Titolo della tesi: Modeling Complex Digital Systems
This dissertation applies Complexity Science to the study of digital environments, emphasizing the effects of recommendation algorithms and the development of agent-based models augmented by Large Language Models (LLMs). The thesis begins by examining power laws in complex systems, discussing their mathematical properties, connections to Heaps' law and Zipf's law, and implications for understanding phenomena like Black Swans and extreme events. Also network theory is revisited and special attention is given to bipartite networks, crucial for the analysis of recommendation algorithms. A large part of the thesis is devoted to the role of recommendation algorithms in online social systems, encompassing a literature review of their effects on inequality, opinion polarization, and echo chamber formation. This is complemented by a detailed analysis of opinion dynamics models influenced by recommendation algorithms and a study of user-user collaborative filtering. On one hand, our results show that recommendation algorithm may fundamentally alter social dynamics in large groups, inducing opinion polarization. However, mathematical models can be used to limit this drawback and to understand how to use these algorithms to improve users' content consumption. As final topic, the Thesis covers the latest advancements in AI, particularly LLMs, and their applications in social sciences. It discusses the concept of generative agents and their potential in simulating online social environments and testing platform interventions. This includes a model of network growth based on generative agents demonstrating emergent scale-free structures and linear preferential attachment among them. Such a finding is a first step in replicating online social environment using in silico LLMs powered simulations.