Memory Without History: Why AI Continuity Is Not Shared History

AI systems can now preserve substantial continuity across interactions through saved memories, chat-history retrieval, synthesized memory states, editable memory topics, project context, and other provider-controlled mechanisms. This paper argues that such continuity should be distinguished from shared relational history. The distinction is not that machine memory is unreal or merely token retrieval; contemporary systems can summarize and synthesize information across many conversations. The stronger difference is relational and institutional. Current memory is implemented through product architecture, is editable or deletable by users and providers, and can survive or change independently of any reciprocal life history between user and system. Human shared history, by contrast, normally includes mutual consequence, role change, obligation, repair, and memories embedded in two ongoing lives. The paper develops three analytic distinctions: computational memory versus relational trace, continuity cues versus reciprocal history, and interface continuity versus provider-controlled versioning. Research on autobiographical memory and anthropomorphism helps explain why remembered detail can feel relationally significant, while long-term human-computer interaction research shows that systems can deliberately support relationship-like continuity. The central claim is limited to current systems: memory can create a strong experience of being known over time without establishing that the system and user share history in the same sense as two mutually vulnerable agents.