Introduction
Instead of searching for a single "perfect" voting rule that doesn't exist, this paper proposes "Voting by Axioms." In this approach, we decide on a case-by-case basis by identifying the most desirable set of normative principles (axioms) that force a unique outcome for a specific situation. It transforms social choice from a search for the best global algorithm into a search for the best local justification.
What Problem Does This Paper Address?
Social choice theory is haunted by "impossibility theorems"—most famously Arrow’s Theorem—which prove that no voting rule can satisfy all desirable properties (axioms) simultaneously. In practice, we usually ignore this by picking one "good enough" rule (like Plurality or Borda) and applying it to every election, even when that rule behaves poorly in specific scenarios.
The authors address the mismatch between theoretical perfection and pragmatic necessity. They ask: What if we didn't commit to one rule? Instead of applying a fixed rule to every preference profile, can we look at a specific set of votes and ask which outcome is most strongly supported by the principles we care about most?
Methods: Ranked Forcing
The authors introduce a framework called Ranked Forcing:
- Axioms as Constraints: An axiom (like "if everyone prefers A over B, B shouldn't win") is treated as a logical constraint. A set of axioms forces an outcome if every possible voting rule that satisfies those axioms would produce that specific outcome for the given votes.
- The Ranked Corpus: Since a single set of axioms might be too restrictive (forcing no outcome) or too weak (forcing many), the authors propose a ranked axiom corpus. This is a list of axiom sets ordered from "most desirable" to "least desirable."
- The Decision Procedure: For any given election, you start at the top of your ranked list. You check if the first set of axioms forces a unique outcome. If not, you move to the next set in the ranking, and so on, until an outcome is forced.
- SAT Solving: To make this practical, the authors encode these axioms into propositional logic. Determining if an outcome is "forced" is a search for inconsistency: if assuming "Outcome X did not happen" contradicts the axioms, then Outcome X is forced.
Key Insights and Ideas
- Instance-Based Justification: The focus shifts from the rule to the decision. We might not agree on one rule for all time, but we might agree that for this specific election, the most important principles point to Candidate A.
- Generalizing Tradition: "Voting by Axioms" generalizes classical social choice. If you put the axioms that uniquely define the Borda rule at the top of your ranking, the system simply behaves like the Borda rule.
- Intraprofile vs. Interprofile: The paper distinguishes between axioms that look at one election at a time (intraprofile, like Pareto efficiency) and those that compare different elections (interprofile, like Monotonicity). They prove that intraprofile axioms are particularly "well-behaved" in this framework, ensuring consistency across different scenarios.
- The "Cost" of Persuasion: The authors suggest that the ranking of axioms could be determined by the "cost" of persuading stakeholders to accept a principle, or by preferring smaller, simpler sets of axioms over complex ones.
Key Results
- Complexity (Theorem 1): The problem of deciding if an axiom set forces an outcome is coNP-complete. While technically "intractable" in the worst case, this is the "sweet spot" for modern SAT solvers, meaning the approach is practically feasible for small-to-medium-sized groups (like committees or boards).
- Consistency (Theorem 2): For "intraprofile" axioms, the "Voting by Axioms" procedure is guaranteed to behave like a consistent voting rule.
- Characterization (Theorem 3): They prove that if you rank a set of axioms that characterizes a specific rule high enough, the entire system will collapse into that rule, providing a bridge back to classical theory.
Limitations and Future Work
While SAT solvers are powerful, the logical encoding of axioms grows with the number of possible profiles. This approach is currently best suited for "high-stakes" decisions with a small number of alternatives and agents. Furthermore, deciding how to rank these principles is itself a difficult social choice problem. Future research is needed to develop faster algorithms and produce "human-readable" explanations (e.g., "Candidate A wins because we value Principle X more than Principle Y").
Summary updated