Scopes of Alignment

TL;DR

This paper argues for expanding AI alignment beyond generic values by introducing three dimensions—competence, transience, and audience—to better tailor models to specific purposes and contexts.

Contribution

It proposes a novel framework with three dimensions for AI alignment, moving beyond the traditional focus on helpfulness, harmlessness, and honesty.

Findings

Introduces three alignment dimensions: competence, transience, audience.

Positions existing and future AI technologies within the new framework.

Highlights the need for context-specific and purpose-driven alignment approaches.

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.