Reformulating AI Alignment as a Social Choice Problem
August 26, 2026
Alignment can be reformulated as linear optimization over a convex impact space by focusing on an algorithm's welfare consequences. This approach applies welfare economics and mechanism design to reconcile divergent human preferences, moving beyond the limitations of standard RLHF.
HOW THIS AFFECTS YOU
●
researcherYou can use tools from welfare economics to provide better social choice guarantees in alignment protocols.
●
policyThis framework helps in translating desired societal welfare constraints into technical alignment requirements.