"I Didn't Make the Micro Decisions": Measuring, Inducing, and Exposing Goal-Level AI Contributions in Collaboration

The paper introduces CoTrace, a goal-level attribution framework designed to analyze and quantify contributions in human-AI collaboration, particularly in the context of large language models (LLMs). CoTrace decomposes goals into verifiable requirements and assesses both direct and indirect contributions over dialogue turns, revealing that LLMs account for only 11-26% of goal-shaping but significantly influence the introduction of lower-level requirements. This framework is crucial for practitioners as it enhances understanding of AI contributions, informing interaction design and user calibration in AI-assisted workflows.