Probabilistic Observation over Deterministic Identity Kernels_ A Derived Architecture for Adaptive AI Systems
Year: 2026
Type: article
Abstract: This note proposes an architectural separation for adaptive AI systems: identity is deterministic, while observation is probabilistic. Real-world AI systems operate under noisy, incomplete, and shifting conditions, making direct application of rigid identity evaluation either brittle or overly permissive. The proposed architecture separates these concerns into three layers. A probabilistic observer estimates which identity regime is active. A deterministic identity kernel evaluates persistence against declared objects, admissible transformations, invariants, drift bounds, and verdict rules. A higher-level governance layer manages lawful adaptation through regime changes, branching, and hierarchical constraint management. Probability therefore does not define identity; it helps discover and select the regime under which identity can be judged. The note argues that confidence should not replace proof, that flexibility requires explicit proof obligations, and that adaptive AI systems should expose their governing structures through white-box artifacts rather than opaque confidence scores. This work is a derived systems architecture within the broader identity-persistence program and serves as a bridge between formal identity theory and practical AI governance infrastructure.
Author: Bostick, Devin
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7165805830
FWCI:
Cites:
Cited by:
Related to: