Resource Rational Contractualism Should Guide AI Alignment

Researchers propose Resource-Rational Contractualism (RRC), a new framework for AI alignment that enables AI systems to make decisions affecting diverse stakeholders through efficient approximations of rational agreements. The approach uses normatively-grounded heuristics to balance computational effort with accuracy in navigating complex human social environments.

• →Resource-Rational Contractualism offers a practical solution to the costly and slow process of securing stakeholder agreements for AI decision-making at scale.
• →The framework employs cognitively-inspired heuristics that trade computational effort for accuracy in approximating rational agreements.
• →RRC-aligned AI agents would dynamically adapt to and interpret changing human social contexts more effectively.
• →The approach addresses the challenge of AI systems making decisions that affect people and other AI agents with divergent goals and values.
• →The framework grounds AI decisions in agreements that diverse stakeholders would endorse under appropriate conditions.