New AI framework reflects human physiology to understand emotional experiences

Emotions are a fundamental part of human psychology and are complex processes that have long distinguished humans from machines. Even advanced artificial intelligence (AI) does not have the ability to feel. But researchers are now exploring whether they can model emotion formation in computers to give machines a deeper, more human-like understanding of emotional states.

In this vein, Assistant Professor Chie Hieda of the Nara Institute of Science and Technology (NAIST), in collaboration with Assistant Professor Kazuki Miyazawa and then master’s student Kazuki Tsurumaki from Osaka University in Japan, is researching computational approaches to modeling the formation of emotions. In a recent study, this team of researchers built a computational model that aims to explain how humans form the concept of emotion. The study was made available online on July 3, 2025 and was published in the journal Volume 16, Issue 4. IEEE Transactions on Affective Computing December 3, 2025.

This model is based on the theory of constructed emotions, which holds that emotions are constructed by the brain in the moment, rather than being innate responses. Emotions arise from the integration of internal bodily signals (interoception, such as heart rate) and external sensory information (exteroception, such as sight and hearing), allowing the brain to create concepts rather than mere reflexes.

“Although theoretical frameworks exist for how emotions emerge as concepts through information processing, the computational processes underlying this formation remain to be elucidated.” Dr. Hieda says.

To model this process, the research team used multilayer multimodal latent Dirichlet allocation (mMLDA). It is a probabilistic generative model designed to discover hidden statistical patterns and categories by analyzing how different types of data co-occur without pre-programming emotional labels.

The developed model was trained using unlabeled data collected from human participants who viewed emotionally evocative images and videos. The system was not told which data corresponded to emotions such as fear, joy, and sadness. Instead, it can now identify patterns on its own.

Twenty-nine participants viewed 60 images from the International Affective Picture System, which is widely used in psychological research. As the researchers viewed the images, they used wearable sensors to record physiological responses, such as heart rate, and collected verbal descriptions. Taken together, these data recorded how people interpret emotions: what they see, how their bodies respond, and how they describe their emotional experiences.

When comparing the trained model’s emotion concepts with participants’ self-reported emotion ratings, the agreement was approximately 75%. This is significantly higher than would be expected by chance, suggesting that the model categorized emotion concepts that closely matched the way people experience emotions.

This research paves the way for more sensitive and responsive AI systems by modeling emotion formation in a way that reflects human experience. “Integrating visual, linguistic, and physiological information into conversational robots and emotional recognition AI systems has the potential to enable more human-like emotional understanding and situational responses.” Dr. Hieda says.

Additionally, the model can infer emotional states that people have difficulty expressing in words, which could be particularly useful for mental health support, healthcare monitoring, and assistive technology for conditions such as developmental disorders and dementia.

”This research has important implications for both society and industry, as it provides a computational framework that connects emotion theory and empirical testing to address the long-standing question of how emotions are formed.” concluded Dr. Hieda.