Séminaire mOeX - Eleni Nisioti

Abstract

Recent developments in AI have revived the search for agents with human-like behaviour: language, tool making and cooperation are skills that we often study in experiments with populations of agents adapting through reinforcement learning or evolutionary optimisation. Yet, despite this close link to human behaviour, research in AI focuses on engineering cognitive architectures and rarely tries to understand how these skills emerged for humans as a result of ecological pressures. In this talk I will summarise the trajectory we followed in the past 2.5 years trying to draw links between the fields of AI and Human Behavioral Ecology. I will present a framework that we came up with to conceptualise skill acquisition in both human and artificial societies and, then, describe two of our computational studies that followed our proposed ecological perspective.

Bio

Eleni Nisioti is currently a researcher at the Flowers group of Inria, where she started in 2021 at a Post-Doctoral Research Visit position and, in 2023, moved to a Starting Research position. She is working on multi-agent learning systems, focusing on how skill acquisition through reinforcement learning and evolutionary algorithms is influenced by features of the environments that these systems inhabit, such as the uncertainty and diversity of ecological pressures. Before, she pursued a PhD at the University of Essex focusing on multi-agent reinforcement learning for communication networks.

https://moex.inria.fr/seminars/2023nisioti.html