Cognitive Architectures of Normative Alignment: Value-Driven Inference versus Discrete Mathematical Computation

Opening paragraph of the published conference abstract

Norms embody constrained cognition of practice. Agents are the principal vehicle for AI deployment. Whereas individual alignment focuses on reasoning about an agent’s internal decision-making, multi-agent alignment entails reasoning about system-level norms, the established mechanism for governing collective behavior across organizations, communities, or simple aggregates. This study contrasts human-value-driven alignment with linear algebraic derivation. Multi-agent experiments are conducted to discern their comparative advantages.