- Statistical and Machine Learning Tools for Consonance Detection

The study of emotions via Artificial Intelligence (AI) is considered a hot topic as technological development is gradually turning its gaze more and more toward understanding and emulating human beings. Emotion detection, however, runs the risk of being a useless investigative tool if it is not accompanied by solid theoretical foundations to reinterpret AI-detected emotions through a broader perspective. The Viable Systems Approach (ASV) may likely be the missing link between studying the emotional sphere and understanding the social dynamics that characterize human beings during each interaction. The concepts of consonance and resonance, typical of ASV, make it possible to describe patterns of information acquisition and to understand the dynamics underlying each individual's decision-making. The purpose of our study is to demonstrate, through field experimentation, the possibility of detecting consonance within groups of people in an automatic way, using statistical analysis techniques and Facial Expression Recognition (FER) algorithms. The trial was conducted in two phases: In the first one, a structured questionnaire categorized participants into distinct value clusters based on their strong beliefs; in the second one FER techniques were used to gauge the change in the initial level of consonance within the same clusters during a common contextual experience. Results demonstrated the possibility of detecting the initial level of consonance and its context-driven variation (resonance) both qualitatively and (at least partially) quantitatively. Research findings certified the feasibility of conso-nance measurement, opening the way to new usage perspectives for emotion detection tools in AI.