Automated design of voting rules by learning from examples

While impossibility results have established that no perfect voting rules exist, efficiently designing a voting rule that satisfies at least a given subset of desiderata remains a difficult task. We argue that such custom-built voting rules can be constructed by learning from examples. Specifically, we consider the learnability of the broad, concisely-representable class of scoring rules. Our main result asserts that this class is efficiently learnable in the PAC model. We also discuss the limitations of our approach, and (along the way) we establish a lemma of independent interest regarding the number of distinct scoring rules.