• AI&Society
• Belief Dynamics
• Mathematical Models of Social Processes

My research has focused on mathematical models of belief exchange and human behaviour of increasing complexity, with an emphasis on emerging domains such as human-AI ecosystems and social media-mediated interactions. The overarching goal is to bridge the gap between the universality and mathematical tractability of simple physics-inspired models and the richness of sociological and psychological insights.
Along with my collaborators, I introduce the Personal-Expressed-Social beliefs model (PES). Personal beliefs are also referred to as internal or private, expressed as public, and social beliefs are the perceptions of others’ beliefs, or second-order beliefs. This distinction allows us to consider a wide range of social-psychological processes, such as social influence, conformity and ego projection, within a model based on dissonance minimization. The utility of this model is twofold. First, it serves as a unifying framework ("meta-model") for many existing models of belief dynamics, which turn out to be special cases of the general model. Second, it enables the exploration of the combined effects of inauthentic expression and belief misperception, thereby offering a more realistic representation of phenomena typical of online social platforms such as the emergence of extremism and the perception of a false social reality. As a possible application, we reproduce with the model the empirically observed underestimation of public support for climate change policies, which can influence individuals’ willingness to engage in pro-social climate action.
I then focus on a series of models, coming from mathematical physics, game theory and opinion dynamics, characterized by a multispecies structure, i.e., two or more groups of individuals with fixed, differing features. Within the Partisanship Voter model and asymmetric games, two groups of individuals with fixed opposite preferences are considered, and the individuals change their opinions through respectively a voter-like dynamics and a coordination game. One is interested in whether the system reaches consensus or a polarized state. Specifically, in the Partisanship Voter Model I consider individuals on a modular network that reflects partisanship, representing the concept of epistemic bubbles, and study how the network structure affects polarization. In asymmetric games, instead, I focus on bounded rationality and map such games into corresponding multispecies Ising models. I rigorously study, using mathematical-physics tools, the equilibrium properties of a similar class of Ising models with a multispecies structure. Finally, in the study of the Prisoner’s Dilemma in an ecosystem composed of interacting humans and AIs, I investigate whether a zealot or a discriminatory AI is more effective at enhancing cooperation.
The usefulness, critical aspects, and limitations of these models (and, more generally, of opinion dynamics models) are discussed as part of a broader exploration of the current frontiers and trends in this field of research.