EXPLORING EMOTIONALLY INTELLIGENT MACHINES: CAN AI TRULY UNDERSTAND HUMAN AFFECT?

Abstract

Artificial intelligence systems now claim the ability to recognize human emotions. Many systems also claim to respond to these emotions appropriately. Researchers question whether such systems truly understand human affect. This study examined that question using a mixed-methods design. The researchers recruited 150 participants for the quantitative component. Participants interacted with three commercially available emotion-recognition AI systems. Each participant completed a structured questionnaire after the interaction. The questionnaire measured perceived accuracy, empathy, and trust. The researchers then conducted semi-structured interviews with twenty participants to explore their subjective experiences in greater depth. The researchers also analyzed system-generated logs. These logs recorded the emotional labels detected by each AI system. The researchers compared these labels against participants' self-reported emotional states. Results showed that AI systems detected basic emotions with reasonable accuracy. However, the systems failed to detect nuanced or mixed emotional states. Participants rated AI empathy lower than AI accuracy. Correlation analysis revealed a moderate relationship between accuracy and trust. Interview data showed that participants valued responsiveness over technical accuracy. Participants described their interactions as helpful but emotionally shallow. The study concludes that AI simulates emotional intelligence at a surface level. Genuine affective understanding remains beyond current AI capability. This finding carries implications for the design of future emotionally intelligent systems