Learning How to Vote with Principles: Axiomatic Insights Into the Collective Decisions of Neural Networks
This work addresses the need for transparent collective decisions in AI and voting theory, offering a mathematically rigorous approach to study bias and value-alignment, though it is incremental in combining these fields.
The paper tackled the problem of applying neural networks to voting theory while ensuring transparency, finding that neural networks often fail to align with core voting axioms despite high accuracy, and that optimizing for axiom satisfaction can synthesize new voting rules that surpass existing ones.
Can neural networks be applied in voting theory, while satisfying the need for transparency in collective decisions? We propose axiomatic deep voting: a framework to build and evaluate neural networks that aggregate preferences, using the well-established axiomatic method of voting theory. Our findings are: (1) Neural networks, despite being highly accurate, often fail to align with the core axioms of voting rules, revealing a disconnect between mimicking outcomes and reasoning. (2) Training with axiom-specific data does not enhance alignment with those axioms. (3) By solely optimizing axiom satisfaction, neural networks can synthesize new voting rules that often surpass and substantially differ from existing ones. This offers insights for both fields: For AI, important concepts like bias and value-alignment are studied in a mathematically rigorous way; for voting theory, new areas of the space of voting rules are explored.