Research Brief · The Persistent Primitive
When AI systems remember you, what should stay centered on you?
Micheal Charles Preble · Operant Dyad — Paper 2
The Question
Once accumulated human–AI relationship state becomes useful or consequential, what governance requirements follow — and around what enduring reference point should a continuity architecture be organized, given that models, providers, and applications can all change?
The Problem
AI systems are becoming better at retaining information about the people who use them, usually framed as a personalization problem: can a system remember preferences and avoid forcing the user to restate information? That framing does not fully describe what happens when interaction continues for weeks, months, or years and produces accumulated relationship state — corrections, decisions, constraints, and working history that alter later interaction. Once that state becomes consequential, memory is the wrong vocabulary: the real questions are about continuity and governance.
The Contribution
Derives ten continuity-governance requirements (R1–R10) from four premises — persistence, consequentiality, heterogeneity, and a human-governance objective: visibility and inspectability; provenance and epistemic status; temporal validity and supersession; correction without historical erasure; human authority and consent; purpose and scope limitation; portability and rehydration; interoperability without governance substitution; revocation and boundary control; and protection of human capacity. It argues the natural person, not the provider account or the cryptographic key, should be the primary design locus for the living, capacitated case — because the surrounding technical components remain substitutable while the person does not.
How the Argument Works
A historically grounded convergence analysis (the “shelf test”): for each of the ten requirements, does credible prior or contemporary work already demonstrate the mechanism is technically or institutionally plausible? User-modeling research, personal-data-store architectures (Solid, MyData), GDPR Article 20 portability, and temporal-database research each predate the current AI wave; 2026-era systems (Stanford's Portable Memory / Human Context Protocol, Portable Agent Memory, the A2A protocol) extend those older traditions. The paper proposes a five-stage human-governed continuity pipeline: canonical continuity state → governance/policy boundary → derived authorized view → interoperability/transport → receiving model or agent, worked through a travel-planning example.
Why It Matters
Two longitudinal, external studies motivate the framework without being its data: Sumida et al. (2026) found perceived memory shaped later self-disclosure and enjoyment across 24 participants and 10 sessions; Jiang (2026) documented a 6-month autoethnography in which maintaining a stateless AI's coherence required sustained human labor — a “caregiving reversal” that directly motivates the requirement that governance mechanisms not transfer excessive coordination cost back to the person.
Evidence Status
Evidence status: A relationship-derived requirements framework, a historically grounded convergence analysis, and a minimum configuration for human-governed continuity — not a claim to invention of the underlying primitives, and not an empirical validation that the ten requirements are jointly sufficient in practice.
What This Does Not Claim
- Does not claim to invent portable memory, provenance, scoped disclosure, temporal correction, user-controlled data, or agent interoperability.
- Does not claim the ten requirements are mathematically necessary, universally sufficient, or already standardized.
- Does not claim the natural person remains the sole operative center across incapacity or death, or exclusive ownership of every record associated with a person — those are explicitly out of scope.
- Does not claim empirical prevalence: the cited longitudinal studies establish that prior interaction can alter later conditions, not how often or how severely.
Open Questions
Incapacity, death, third-party rights, and fiduciary administration “require institutional and legal analysis beyond the scope of this paper” — the next paper in the series (Authority Does Not Travel) begins exactly where this one stops. No single reviewed source establishes a mature, generally adopted configuration that jointly governs the full continuity problem.
Read / Cite the Research
Read the canonical paper record → · Full paper PDF · View on SSRN
Preble, Micheal Charles. “The Persistent Primitive: Configuring Human-Governed Continuity for Persistent Human–AI Interaction” Available at SSRN 7328783, 2026.