Agency Over Transparency: Empirical Evidence for the Understanding-Trust Gap in AI-Assisted Camouflage Analysis

Abstract

Explainable AI assumes a simple chain: transparency produces understanding, and understanding produces trust. We empirically debunked this assumption. In a controlled study (N = 150) comparing two AI-assisted camouflage detection interfaces, users who fully understood how a system worked still did not trust it. This understanding-trust gap persisted across all conditions. Adding interactive explainability produced a very large effect on perceived agency (d = 1.19) but only a small effect on trust (d = 0.33) and no effect on understanding (d = -0.02). Agency increased substantially while trust improved only marginally. We introduce the Mixed-Initiative Camouflage Analysis (MICA) framework, which repositions structured user control rather than transparency as the operative mechanism for trust calibration.