Research
What persists, and under whose authority?
Repeated interaction can make an AI system more useful because earlier corrections, constraints, and decisions remain available. Usefulness does not answer the harder questions created by persistence.
The research problem
A retained correction may improve a later answer. It may also outlive the context that made it accurate. A permission valid in one system may be treated as though it survived migration into another. An inference that began as a tentative prediction may become part of a durable identity record. These are different failures, but they share a structural source: relationship state has become consequential without an adequate account of its provenance, validity, correction, and authority.
My work examines that gap. The central concern is not persistence by itself. It is the configuration required when persistent state begins to shape later interaction: what must be recorded, what must remain contestable, what can travel, and what has to be reevaluated before a receiving system acts.
Current lines of inquiry
The working relationship
The Operant Dyad describes a task-bound relationship formed through repeated correction between one human and one functionally bounded adaptive system. The proposal keeps human authorship visible as a governing condition rather than assuming that the system already preserves it.
Human-governed continuity
The Persistent Primitive develops requirements for continuity that remains attached to the person whose interaction gives it value. The framework distinguishes retained state from legitimate use and treats correction, revocation, purpose, and temporal validity as part of the architecture.
Legal provenance
Authority Does Not Travel by Default asks what happens when relationship state crosses providers, applications, roles, or jurisdictions. The paper argues that a receiving system must reevaluate current operability instead of treating source-side authority as a permanent technical attribute.
Emergent novelty
Emergent Novelty proposes a way to study outputs that may arise from the interaction dynamics of a particular human–AI pairing. The evidence is preliminary and comes from a single longitudinal case study; the framework remains a research agenda requiring outside testing and revision.