New arXiv paper redefines AI affect as coordination layer for agent collaboration
How emotion-like behaviors govern human trust and delegation of consequential tasks to autonomous agents.
A new review paper on arXiv from researchers including Junjie Xu tackles the fragmented landscape of affective computing in AI agent collaboration. Current research on simulated empathy in LLMs, trust in automation, and AI safety remains siloed, lacking an integrated account of how affective cues operate when humans delegate, monitor, and correct consequential tasks to agents with memory, tool use, and partial autonomy.
The authors synthesize existing work to propose a framework that treats affect not as an internal property of AI but as a dynamic coordination layer. This layer governs reliance, repair, and oversight through interaction loops driven by model-generated affective signals. The framework provides a foundation for calibrated measurement, purposeful design, and informed governance of human-AI agent partnerships, emphasizing the critical role of perceived agent affect in trust calibration and delegation decisions.
- Synthesizes fragmented literature on affective computing, simulated empathy in LLMs, trust in automation, and AI safety.
- Proposes affect as a coordination layer for trust calibration, delegation, error correction, and governance in agentic collaboration.
- Provides a unified framework for measuring, designing, and governing affective dynamics in human-AI agent interactions.
Why It Matters
As autonomous agents grow capable, managing human trust through affective cues becomes critical for safe and effective collaboration.