Ai As Social Actor And Decision Partner: Behavioral Evidence And Design For Human-centered Systems

Abstract

Artificial intelligence increasingly participates in human systems not as a tool, but as a social actor and decision partner. This shift raises two questions at the core of information systems research: How does AI participation reshape human behavior and social dynamics? And how should AI be designed to act effectively in human-centered domains? I present a research program that answers them by pairing causal empirical analysis with theory-grounded computational design. The first study turns to Social AI Agents—LLM-powered agents that reply to users on behalf of influencers on a major social media platform. Exploiting the staggered rollout of an AI reply feature in a difference-in-differences design, we find that receiving an AI-generated reply increases users’ subsequent engagement, and that the effect strengthens when replies reinforce the influencer’s social presence through relevant content, stylistic alignment, and timely interaction. The results recast social presence as something that can be delegated, and show how behavioral evidence can directly inform the design of AI in social interaction. The second study asks how AI should be designed for high-stakes domains that demand both rigorous reasoning and human-centered interaction. We introduce WiseMind, a knowledge-guided multi-agent framework for psychiatric assessment inspired by Dialectical Behavior Therapy. WiseMind pairs a Reasonable Mind agent for structured diagnostic reasoning with an Emotional Mind agent for empathetic communication, both guided by a DSM-5 knowledge graph that structures clinical inquiry and curbs hallucination. Across simulated patients and real user sessions, WiseMind approaches clinician-level diagnostic performance while sustaining empathetic, clinically appropriate dialogue. Together, these studies sketch a broader agenda for AI in socio-technical systems—one that integrates causal evidence on human behavior with theory-grounded design to guide how AI is built and deployed in complex human environments.