Neural mechanisms of error-driven learning in retrieval practice: confidence gates memory and metamemory networks

Retrieval practice is one of the most effective strategies for enhancing long-term memory. Building on recent evidence, we investigated error-driven learning (EDL) as a unified mechanistic account, which posits that memory updating is shaped by two key signals: objective error-quantified as retrieval inaccuracy-and subjective error estimates, reflected in confidence judgements, which regulate how objective errors are processed. Using a continuous colour-wheel paradigm with functional magnetic resonance imaging, this study examined the neural substrates of EDL and the modulatory role of confidence. Behaviourally, both larger objective error and lower confidence independently predicted greater subsequent memory improvement. Neurally, analyses of univariate activation, functional connectivity and multivoxel pattern similarity revealed two dissociable yet interacting systems gated by confidence. Memory-related circuits, such as functional connectivity between the hippocampus and inferior frontal gyrus, were preferentially engaged under low confidence, leveraging objective retrieval errors to drive memory updating. In contrast, metamemory-related regions-including the striatum, thalamus and supramarginal gyrus-were recruited under high confidence to process confidence violations elicited by unexpected retrieval failures. Together, these findings demonstrate how confidence dynamically regulates the recruitment of distinct neural circuits in support of retrieval-based learning, offering broad implications for educational interventions and memory rehabilitation. This article is part of the theme issue 'The role of hippocampal predictions in cognition: bridging perception and memory'.