Establishing and evaluating trustworthy AI: overview and research challenges.

Area of Science:

• Computer Science
• Ethics
• Technology Policy

Background:

• Artificial intelligence (AI) technologies are transforming society, necessitating a focus on trustworthy AI systems.
• Concerns regarding unexpected outcomes and misuse of AI have spurred discussions on AI trustworthiness.

Purpose of the Study:

• To synthesize existing conceptualizations of trustworthy AI.
• To consolidate discussions on AI trustworthiness across academic and public forums.
• To provide a reference for a broad audience and guide future research.

Main Methods:

• Synthesis of existing literature on trustworthy AI.
• Identification and definition of six core AI trustworthiness requirements.
• Analysis of research challenges specific to each requirement and overarching issues.

Main Results:

• Six key requirements for trustworthy AI identified: human agency and oversight, fairness and non-discrimination, transparency and explainability, robustness and accuracy, privacy and security, and accountability.
• Definitions, evaluation methods, and requirement-specific research challenges are detailed for each aspect.
• Overarching research challenges include interdisciplinary collaboration, conceptual clarity, context-dependency, system dynamics, and real-world validation.

Conclusions:

• Consolidated framework for trustworthy AI presented, synthesizing diverse perspectives.
• Highlights critical research gaps and future directions for developing responsible AI.
• Aims to serve as a foundational reference for researchers, policymakers, and the public.