Hi! I’m Cédric Colas, AI researcher and digital artist.
The title is: Towards Vygotskian Autotelic Agents: Learning Skills with Goals, Language and Intrinsically Motivated Deep Reinforcement Learning (manuscript, defense). In short, I’m interested in the study of artificial open-ended skill discovery, the ability for an artificial agent to grow an open-ended repertoire of skills along its life. This leads me to focus on intrinsically motivated agents that sets their own goals, a.k.a. autotelic agents.
I’m also convinced that to develop such abilities, agents will need to be immersed into rich socio-cultural worlds, to interact with their peers and with humans, to participate in a shared cultural evolution, with us. This idea is developed in our recent perspective article.
I’m currently working with Joshua Tenenbaum at MIT, where I look into how program synthesis methods could help intrinsically motivated agents to grow open-ended repertoires of skills in partnership with humans.
I often think of ways to reuse the algorithms I encounter in my research for alternative, mostly useless purposes. This leads to the projects presented here.
The website robot logo is modified from a Flaticon logo, created by Slidesgo.