Representational reversibility concerns how information carried forward in longitudinal AI systems should change in status and authority as the evidence, circumstances, or person represented changes over time. It distinguishes retaining information from continuing to treat that information as authoritative for subsequent inference.
The project examines how provenance, temporal change, evidentiary warrant, and supersession should shape what a system continues to rely on, rather than simply what it continues to retain. It brings together philosophy of AI, philosophy of psychiatry, and the evaluation of longitudinal AI systems.
Current work
When Should an AI System Stop Representing a Person as It Once Did? Reversibility and Longitudinal Representation
Workshop submission, 2026.
Related work
Narrative and diagnostic capture
This research examines how interpretations are preserved, circulated, and repeatedly reactivated across institutional systems, shaping understanding as people and circumstances change.
NIST AI TEVV-Athlon
Work on test, evaluation, verification, and validation (TEVV) approaches for longitudinal clinical AI, including how systems should respond when previously represented information changes in evidentiary status or authority. A public comment was submitted in an individual capacity in 2026; a public link is forthcoming.