The rise of machine learning and generative AI has brought challenges, particularly around how individuals perceive and interact with these technologies. People often misalign their reliance on automation—either overtrusting poor systems or distrusting reliable ones—leading to overreliance or underuse. Automation biases also lead individuals to follow AI without checking its accuracy and high performance. Trust in automation is hard to measure, and neural correlates are emerging as promising tools. Event-related potentials (ERPs) like error-related negativity (ERN) and error positivity (Pe) reflect how people respond to errors, both their own and those of others. Participants collaborated (collected points together) or competed with an algorithm in an experiment using a modified Eriksen Flanker task. We measured the neural responses (oERN and oPe) of participants while performing the task and observing the algorithm performing the same task. During the algorithm monitoring, ERPs revealed that participants tracked algorithm errors in both groups. However, performance profiles differed, with shorter reaction times and fewer errors during the performance competition relative to the collaboration condition. Our results suggest that participants attended to the algorithm performance but performed differently depending on the social context. We conclude that collaborating and competing with artificial intelligence elicit brain activity and behavioral performance similar to those of interacting with humans.