https://www.mdu.se/

Responsible AI (RAI) approaches have expanded across policy, technical, and industry domains, creating a broad normative landscape. Despite this proliferation, a significant gap remains between high-level ethical principles and their practical implementation in AI engineering. This paper reviews key RAI approaches in these domains and identifies three structural operationalisation gaps: the Abstraction Gap which principles are expressed at too general a level to guide engineering practice; the Contradiction Gap when approaches present conflicting principles without offering methods for resolving them; and the Technocentric Gap that ethical concerns are reduced to measurable indicators that overlook human and sociotechnical context. These gaps stem from the institutional separation of policy, technical research, and industry practice. Overcoming these challenges requires integration of operational methods that link ethical commitments, engineering processes, and community participation throughout the software development lifecycle. This analysis provides a foundation for developing more effective strategies for operationalising Responsible AI.

As Artificial Intelligence (AI) becomes increasingly influential, ethical concerns surrounding its impact on society and people grow. Current AI ethics frameworks often prioritize efficiency and optimization, neglecting core humanistic values such as dignity, autonomy, and social justice. Digital Humanism offers an essential philosophical foundation to ensure that AI aligns with human values and societal well-being. This paper explores the integration of Responsible AI within the philosophical framework of Digital humanism. It argues that Digital Humanism principles, such as Human-Centered Design, inclusivity, transparency, and ethical universalism, offer critical guidelines for developing and deploying AI that serves humanity. In doing so, this work advocates a humanistic approach to AI development that transcends the limitations of technocentric paradigms.

The fast advancement of artificial intelligence presents ethical challenges that exceed the scope of traditional moral theories. This paper proposes a value-centered framework for AI ethics grounded in axiology, which distinguishes intrinsic values like dignity and fairness from instrumental ones such as accuracy and efficiency. This distinction supports ethical pluralism and contextual sensitivity. Using Multi-Criteria Decision Analysis (MCDA), the framework translates values into structured evaluations, enabling transparent trade-offs. A healthcare case study illustrates how ethical outcomes vary across physician, patient, and public health perspectives. The results highlight the limitations of single-theory approaches and emphasize the need for adaptable models that reflect diverse stakeholder values. By linking philosophical inquiry with governance initiatives like Responsible Artificial Intelligence (AI) and Digital Humanism, the framework offers actionable design criteria for inclusive and context-aware AI development.

AI systems generate ethical tensions that cannot be addressed through principle-based guidance alone. This paper brings forward an Integrated Axiology--MCDA Framework for AI ethics that distinguishes intrinsic, instrumental, and relational values and uses multi-criteria analysis to operationalise value pluralism in practice. The framework structures ethical evaluation by making value commitments explicit, enabling transparent examination of trade-offs, and supporting context-sensitive judgment. A healthcare hyper-scenario with sensitivity analysis shows how different weight configurations influence the relative acceptability of diagnostic systems and clarifies the thresholds at which accuracy considerations outweigh privacy or fairness. Cross-domain applications in education, criminal justice, and finance further illustrate how domain-specific value tensions require distinct criteria sets and weighting structures. The analysis shows that ethical challenges in AI arise from genuine value pluralism. Explicit value classification enables more accountable decision-making across the AI lifecycle.

 This paper introduces the Digital Humanism AI Ethics Framework for embedding ethical considerations across the AI development lifecycle. Grounded in the principles of human dignity, autonomy, cultural pluralism, and democratic participation, the framework addresses the limitations of compliance‑oriented Responsible AI approaches. It is structured around three interconnected layers: Foundational Principles, Design and Development Tools, and Governance and Oversight. The H.E.A.R.T. model serves as a practical method within the toolkit for embedding ethical reasoning throughout the lifecycle and is explicitly grounded in a tripartite value taxonomy that distinguishes intrinsic, instrumental, and relational values. Multi‑Criteria Decision Analysis (MCDA) is included as a supplementary toolkit method for structured value trade-off analysis when ethical review reveals unresolved conflicts in complex and pluralistic contexts. By integrating with CRISP‑ML(Q), the toolkit operates as an overlay on standard machine‑learning engineering practice, treating ethics as a core quality concern at each development phase. Framing ethics as integral to system design rather than as a post‑hoc consideration, the framework and toolkit enable more participatory and context‑sensitive AI development, supporting stronger alignment with human values and greater societal legitimacy.