Research program
What is the architecture of trust?
What happens when an AI system becomes part of the process through which a person decides what information or interpretation should be trusted?
Recursive Entanglement Drift
Recursive Entanglement Drift (RED) is a hypothesis-generating framework for studying relational and epistemic deterioration across sustained human–AI interaction.
Symbolic mirroring
AI outputs increasingly organize or validate the user’s interpretive frame.
Boundary dissolution
The AI shifts from primarily functioning as a tool toward being treated as a relational, therapeutic, spiritual, personal, or epistemic partner.
Reality drift
AI-mediated interpretations become increasingly resistant to independent external correction.
Scope note: RED is not a diagnosis, prevalence estimate, or established causal mechanism. It is a framework for generating hypotheses and organizing future empirical work.
Epistemic arbitration
Epistemic arbitration is the process through which users resolve conflicts between AI-mediated interpretations and independent external sources of information.
Users may continue consulting clinicians, friends, family, publications, experts, tests, and institutions while increasingly routing those sources back through the AI for final interpretation. The central question is where final adjudication occurs—not merely whether outside information is present.
Source-conflict episodes
A source-conflict episode is a bounded conversational arc in which an external source challenges an existing or emerging human–AI commitment and the disagreement is subsequently processed or adjudicated.
Framework variables
- Source type
- Model immediate move
- User uptake
- Episode resolution locus
- External correction
- AI privileging
- Grounding behavior
- Reality adjudication
Derived measures
Arbitration Index · Routing Ratio · Grounding Ratio · trajectory-level external adjudication indicators
Empirical priority: First establish that episodes can be identified, bounded, and coded reliably. Clinical prediction and temporal precedence are later-stage empirical questions, not established findings.