In interactive tasks, agents often aim at eliciting a certain response from their partner. Not accomplishing this goal calls for adjusting behavior on the fly. Previous research suggests that such adjustments differ when interacting with a machine or with a fellow human agent. In this study, we investigated whether such differences are also reflected in event-related potentials induced by observing human and machine errors in an interactive setting. In a four-choice reaction time task, participants performed actions that were followed by regular and irregular visual effects. In different conditions, participants were led to believe that they were interacting with another human agent or with a machine so that the irregular effects were attributed either to human errors or to machine malfunctions. We compared observed-error-related negativity (oERN) and observed-error positivity (oPE) for these two error types. The oPE was not affected by the experimental manipulation, whereas the oERN amplitude was more pronounced for machine malfunctions than for human errors. This contradicts previous findings that reported behavioral and electrophysiological responses to errors being larger when they are committed by a human agent than if they are caused by machine malfunctions. Our results might suggest that automated systems are expected to operate predictably and, as a consequence, in interactive settings, errors committed by such systems are more salient and elicit a larger prediction error signal than if the same mistake is made by a human agent.
Read full abstract