Scopes of Alignment

TL;DR

The paper argues that AI alignment has mostly targeted mass semantic alignment to generic values, neglecting context, temporal variation, and audience diversity. It introduces a three-scope framework—competence ($KSB$: knowledge, skills, behaviors), transience (semantic vs episodic), and audience (dyadic to mass)—to tailor alignment to different use cases and cultures, illustrated with a motivating example. Key contributions include formalizing scope-based alignment, clarifying data and algorithmic implications (e.g., adapter-based episodic alignment and memory-informed semantic alignment), and outlining pathways toward pluralistic alignment via reflective equilibrium. This framework enables context-aware, culture-sensitive, and audience-specific alignment workflows, which can improve safety, usefulness, and governance of frontier models in real-world deployments.

Abstract

Much of the research focus on AI alignment seeks to align large language models and other foundation models to the context-less and generic values of helpfulness, harmlessness, and honesty. Frontier model providers also strive to align their models with these values. In this paper, we motivate why we need to move beyond such a limited conception and propose three dimensions for doing so. The first scope of alignment is competence: knowledge, skills, or behaviors the model must possess to be useful for its intended purpose. The second scope of alignment is transience: either semantic or episodic depending on the context of use. The third scope of alignment is audience: either mass, public, small-group, or dyadic. At the end of the paper, we use the proposed framework to position some technologies and workflows that go beyond prevailing notions of alignment.

Figures (2)

• Figure 1: Typical application development lifecycle for LLMs considered by frontier model providers.
• Figure 2: Different audiences of alignment.