Who can re-enter the loop? A scoping review on cognitive abilities and task re-engagement in human-AI interaction

Rapid advances in artificial intelligence (AI) promise a near future in which humans and AI systems work together to solve complex problems. Although offloading tasks to AI reduces immediate cognitive demands, the system may later require human input or decision-making, at which point the human must “re-enter the loop” by redirecting attention, reconstructing the task state, and responding appropriately. Whether individual differences in cognitive ability predict the capacity to re-enter the loop remains unclear. Here, we conducted a scoping review of studies relating cognitive ability to loop re-entry performance. Thirteen studies met the inclusion criteria. Among them, most studies examined conditionally automated driving, and none examined interactions with generative AI systems. Working memory capacity was the most frequently studied ability and was generally associated with better loop re-entry performance, whereas results for attentional control were mixed. The evidence was further limited by small samples, single-task measures of cognitive abilities, and inconsistent outcome measures. As human oversight becomes increasingly central to AI safety, understanding who can re-enter the loop effectively remains an important open question. We offer concrete recommendations to guide future research.