https://www.mdu.se/

 This licentiate thesis establishes a normative and methodological foundation for operationalising Responsible AI (RAI), grounded in the philosophical commitments of Digital Humanism. Despite the proliferation of AI ethics guidelines across policy, technical research, and industry domains, a persistent and well-documented implementation gap remains between high-level ethical principles and their practical implementation in AI engineering. This gap is structural, arising from institutional separation among policy, technical research, and engineering practice, as well as systematic failures to translate abstract values into actionable engineering processes.

The thesis argues that addressing this gap requires three elements: a normative foundation that goes beyond compliance-oriented metrics, a principled method for making value trade-offs explicit and open to deliberation, and a concrete mechanism for integrating ethical reasoning across the AI lifecycle. Drawing on Digital Humanism, axiology, and Multi-Criteria Decision Analysis (MCDA), it develops the Digital Humanism AI Ethics Toolkit. Within this toolkit, the H.E.A.R.T. model functions as a decision-support mechanism embedded across design, feedback, and continuous improvement processes. Rather than treating ethics as an external constraint or post-hoc evaluation layer, the toolkit supports reflective and accountable decision-making within existing engineering and governance workflows. Across the included studies, the thesis connects a structural diagnosis of Responsible AI operationalisation barriers with the development of methodological and engineering support for value-sensitive AI design and governance.