Survey of Trustworthy AI: A Meta Decision of AI

Trust in AI requires a structured framework: The paper proposes organizing trustworthy AI into three levels—basic, authentic, and articulate—to help organizations determine what degree of trust is appropriate for different AI applications, rather than treating all AI systems as equally trustworthy or untrustworthy.

Ten measurable dimensions define trustworthy AI: The framework identifies specific, concrete criteria for evaluating AI systems including explainability, fairness, privacy protection, safety, accountability, and reproducibility—providing organizations with practical checkpoints beyond vague notions of "trustworthiness."

Transparency alone is insufficient: While explainability and transparency are included as key dimensions, the framework recognizes these must be paired with other safeguards like fairness checks, privacy protections, and accountability mechanisms to create genuinely trustworthy systems.

AI trustworthiness is a strategic decision, not a technical afterthought: The paper frames trust in AI as a meta-decision—a higher-level choice about which information and systems to rely on—suggesting that organizations should deliberately assess trustworthiness before deploying AI in important decision-making contexts rather than addressing concerns after implementation.

Multiple dimensions must work together: The framework emphasizes that trustworthy AI requires attention to interconnected factors including data governance, sustainability, and reproducibility alongside more commonly discussed concerns like fairness and safety, indicating that responsible AI requires comprehensive rather than piecemeal approaches.