KurisutinaResearch date: 2026-09-22.
Citation and status. Maria Wimber, Arjen Alink, Ian Charest, Nikolaus Kriegeskorte, and Michael C. Anderson. Retrieval induces adaptive forgetting of competing memories via cortical pattern suppression. Peer-reviewed, Nature Neuroscience 18, 582–589; published 2015-03-16. Article DOI; public accepted manuscript.
Acquisition and reading. Saved PDF, readable text, and provenance with SHA-256 hashes. Read the entire main article, Online Methods, figure legends, and references. Separate supplementary figures, tables, and checklist were not obtained or read; their descriptions in the main article were read. The repository supplies an accepted manuscript, with no open redistribution license identified.
Source findings and limits (factual paraphrase). Twenty-four adults learned two pictures per word cue. Repeated selective retrieval of one picture reduced later recognition of its competitor: 75.2% versus 82.1% for unpractised baseline items. Across four retrievals, competitor-specific visual-cortex pattern similarity declined below a baseline estimated from other pictures; this suppression related to subsequent forgetting. Target reinstatement increased. Recognition used pictures without the original word cues, weakening a simple explanation limited to blocking at those cues. The imaging readout matched known pictures against separately measured templates; it did not extract arbitrary memory contents. Prefrontal recruitment correlated with suppression, but the study did not causally perturb the proposed inhibitory mechanism. Targets and competitors had unequal initial training, and participants were instructed against integrating the paired pictures. Most directional tests were one-tailed. These results support competition-dependent changes in later accessibility; they do not establish permanent deletion, recovery failure under every cue, autobiographical generality, or an independently validated individual forecasting model.
Correction checked. The 2018 author correction adds that unanswered recognition trials within 3.5 seconds counted as errors. Forgetting remains significant when those trials are excluded instead. The original article was not updated. Saved and read the Wimber correction in full: PDF, text. The publisher's PDF contains three corrections; its hosting URL names a different correction on that shared page.
Interpretation for functional transfer — our reasoning. Accurate stored records and human-like access are distinct objectives. A system could preserve every episode accurately yet predict a person's next recollection poorly because it ignores which episodes have recently been retrieved and which compete. Conversely, a retrieval failure does not justify deleting the underlying record. An immutable evidence archive plus a changing access model is a useful engineering hypothesis, not an architecture demonstrated by this experiment.
The critical uncertainty is what changed: stored discriminative information, context-dependent access, attention during retrieval, or some combination. A known-template fMRI correspondence is evidence about the process, not a complete measurement of content. A functional replica should therefore keep “unavailable in this test” separate from “absent from storage.” Whether altered access is reversible needs distinct tests, not a narrative explanation attached to each lapse.
Falsifiable experiment — proposal, not an existing result. Give consenting participants controlled episode pairs with a shared cue. Randomize selective practice across pairs and retain untouched pairs. Reserve different item subsets for free recall, recognition, alternate-cue rescue, and delayed tests so that one test does not silently train the next. Give every model the same episode records and practice history. Compare fixed retrieval with an access model updated by retrieval history; retain the same underlying records in both. Require advance probabilities for target recall, competitor intrusion, omission, and recognition errors, scored on held-out people and item families. A gain confined to reproducing the practised items would not establish transfer. A predictive benefit across alternate cues and delays would justify modelling access dynamics; it would still not identify a biological inhibition mechanism or prove irreversible content loss.
Connection to extraction. Asking repeatedly about an episode is a potential intervention on later access, not merely another independent sample. Extraction protocols should record retrieval history and assess what ceases to surface, alongside what is successfully recovered.