Preface to the Speical Issue on Reasoning about Preferences, Uncertainty, and Vagueness

Thomas Lukasiewicz, Rafael Peñaloza, Anni-Yasmin TurhanFLAP 2017 · OpenReview

This special issue of the IFCoLog Journal of Logics and Their Applications (Volume 4, Issue 7) explores the intersection of three critical domains in artificial intelligence: reasoning about preferences, uncertainty, and vagueness. As AI systems move from controlled environments to the "messy" real world, the ability to handle subjective choices, probabilistic outcomes, and fuzzy concepts becomes paramount.

Cover of the IFCoLog Journal of Logics and Their Applications, Volume 4, Issue 7.

Addressing the "Brittleness" of Classical Logic

Classical logic is often too rigid for practical applications, operating on binary truths. In contrast, real-world data is frequently:

  • Vague: Concepts like "affordable" or "nearby" lack sharp boundaries.
  • Uncertain: Information is often incomplete or probabilistic.
  • Preference-based: Decisions depend on subjective user rankings.

The contributions in this issue aim to integrate these "soft" concepts into formal logical frameworks without compromising mathematical rigor or computational efficiency.

Methodological Breakthroughs

The issue highlights several advancements in formalisms and algorithms:

  • Multi-valued SAT Solving: David Mitchell adapts Conflict-Driven Clause Learning (CDCL) to "Signed CNF," proving that "restarts" allow these solvers to efficiently simulate general resolution proofs.
  • Fuzzy Stable Models: Lee and Wang extend Stable Model Semantics to fuzzy logic using a simplified "reduct-based" definition, bridging the gap with Fuzzy Equilibrium Logic.
  • Weighted Description Logics: Acar et al. combine Description Logics with Multi-Attribute Utility Theory, implementing the uDecide Protégé plugin for ranking choices based on weighted attributes.
  • Modular Preference Systems: Ensan and Ternovska propose a framework to combine disparate preference models (e.g., CP-nets and Answer Set Optimization) across different logical languages.
  • Maximum Entropy (ME): Beierle et al. apply the Principle of Maximum Entropy to complete knowledge in probabilistic systems in an unbiased manner.

Practical Applications and Scalability

Theoretical advancements are validated through practical case studies:

  • Semantic Web Scalability: Using a subset of DBpedia with over 300,000 axioms, Acar et al. demonstrated that their preference-ranking plugin scales linearly, processing large datasets in seconds.
  • Medical Diagnosis: Beierle et al. tested their MEcore system on clinical brain tumor data. The probabilistic results were found to be highly plausible by neurosurgeons, suggesting that Maximum Entropy reasoning can effectively mirror expert human intuition.

Key Insights and Results

A significant finding by Mitchell is that "restarts" in solvers are not merely a heuristic but are mathematically essential for power. In terms of complexity, Wilson et al. identified "tractability islands," showing that preference inference becomes polynomial-time when importance orderings are simple total orders. Furthermore, Lee and Wang’s proof of semantic equivalence simplifies the implementation of fuzzy logic systems.

Future Directions

While promising, challenges remain. Future research will focus on balancing more expressive languages (including arithmetic functions) with computational speed and moving from "point" probabilities to "interval" probabilities to better capture true uncertainty. This collection underscores the ongoing evolution of logic-based AI toward systems that can reason with the nuance and flexibility required for human-centric applications.

Summary updated

BibTeX
@inproceedings{lukasiewicz2017preface,
  author = {Thomas Lukasiewicz and Rafael Peñaloza and Anni-Yasmin Turhan},
  title = {{Preface to the Speical Issue on Reasoning about Preferences, Uncertainty, and Vagueness}},
  booktitle = {FLAP},
  year = {2017},
  url = {https://openreview.net/forum?id=nltXmg6pLEM},
}
Figure