Abstract
Persistent digital systems increasingly treat identity as a continuity problem: the more faithfully a system can retain our preferences, statements, relationships, and prior decisions, the more useful it appears. This essay argues that continuity is only one condition of a humane identity system. People also need room to contradict a prior self, reinterpret experience, and become less predictable to the infrastructures around them.
The right to an unstable self is not a right to rewrite public facts or escape responsibility. It is a narrower claim about proportionality. Records should remain durable where other people depend on them, while personal models should preserve the possibility that an old pattern is no longer an active identity. Stability can support agency; compulsory stability can quietly replace it.
Identity is not a snapshot
A photograph, profile, or model can be accurate and still be wrong about a person. Accuracy describes a relationship to a moment; identity stretches across moments that do not always agree. Human lives contain changed beliefs, abandoned habits, repaired relationships, and memories whose meaning alters even when the underlying event does not. A system that collapses those states into one durable profile mistakes consistency for truth.
Research on autobiographical memory describes recollection as constructive and responsive to present goals. Philosophical accounts of narrative identity likewise treat the self as open to revision rather than fixed like an object. Neither tradition implies that facts are optional. They suggest something more modest and more demanding: a person cannot be represented adequately by preserving facts alone. The relation among those facts must remain capable of change.
When continuity becomes pressure
Continuity pressure appears when a system repeatedly returns a person to its prior model of them. A recommender keeps serving a preference after interest has faded. A workplace assistant carries an old role into a new one. A health interface treats a resolved crisis as the dominant context. None of these actions needs to be dramatic to narrow the future. Repetition can make a prediction feel like a verdict.
The burden is unevenly distributed. People with stigmatized histories, unstable housing, changing family structures, or contested public identities have more to lose when old context becomes ambient. The most privileged users may experience persistence as convenience; others experience it as the inability to leave a room whose walls have become data.
The same retained context can move from useful continuity to compulsory identity as its authority and duration increase.
Past context is retrieved for a specific task.
The interface proposes context that can be ignored.
Old context begins shaping choices before review.
The retained model determines eligibility, trust, or access.
The relevant variable is not retention alone, but the authority granted to retained context.
Responsibility without fixity
The strongest objection is that instability can become a polite name for evasion. Shared life depends on promises, evidence, and the ability to say that an action occurred even when its author has changed. A right to revision cannot grant private control over another person's memory or dissolve records needed for safety, justice, or historical accountability.
That objection sets a boundary rather than defeating the argument. Public records and personal models perform different functions. A court record may need integrity; a shopping profile does not need to preserve every past preference. A survivor's account cannot be deleted because another person wishes to move on; an assistant's inference that someone is still grieving should not become permanent. Humane design starts by distinguishing obligations to others from predictions about the self.
Designing for revision
An identity system should make time visible. It should show when an inference was formed, what evidence sustains it, whether it is still active, and how strongly it shapes current behavior. Users should be able to expire a preference, separate contexts, annotate a record, or begin a new profile state without deleting the historical fact that a previous state existed.
The system should also notice discontinuity without pathologizing it. A sharp change can indicate fraud or harm, but it can also indicate recovery, learning, migration, conversion, aging, or ordinary surprise. Escalation should be proportional to consequence. The closer a model comes to determining access or credibility, the stronger the requirements for provenance, review, and human appeal.
Limitations and open questions
This essay offers a normative framework, not a universal retention schedule. Different domains carry different duties, and identity is shaped by culture, law, disability, and social position in ways no single interface model captures. More empirical work is needed on how people experience persistent personalization over long periods and how controls can support change without creating new opportunities for coercion.
The right to an unstable self should therefore be treated as a design question with a standing dissent: who else depends on the continuity being loosened? The answer will sometimes justify durable memory. But systems should have to make that case, rather than assuming that maximum continuity is the neutral default.
References
- Update on Emotion and Autobiographical Memory. Elizabeth A. Kensinger and colleagues. Affective Science, 2025.
- An Update on Memory Reconsolidation Updating. Jonathan L. C. Lee, Karim Nader, and Daniela Schiller. Trends in Cognitive Sciences, 2017.
- Paul Ricoeur. Bernard Dauenhauer and David Pellauer. Stanford Encyclopedia of Philosophy, 2024.
Suggested citation
Mara Voss. “The Right to an Unstable Self.” Drifting Reality Institute, 2026. DRI-ES-2026-03, version 1.0.
Revision history
First published.
Expanded with design requirements and limitations.