Traceable judgments, critical reflection and responsible AI governance
Responsible AI is often defined through principles and commitments. But how can we examine the reasoning behind AI assessments, challenge their conclusions, and ensure that human judgment remains central?
rAISE (Responsible AI for Self-directed Enquiry) explores how AI systems can support critical reflection and accountable decision-making without replacing human judgment. Building on aihorizon R&D’s Responsible AI Framework, the project develops approaches that make normative commitments, supporting evidence, underlying assumptions, and the limits of AI assessments transparent and contestable.
At its core is a distinction between four dimensions of responsible AI: epistemic justification, normative legitimacy, legal compliance, and contextual appropriateness. Rather than collapsing these dimensions into a single assessment, rAISE seeks to make their different conclusions, potential conflicts, and underlying reasons visible, allowing people to examine and weigh them independently.
The project develops an interactive Reflection App in which individuals can critically revisit their use of generative AI, examine the reasoning behind assessments, and challenge underlying assumptions and normative standards. Specialised AI agents support this process through structured, traceable assessments while leaving the weighing of potentially conflicting considerations to human users.
A central research question is whether AI-assisted reflection strengthens people’s capacity for independent judgment or creates new forms of dependence on AI itself. By investigating how people engage with, challenge, and respond to AI-generated assessments, rAISE explores the conditions under which AI can support self-directed enquiry, responsible delegation, and accountable human–AI interaction.