Once accumulated relationship state becomes consequential, the design problem exceeds memory retrieval and requires provenance, temporal validity, correction, authority, and portability.
Research Poster · The Persistent Primitive
The Persistent Primitive: Configuring Human-Governed Continuity for Persistent Human–AI Interaction
Micheal Charles Preble · Operant Dyad — Paper 2 · SSRN 7328783
Problem · Background · Research Question
User modeling, personal data stores, portability law, and temporal databases predate the current AI wave; the paper claims convergence, not invention.
What continuity-governance requirements follow once persistent relationship state is both persistent and consequential?
Framework: the human-governed continuity pipeline
- 01Canonical continuity state
The provider-independent representation of consequential relationship state.
- 02Governance / policy boundary
Determines authorized purpose, recipient, scope, temporal validity, and applicable constraints.
- 03Derived authorized view
Contains only the state necessary for the task — not the entire underlying store.
- 04Interoperability / transport
Mechanisms such as A2A carry the authorized view onward, after the governance decision, not instead of it.
- 05Receiving model or agent
Uses only the authorized derived view; the canonical store is never replicated indiscriminately.
Central diagram: requirements-to-landscape mapping
| Requirement | Representative precedent | Remaining integration issue |
|---|---|---|
| R1 Visibility / inspectability | Personal data stores; user-modeling transparency | Trace consequential generative effects to intelligible state |
| R2 Provenance / epistemic status | PAM Merkle-DAG provenance | Preserve declared / observed / imported / inferred distinctions across systems |
| R3 Temporal validity | Temporal databases; StateMem | Reliable current-state use in generative and agentic workflows |
| R4 Correction / downstream repair | Versioned data practices; dependency-guided rollback | Propagate correction to exported or externally held derivatives |
| R5–R6 Authority, consent, purpose, scope | Solid; MyData; GDPR; Stanford Portable Memory; PAM | Common policy layer tied to the person across heterogeneous AI consumers |
| R7 Portability / rehydration | GDPR portability; Solid; PAM; HCP | Semantic fidelity and usable continuity across different model behaviors |
| R8 Interoperability | A2A | Ensure transport operates on an authorized derived view, not raw continuity state |
| R9 Revocation / boundary control | Access-control traditions | Enforce downstream handling after information is copied, transformed, or observed |
| R10 Human capacity | Jiang (2026) | Preserve meaningful control without reproducing caregiving reversal |
Propositions / Observable Implications
- Substitutability: a continuity architecture centered on the person should survive replacement of any single model, provider, or application.
- Non-erasure correction: correcting outdated state should update what remains operative without necessarily destroying the historical record.
- Bounded governance: requiring approval of every memory operation defeats continuity’s value — meaningful control needs risk-sensitive defaults, not maximal manual intervention.
Evidence Required
Sumida et al. (2026, accepted ICMI): 24 participants, 10 sessions with a memory-augmented conversational agent; perceived memory shaped later enjoyment via self-disclosure. Jiang (2026, Design and Artificial Intelligence): 6-month autoethnography of a user-built relay system for a stateless AI, documenting labor and a “caregiving reversal.” Both are cited, external findings, not new data collected by this paper.
Limitations
Technical configuration cannot by itself establish legal authority. Does not resolve succession, incapacity, or death. Empirical evidence is thin relative to the breadth of the design problem. Several cited 2026 technical sources are preprints or prototypes, not mature deployed standards.
Falsifiers / Open Questions
“No reviewed source establishes a single mature, generally adopted configuration that jointly governs the full continuity problem around the natural person.” Whether the ten requirements are jointly sufficient, and not merely individually plausible, remains untested.
Citation / QR
Suggested citation
Preble, Micheal Charles. “The Persistent Primitive: Configuring Human-Governed Continuity for Persistent Human–AI Interaction” Available at SSRN 7328783, 2026.
Preble, Micheal Charles. “The Persistent Primitive: Configuring Human-Governed Continuity for Persistent Human–AI Interaction” Available at SSRN 7328783, 2026.