Expertise-Indexed Trust Calibration: Why Observed AI Reliability Emerges from Model Capability, Evidence Grounding, and User Expertise

The public debate on artificial intelligence focuses almost entirely on the model: which AI is best, which hallucinates least, which has the largest context window. I propose that this is the wrong question. The observed reliability of an AI system is not a fixed property of the model, but a property that emerges from the interaction of three factors — model capability, available evidence, and the user's expertise in the domain of the question. I propose to term this relationship the Kraemer Asymmetry (K-Asymmetry), represented by two complementary artifacts: a descriptive curve of the maturation of user trust over time, and a four-pillar model integrating three established traditions — the Dunning–Kruger effect, trust calibration in automation, and human–AI teaming — with a fourth, empirical pillar: evidence grounding. Findings from the EvidenX citation-integrity study are consistent with the model, although they were not designed to isolate user expertise, which remains a hypothesis to be tested. This is an integrative conceptual model, not a universal law.