Working Paper
In this paper, the authors study the impact of algorithmic advice on team decisions through two lab experiments. In Study 1, they vary whether and when advice is provided to a team of that is engaged with a prediction task. The authors find that decision quality (e.g., team performance) is higher when advice is provided earlier, since it is more readily adopted by the team. The authors also investigate how algorithmic advice impacts team interactions; they find that early advice leads to a reduction in the teams’ efforts to jointly make sense of the problem context and develop mental models linking predictors and outcomes. This points to a downside of early algorithmic advice.
In Study 2, in addition to replicating these results, the authors investigate the impact of algorithmic quality and its interactions with timing. The authors show that when algorithmic quality is low, the greater reliance on early advice, combined with the reduction in effort for sense-making, diminishes team performance. These findings emphasize the importance of when and how algorithms are introduced into team processes, revealing a fundamental trade-off in human–AI collaboration: early algorithmic advice can improve immediate performance but undermine collective reasoning, whereas delaying advice promotes analytical depth and shared understanding that support long-term team expertise development.
