Main article

Wei-Ling Tan
Department of Information Systems, Faculty of Computing, Universiti Malaysia Pahang Al-Sultan Abdullah, Pekan, Malaysia
Faridah binti Hashim
Faculty of Business and Management, Universiti Teknologi MARA Selangor Branch, Puncak Alam Campus, Selangor, Malaysia
Mohd Azlan Ibrahim*
School of Computer Sciences, Universiti Sains Malaysia, Penang, Malaysia
m.azlan@usm.my

DOI: https://doi.org/10.63646/jaiaa.2024.020202

Abstract

Generative artificial intelligence (GenAI) has rapidly moved beyond content production into operational roles where its outputs influence consequential decisions across industry, healthcare, education, finance, and public governance. As deployment scope expands, a recurring failure mode becomes visible: human operators systematically miscalibrate their trust in GenAI systems, either over-relying on confident but ungrounded outputs or under-using reliable recommendations because of opaque reasoning. This paper proposes a cross-domain framework, Trust Calibration Analytics (TCA), that treats trust calibration not as a one-time design decision but as a continuous analytical process embedded in deployment. Drawing on 71 peer-reviewed sources spanning 2014–2025 and an empirical synthesis of 312 documented GenAI deployments across five application domains, we develop a three-layer architecture comprising a generative substrate, a calibration analytics engine, and an operational context layer. Quantitative analysis demonstrates that the full TCA pipeline reduces expected calibration error by 76% on average across domains, while deployments with explanation modules and adaptive oversight show a 24-percentage-point reduction in inappropriate operator overrides. Five operational scenarios are analysed to derive three transversal design principles: situation-anchored confidence reporting, drift-aware trust monitoring, and risk-graded oversight orchestration. The paper concludes with a research roadmap identifying confidence-calibrated outputs, cross-domain drift detection, federated oversight, audit-grade provenance, and adaptive escalation as priorities through 2028. TCA reframes human-AI collaboration around measurable trust dynamics rather than assumed reliability.

Article details

How to Cite

Tan, W.-L., Hashim, F. binti, & Ibrahim, M. A. (2024). Trust Calibration Analytics for Generative AI Systems: A Cross-Domain Framework for Operational Intelligence and Human Oversight. Journal of AI Analytics and Applications, 2(2), 23-39. https://doi.org/10.63646/jaiaa.2024.020202