Are we going towards Human-AI Coevolution? A multi-leveled framework to study the impact of human-AI interaction on societal behaviors

Human-AI Coevolution (HAIC) is a framework theorised by Pedreschi (2025), aimed at filling an explanatory gap among approaches to machine behavior, which currently succeed in modeling user-AI interactions (i.e. as feedback loops), but fail to account for how these can ripple out into societal effects (such as polarization, homogenization or discrimination). On the contrary, HAIC proposes to study the impact of human-AI interactions on social behaviors by taking a multi-leveled viewpoint, as defined by the literature on the coevolutionary history of technology and the modeling tools of complexity science. In this paper, we present two case studies for the application of HAIC, based on studies conducted by Doshi (2024) on human-LLM collaboration for creative writing and by Aridor (2020) on the impact of recommender systems on users’ behavior. We argue that HAIC can uniquely be used to interpret the results of these studies as the evolution of human behaviors across different societal levels, through their interaction with AI systems.