RE-centric Recommendations for the Development of Trustworthy(er) Autonomous Systems

Ethical AI guidelines lack consistency: A review of major ethical AI frameworks revealed significant inconsistencies in both the principles they address and the terminology used to describe them, making it difficult for practitioners to understand what "trustworthy AI" actually means in practice.

Requirements engineering is missing from most frameworks: The paper identifies that requirements engineering (RE)—the process of clearly defining what a system should do—is largely absent from existing ethical AI development frameworks, despite being critical for building trustworthiness from the early stages of development.

EU AI Act compliance needs practical guidance: While the EU AI Act will soon mandate compliance with ethical guidelines, practitioners currently lack "actionable instructions" for actually implementing ethics during AI system development, creating a gap between policy requirements and real-world practice.

Existing frameworks have limited applicability: A meta-analysis of ethical AI development frameworks revealed significant limitations in their ability to support effective requirements engineering for trustworthy AI systems, suggesting current tools are insufficient for developers.

Standardized terminology and concrete practices are needed: The paper recommends developing clearer, standardized language across ethical guidelines and establishing concrete development practices to make trustworthy AI development more achievable and consistent across organizations.