Social Alignment

Social alignment concerns AI’s effects on society beyond its operator. Korinek and Balwit distinguish it from Direct Alignment, linking externalities to governance. See Aligned with Whom? Direct and Social Goals for AI Systems and Society-in-the-Loop. ^[1]

The Challenge of Value Alignment: from Fairer Algorithms to AI Safety connects technical safety with plural social values. Using the Veil of Ignorance to align AI systems with principles of justice tests one bounded procedure for choosing principles; Multi-Agent Risks from Advanced AI asks how interactions create harms beyond individual alignment.

Dynamic value alignment through preference aggregation of multiple objectives tests a bounded approach to combining changing preferences in control decisions.

Distributional objectives

Collective Good and Optimization in Socioeconomic Systems makes average and worst-case predictive loss explicit competing objectives. Artificial Intelligence and Inequality: Challenges and Opportunities supplies a broader narrative of possible harms; it does not quantify them independently.

Institutional perspectives

Learning Designers and AI examines perceived values in a professional setting. Longtermist Political Philosophy asks how future people should count in institutional decisions. These raise different questions; survey preferences do not settle normative legitimacy.

Institutions and identities

Surfing the AI Waves offers an overlapping historical framework for organizational adoption. Who’s in the Mirror proposes an identity-based perspective; that entry is restricted to its accessible abstract. Neither establishes a general causal effect of AI on organizations.