The review pulls together six human patterns of transient forgetting — arising from how we learn, retrieve memories, lapse spontaneously, overgeneralize, switch tasks, or multitask — and considers where their causes might overlap with mechanisms seen in neural networks. Thinking across these domains invites practical experiments: targeted training that alternates tasks, careful timing of tests, and measures of interference that track recovery. Those approaches would reveal whether human forgetting sometimes follows the same temporary, substantial drop-and-rebound pattern as the stability gap.
If a stability-gap-like effect exists in people, it could reshape strategies for teaching complex skills and for deploying AI systems that learn alongside humans. The proposed experiments point to concrete ways to probe these links between biological and artificial learning, and to discover how we might leverage transient forgetting to support growth, inclusion, and sustained performance rather than simply prevent loss.
Forgetting can be permanent or transient. An intriguing, recently described form of transient forgetting (TF) in artificial neural networks is the stability gap: a network that has already learned a first task exhibits transient yet substantial forgetting when it begins additional training on a second task alongside the first. Could an analogous phenomenon be found in humans? Following a literature survey, six types of TF in humans were identified, categorised by their sources: learning, retrieval, spontaneous, overgeneralisation, task-switching, and multitasking. Although all differ in important ways from the stability gap, it is hypothesised that some underlying mechanisms are shared. This opinion article concludes by proposing experiments to test whether a ‘stability gap-like’ form of TF also occurs in humans.
