Emotion Recognition Systems Must Embrace Ambiguity

Automatic Emotion Recognition (AER) systems aim to emulate human-like perception of emotions. However, existing techniques often neglect the inherent ambiguity in emotion perception. This paper argues that this oversight fundamentally inhibits the ability of such systems to mimic human-like abilities and advocates for a more nuanced approach to emotion modeling-one that embraces ambiguity. We provide a thorough exploration of positions related to ambiguity-aware emotions, delving into both supportive and opposing viewpoints. Our argument emphasises that ambiguity should be represented through probability distributions, and continuous modelling of emotions over time is crucial to capturing human-like perceptions. By integrating these elements, AER systems can achieve more natural and effective interactions with users, better reflecting the complexities of human emotional perception. We conclude by offering guidelines for advancing ambiguity-aware emotion modeling in future research. In summary, embracing ambiguity in AER systems leads to more robust and contextually sensitive emotion recognition, enhancing their practical utility and bridging the gap between human and machine understanding.