AI role in LLM-assisted writing: collaborator/advisor versus tool

Investigate design approaches to transform large language model-powered writing systems (such as GhostWriter) to enable the AI to act as a collaborator or advisor rather than solely as a tool, and determine whether most end-users prefer or require such collaborative roles or are instead satisfied with AI operating strictly as a tool.

Background

The paper studies GhostWriter, an LLM-powered writing design probe that aims to enhance personalization and user agency. Through two user-paper sessions with 18 participants, the authors observed diverse perceptions of the AI’s role, with participants variably describing it as a tool, collaborator, or advisor.

Synthesizing these observations, the authors highlight a key unresolved direction: whether and how to design LLM-assisted writing systems to support AI in more collaborative or advisory roles, or whether users are adequately served by AI functioning primarily as a tool. This motivates a focused inquiry into concrete design transformations that enable these roles and into user preferences regarding them.