Memory, Forgetting, and the Right to Revision: Artificial Intelligence and the Continuing Authority of the Past

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Accurate records protect accountability, but they can also govern people after the conditions they describe have changed. AI systems can retrieve, combine, infer from, and repeatedly circulate historical representations faster than evidence of later change can follow. A person may change while the informational version through which institutions judge that person remains fixed. This paper asks how institutions can preserve truthful history without giving it automatic and indefinite authority over present decisions. It distinguishes historical accuracy from continuing applicability, record correction from representational revision, and availability from actionability. It introduces revision-sensitive memory: an information environment’s capacity to preserve warranted historical continuity while allowing sufficiently supported later developments to alter an earlier representation’s status, interpretation, weight, or use for a specified purpose. Its corresponding failure, representational foreclosure, occurs when the past remains practically authoritative while relevant evidence of change cannot become comparably consequential. The paper argues for a qualified right to revision in consequential institutional settings. This guarantees neither deletion nor a favorable outcome. It requires a meaningful opportunity for warranted evidence of change to enter present judgment and for institutions to justify the authority they assign to the past. Current AI design choices can deepen or reduce the problem.