Work

Am I as Effective at Identifying Emotions as Artificial Intelligence? A Comparative Study of Emotion Recognition

Year: 2025

Type: article

Abstract: Background Learning conversations, or dialogues aimed at deepening understanding and reflection, are deeply influenced by emotions. Effective communication is influenced by emotional intelligence - the ability to recognize, understand, and manage both one’s own and others’ emotions. While advances in artificial intelligence (AI) offer new tools for emotion recognition, these technologies still struggle with accurately interpreting subtle and culturally diverse emotional expressions, sparking debate about their reliability and effectiveness. This article provides a comparative analysis of human versus AI recognition of emotions during an end-of-course reflective learning conversation. Methods Emotions during a structured post-conference debriefing were analyzed and coded by MorphCast (“AI data”) in real time and human researchers (“peer data”) from a video recording of the conversation in which audio was removed. Each individual in the learning conversation also self-assessed and coded their own emotions (“self data”) from the same recording. The AI and peer data were compared to the self data. Findings The AI dataset captured a wider range of emotions - including anger, sadness, and disgust - not detected by Self or Peer assessments, which predominantly identified happy and neutral emotions. Happiness was lower in MorphCast assessment (20.86%), higher in human assessment (59%), and highest in self-assessment (65%). Conclusions Human identification was more similar to self-identification of emotions during the learning conversation than AI. AI identified a broader range of emotions, especially negative ones which were largely absent in self- and peer-reports that leaned toward happy and neutral states. These differences highlight how AI, lacking contextual understanding and relying solely on facial cues, may misinterpret emotions compared to humans who integrate verbal, situational, and social information into their assessments.

Source: Cureus Journal of Computer Science.

DOI: 10.7759/s44389-025-07224-y