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

Researchers introduced CoTrace, a framework that breaks tasks into verifiable requirements and traces AI influence across human-AI dialogue.

In 638 real collaboration logs, LLMs accounted for only 11–26% of direct goal-shaping contributions, but had much stronger impact on concrete requirements and indirect influence.

Simulations showed that interface design can materially change how much influence AI has, while a user study found that goal-level analysis shifted perceived contribution by nearly 2 points on a 5-point scale.

The work offers a way to measure hidden AI influence more fairly, helping assess reliance, credit, and fairness in collaborative AI tasks.