Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
A decision maker (DM) uses an AI agent to estimate an unknown state, for which both possess informative private signals. Conditional on the state and the DM's final assessment, he prefers the AI's recommendations to be incorrect, thus affirming his own superiority or sharing the blame. Our analysis indicates that the correctness of the process is not a monotone function of participants' expertise levels: (i) a less accurate AI may lead to improved outcomes by reducing the DM's reliance on it, and (ii) a less accurate DM can enhance information aggregation leading to a superior result.