication and consistency: The introduction of quantitative measures for implication and consistency allows for a more nuanced understanding of how propositions relate to each other in the presence of uncertainty.
  • Generalized modus ponens as extrapolation: Viewing the generalized modus ponens as an extrapolation procedure highlights its ability to handle approximate reasoning by bridging gaps in knowledge through similarity.

Key Results

The paper presents a formal semantic model for fuzzy logic based on similarity. It demonstrates how:

  • Possibility and necessity distributions can be interpreted as measures of proximity between possible worlds.
  • The generalized modus ponens of Zadeh can be derived as a sound inferential procedure based on the transitive properties of similarity.
  • The model provides a unified framework that clarifies the relationship between fuzzy logic, modal logic, and metric spaces.

Limitations and Future Work

  • Scope: The paper focuses on propositional formulations for simplicity and does not delve into predicate logic.
  • Detailed derivations: The authors mention that a complete account of all relevant results and derivations is presented in a related technical note [8], implying this paper is a summary.
  • Future work: While not explicitly detailed, the framework suggests avenues for exploring more complex logical systems and applications where similarity plays a crucial role in reasoning. The authors' emphasis on the "natural extension" of modal logic hints at further integration possibilities.

Summary updated

BibTeX
@inproceedings{ruspini1990possibility,
  author = {Enrique H. Ruspini},
  title = {{Possibility as similarity; the semantics of fuzzy logic}},
  booktitle = {Conference on Uncertainty in Artificial Intelligence},
  year = {1990},
}
Figure