Learning From Our Mistakes: Opportunities for Self-Correction Following Fast Errors Enhance Adaptive Fact Learning

Adaptive fact-learning systems optimize retention by tailoring review schedules to learner behaviour, yet most models consider only accuracy or the latency of correct responses, ignoring error timing. Decision-making research and dual-process theories suggest errors arise from distinct mechanisms: impulsiveness produces fast errors, whereas uncertainty yields slow errors, with corresponding differences in internal error monitoring. This implies error speed may differentially influence subsequent learning. In a spaced-repetition recognition task, we found differences in subsequent performance between fast and slow errors. Building on this, we investigated whether fast errors can be self-corrected by prompting participants to reconsider responses, using slow errors as a benchmark. Fast-latency errors were corrected significantly more often than slow-latency errors following prompting (44% vs. 24%). Accurate self-correction of fast errors improved subsequent item-level accuracy, whereas prompting slow-latency responses did not. These findings show fast and slow errors have distinct learning consequences and that latency-tailored feedback can enhance learning.