To Split or Not to Split? Evaluating IA Roles Providing Knowledge and Emotional Support

As intelligent agents (IAs) become increasingly integrated into human-machine systems, understanding the dynamics of collaboration between individuals and IAs is crucial. This study explores the impact of different IA roles and users’ knowledge levels on collaborative task performance and subjective evaluations. Three IA role design approaches are investigated: expert-type IA providing only knowledge support, mentor-type IA (combined-persona) providing both knowledge and emotional support, and IA group (split-persona) providing knowledge and emotional support separately. Participants with high and low task-related knowledge levels were involved. Results reveal that the combined-persona IA enhances task performance but receives lower subjective evaluations, whereas the split-persona IA receives higher subjective evaluations but reduces task performance, indicating a trade-off between objective and subjective measures. Additionally, users’ knowledge levels influence preferences for IA roles, with low-knowledge participants favoring emotional support. Recommendations are made for adjusting IA roles based on task requirements and user characteristics.