Human-Centered Explainable AI: Enhancing User Trust and Interpretability in Generative and Autonomous AI Systems

Abstract

ABSTRACT As generative artificial intelligence (GenAI) and autonomous intelligent agents transition into high-stakes operational domains (such as healthcare diagnostic support, autonomous vehicle navigation, financial auditing, and automated legal analysis), the demand for human-centered explainable AI (HC-XAI) has become critical. Traditional Explainable AI (XAI) frameworks have focused primarily on post-hoc mathematical fidelity, offering feature attribution heatmaps or surrogate model weights designed for AI engineers. However, these developer-centric techniques fail to establish cognitive alignment or calibrated trust for non-expert human decision-makers. This review paper presents a comprehensive critique and taxonomy of HC-XAI mechanisms tailored specifically for generative foundation models and dynamic autonomous agents. We analyze the cognitive, ergonomic, and psychological dimensions of user trust, examining how operator cognitive load, domain expertise, confirmation bias, and decision agency interact with explainability interfaces. Furthermore, we synthesize recent state-of-the-art developments in interactive, counterfactual, and natural language explanation frameworks. Through benchmark analysis across four core operational metrics—trust calibration, cognitive overload, joint decision accuracy, and explanation latency—we demonstrate that adaptive human-centered explainability substantially outperforms static feature attributions. Finally, we propose a unified architectural framework for human-centered autonomous XAI, address ethical and adversarial vulnerabilities, and delineate future research directions toward ethical, transparent, and human-trusted AI deployment.

KEYWORDS: Human-Centered Explainable AI (HC-XAI), Trust Calibration, Generative AI, Autonomous Systems, Interpretability, Cognitive Load, Counterfactual Rationales, Foundation Models.