Researchers built a gamified platform called Spook the Machine in which participants craft and interact with AI characters that display fear. The study examines whether people respond differently depending on how a machine expresses fear, and what those reactions suggest for designing affective human-AI interaction. It is an arXiv preprint focused on emotion-oriented AI design.
A revised preprint on arXiv examines task-oriented dialogue systems, which help users complete goals through natural language. The authors argue that emotional factors shape learning in these systems, and that success depends not only on completing tasks but also on sustaining positive emotional exchanges and conveying information accurately.
Researchers have introduced RESCUE-BENCH, a benchmark designed to push emotional support conversation systems beyond traditional one-on-one interactions. The work addresses the gap in handling multi-party scenarios, where the relationships between participants shape how support should be delivered. It marks a step toward support systems that reason about interpersonal dynamics rather than just individual emotional states.
A new arXiv paper argues that affective computing's dominant approach, which assigns discrete emotion labels or arousal/valence scores to individual speakers, is inadequate for understanding emotional life as it happens between people. The authors instead build on the ideas of affective resonance and vitality affects to model vocal exchanges as dynamic, shared interaction fields. The work reframes emotion recognition as a relational process rather than the classification of isolated internal states.