Align AI to Dynamic Human-AI Workflows

arXiv cs.AI · 6 hours ago

Researchers argue that AI alignment should shift from static preference matching to dynamic, interactive approaches where human and AI behavior co-evolve over time. The paper grounds this perspective in interdisciplinary workshop insights and social science accounts of human collaboration, identifying how human-AI systems introduce new asymmetries and coordination challenges not present in human-human interaction. The proposed research agenda requires combining machine learning with social and decision sciences to develop AI systems that align through ongoing interaction rather than fixed preference representations.