Developing the Generative Artificial Intelligence Trustworthiness and Accuracy Index (GAITAI): A Multidimensional Framework for Evaluating Large Language Models

Abstract

The rapid advancement of generative artificial intelligence (GenAI) and large language models (LLMs) has transformed sectors such as healthcare, education, finance, law, and public administration. Despite significant improvements in computational capability, concerns regarding hallucinations, misinformation, inconsistency, bias, opacity, and governance have exposed important limitations in existing benchmark-based evaluation methods. Current frameworks primarily assess task performance and reasoning ability but provide limited insight into the broader characteristics that determine whether AI systems can be trusted in high-stakes environments. This study addresses this gap by developing the Generative Artificial Intelligence Trustworthiness and Accuracy Index (GAITAI), a multidimensional conceptual framework for evaluating AI trustworthiness beyond traditional benchmark metrics. Using a conceptual research design, the study integrates Information Quality Theory, Trust Theory, the Technology Acceptance Model, Measurement Theory, Composite Index Theory, Artificial Intelligence Governance Theory, and Institutional Theory to establish an interdisciplinary foundation for trustworthy AI evaluation. GAITAI conceptualises Generative Artificial Intelligence Trustworthiness as a second-order latent construct comprising four first-order dimensions: Accuracy, Reliability, Robustness, and Transparency, operationalised through twenty-eight observable indicators. The study contributes theoretically by redefining AI trustworthiness as a multidimensional construct and methodologically by extending psychometric measurement and composite indicator principles to AI evaluation. Practically, GAITAI provides organisations, AI developers, policymakers, and regulators with a structured framework for assessing AI trustworthiness and supporting responsible AI governance. Although conceptual, the framework establishes a foundation for future empirical validation, comparative benchmarking, and international standardisation. GAITAI advances the development of scientifically rigorous methods for evaluating trustworthy generative artificial intelligence and supports evidence-based AI governance across diverse organisational and societal contexts.