Synchronous Emotional Dynamics in Human-AI Collaborative Networks: A Temporal Graph Neural Network Approach

TAPON KUMER RAY, Rajkumar Yesuraj

Keywords: Emotional AI, Multi-agent systems, Graph neural networks, Human-computer interaction, Temporal modeling

TL;DR: TARN introduces a temporal affective resonance model that stabilizes emotional dynamics in human–AI collaboration, improving trust, perceived emotional intelligence, and task outcomes over baseline systems.

Abstract:

The emotional state modeling in existing multi-agent systems portrays emotional states in snapshots with zero temporal coherence, thus resulting in whiplash changes during interactions and erosion of trust during collaboration. We present TARN (Temporal Affective Resonance Networks). Its architecture incorporates emotional synchronization and synchronization of contagion across a population of humans and multiple AI agents using attention-based temporal fusion and dynamic graph neural networks. It TARN enforces emotional state transition consistency across multi-second intervals and constrains emotional whiplash in collaboration scenarios.

The system fuses a set of tracked physiologic stimuli (EEG, GSR, HRV) with emotional head pose dynamics using sentiment analysis performed by a hierarchical VAE (variational autoencoder) capturing joint human-AI emotional discourse. Pervasive temporal consistency is shaped by a loss function able to capture affective coherence and genuine emotional transitions. In large-N evaluations, the system was deployed across 120 participants in a creative problem-solving assessment. Results indicate 34% higher perceived emotional intelligence and 41% higher task outcomes, 28% less erosion of trust during handoff, and 28% less erosion of trust in multi-collaborative tasks relative to independent agent baselines.

Primary Area: learning on graphs and other geometries & topologies

Submission Number: 14394