Explainable Artificial Intelligence in Human–AI Collaboration: Enhancing Trust and Interpretability in Decision Systems

Abstract

Explainable artificial intelligence (XAI) addresses the challenge of understanding, validating, and appropriately trusting model outputs in consequential decision environments. This article introduces a formal evaluation framework comprising three quantitative measures for assessing XAI effectiveness in human–AI collaborative decision systems: the Trust Calibration Score (TCS), which quantifies alignment between user trust and actual system accuracy; the Explanation Utility Index (EUI), which combines comprehension gain and decision quality improvement in a weighted composite; and the Human–AI Collaboration Gain (HACG), which measures performance surplus or deficit relative to individual baselines. Utilizing existing XAI methods (LIME, SHAP, Grad‑CAM, and counterfactuals), the framework is applied across three deployment conditions and three application scenarios. XAI‑supported conditions yield TCS values above 0.95, compared to 0.74 under the black‑box condition, with HACG results showing a 7–9% improvement over the strongest individual baseline. The proposed framework provides a replicable methodology for evaluating and optimizing XAI implementations across clinical, financial, and organizational decision environments.