From Recommendation to Reflection

Measuring Moral Value Stability in Human–AI Collaboration Using Cognitive Value Recontextualization

Abstract

High-stakes human–AI collaboration is typically evaluated on outcome quality or efficiency. In morally charged domains such as disaster response, however, a key question is whether decisions align with the decision-maker’s core moral values. Research in moral psychology shows that individuals may accept identical outcomes under one framing but reject them under another, signaling value stability, not noise.

We propose Value-Recontextualizing Decision Support (VRDS), demonstrated in a wildfire crisis simulation. The system employs Cognitive Value Recontextualization (CVR), which re-presents decisions under morally intensified framings, and Adaptive Preference Alignment (APA), which clarifies whether contradictions reflect situational reasoning or genuine value change. We hypothesize that value-aligned decisions—even when objectively suboptimal—yield higher user satisfaction and well-being, reframing AI decision support from optimizing outcomes to supporting reflective moral reasoning.