Invited Talk 2 by Mohamed Rayan Barhdadi (Undergraduate Student and Researcher at Texas A&M University): EMPATHIA: Multi-Agent Reasoning for AI That Understands and Supports Human Needs

Current AI systems often optimize narrow objectives and overlook the emotional, cultural, and ethical dimensions that shape real human outcomes. We present EMPATHIA, a multi-agent framework designed to teach AI to reason with multiple human perspectives and support decisions that preserve dignity rather than reduce people to metrics. Grounded in Kegan’s Constructive Developmental Theory, EMPATHIA models human development through three modules: SEED for initial assessment, RISE for early independence, and THRIVE for long-term adaptation. SEED uses a selector–validator architecture with multi-faceted agents that deliberate transparently to generate explainable recommendations. Evaluated on the UNHCR dataset, EMPATHIA achieves high validation convergence (87.4%) and consistent, interpretable assessments across host countries. By integrating multiple forms of human reasoning and maintaining practitioner oversight, EMPATHIA shows how multi-agent architectures can extend beyond accuracy to support value-sensitive, human-centered decision-making.