Sharing control with a machine impairs human response inhibition

To fulfill goals, humans make use of cognitive control, which is a suite of processes toplan and manage thoughts and actions. One such process is response inhibition, which entailsstopping a response as goals change. Traditionally, response inhibition is measured inexperiments in which humans have unilateral responsibility for stopping the action. However, inthe real world, humans are increasingly sharing control with artificial intelligence (AI), with theparadigmatic case being partially automated cars. We designed an experiment that models thesharing of control between humans and AI in partially automated cars and found that whenhumans (N=40) share control with an automated system that usually stops, human responseinhibition is impaired when the automated system fails. This reveals a specific cost of sharingcontrol to human response inhibition, suggesting that the benefits of partial automation shouldbe weighed against the costs of impaired human cognitive control.