Introduction
In the field of Artificial Intelligence, knowledge bases (KBs) frequently encounter logical contradictions, particularly when integrating information from multiple, potentially conflicting sources. While significant research has focused on global measures to quantify the overall inconsistency of a KB, there has been a notable lack of rigorous local measures. Local measures are essential for identifying the specific "culpability" of individual formulas within a set. Current local metrics often rely on heuristic counting—such as membership in minimal inconsistent sets—which can fail basic logical "sanity checks." This paper introduces a formal, principle-based framework to move beyond ad-hoc metrics toward logically sound assessments of formula contribution to conflict.
Methodology and Rationality Postulates
The authors adopt a postulate-based approach to evaluate the rationality of inconsistency measures. They define 13 desiderata (rationality postulates) that any robust local measure should satisfy. Key postulates include:
- Monotonicity: The degree of conflict attributed to a formula should not decrease as more information is added to the KB.
- Maximality: A formula that is inherently self-contradictory (e.g., $p \wedge \neg p$) must always receive the maximum possible inconsistency value.
- Free Formula Independence: The presence of "innocent" formulas (those not involved in any conflict) should not affect the inconsistency values of other formulas.
To address the shortcomings of existing metrics, the authors propose several new measures, including:
- Problematic-based ($I_{\text{prob}}$): Quantifies how many other formulas are "freed" from conflict upon the removal of the target formula.
- Conflict-based ($I_\wp$): A novel measure that evaluates a formula against the entire powerset of the KB. It calculates the ratio of consistent subsets that become inconsistent when the specific formula is introduced.
As a theoretical study in formal logic, the framework is validated using synthetic, illustrative examples of propositional logic formulas to provide counter-examples and proofs for postulate satisfaction.
Key Results
The primary contribution of the research is a comprehensive comparative analysis of nine different measures against the 13 proposed postulates. As shown in Table 1, existing measures—including the widely used Shapley Inconsistency Value ($S^I_K$) and simple conflict counting ($I_\#$)—fail to satisfy between 3 and 7 postulates each. Notably, these popular metrics often violate Maximality and Dominance, potentially ranking a minor conflict higher than a direct self-contradiction.
The authors prove that the new $I_\wp$ measure is the only one to achieve a "perfect score," satisfying every single postulate. This demonstrates that it is mathematically possible to construct a local measure that is entirely consistent with the proposed logical framework.
Conclusion and Future Directions
While $I_\wp$ provides a theoretically ideal measure of local inconsistency, it faces practical challenges regarding computational complexity. Because it requires evaluating the powerset of the KB, its calculation is exponential ($2^n$) relative to the number of formulas, making it NP-hard.
Future research will focus on identifying "characterization results"—the minimal set of postulates required to uniquely define a measure—and exploring practical applications. These include belief revision, where local measures can guide which information to discard to restore consistency, and deontic logic, for reasoning about conflicting obligations.
Summary updated