"I Didn't Make the Micro Decisions": Measuring, Inducing, and Exposing Goal-Level AI Contributions in Collaboration
arXiv:2605.21363v2 Announce Type: replace Abstract: As large language models (LLMs) increasingly shape how users form, refine, and extend their goals, attributing contributions in human-AI collaboration becomes critical for users calibrating their own reliance and for evaluators assessing AI-assisted work. Yet existing methods focus on final artifacts, missing the process through which goals themselves are jointly shaped. We introduce a goal-level attribution framework, CoTrace, that decomposes explicit goals into verifiable requirements and traces both direct contributions and indirect influe
The proliferation of advanced LLMs in collaborative settings necessitates new methods for understanding and attributing contributions, especially as AI's role becomes more nuanced than simple artifact generation.
Accurate attribution in human-AI collaboration is crucial for calibrating user reliance on AI, establishing accountability in AI-assisted work, and ensuring fair evaluation of both human and AI contributions.
The introduction of frameworks like CoTrace shifts the focus from merely evaluating final outputs to analyzing the process of goal formation and refinement, offering deeper insights into human-AI interaction.
- · AI accountability researchers
- · Human-computer interaction designers
- · AI ethics and governance bodies
- · Users of collaborative AI tools
- · Over-simplified AI evaluation methods
- · Organizations with opaque AI-human workflows
Improved methodologies for assessing the quality and impact of AI in collaborative problem-solving.
Increased trust and more effective integration of AI into complex, multi-stakeholder projects as accountability becomes clearer.
Potential for new regulatory frameworks around AI-assisted intellectual property and responsibility in cases of error or harmful output.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL