Sites Inria

Version française

Inria Awards 2018

Pierre-Yves Oudeyer : Inria – French Académie des Sciences Young Researcher Award

© Inria / G. Scagnelli

Can artificial intelligence help us to better understand natural intelligence? Pierre-Yves Oudeyer cultivates a vision of artificial intelligence that is as close as possible to the living world. A vision in which machines are at the service of humans. His guiding principle is looking to mechanisms of intrinsic motivation to enable humans and machines to progress in autonomous learning.

Pierre-Yves Oudeyer's early interest in artificial intelligence was above all for understanding human intelligence. He worked on the assumption that computers could be a valuable aid in deciphering brain functions, in the same way as computer climate simulations facilitate the work of geophysicists. A stance that led him, upon graduating from the École Nationale Supérieure (ENS) de Lyon, to join the Sony Computer Science Lab (CSL) in Paris, where, during a PhD directed by Luc Steels, he concentrated on searching for the origins of language and, more precisely, on modelling speech formation. “At the time, research into the origins of language was in full swing, mobilising a great number of people from different scientific backgrounds: linguists, biologists, philosophers, anthropologists, ethnologists and neuroscientists. But the idea of using algorithms to develop this work was still quite innovative ,” says the researcher.

When curiosity arouses curiosity

Over the course of 8 years spent at the CSL, Pierre-Yves Oudeyer gradually changed his focus, questioning what prompted our distant ancestors to explore the possibilities offered by the speech organs. “I then turned my attention to what the cognitive sciences call intrinsic motivation – a central mechanism of human sensorimotor development and of autonomous learning, more commonly referred to as ‘curiosity’. This is what encourages young children to experiment constantly in order to understand how their bodies work and what they can do with them. ” For the researcher, this is where his long project of creating algorithmic models of intrinsic motivation began, which he still carries out today with the Flowers project team (Inria Bordeaux Sud-Ouest) that he created in 2008. “One of my greatest satisfactions is seeing that ‘artificial curiosity’ has now become a real discipline that mobilises many scientists around the world, at the crossroads of artificial intelligence, neuroscience and psychology. A real community is forming, in particular through the collaboration that we have established with Jacqueline Gottlieb, who directs a cognitive neuroscience laboratory at Columbia University.

Motivating machines: a new approach to artificial intelligence?

Over time, Pierre-Yves Oudeyer's ‘playground’ has expanded considerably. As a result, part of Flowers team is now involved in the emerging field of developmental robotics, through the development of machines capable of learning repertoires of tasks more autonomously by applying the concept of ‘artificial curiosity’. The central idea is to overcome one of the main pitfalls of artificial intelligence as it stands currently, which lies in a hyperspecialisation of skills and a lack of flexibility. “Today, even the most powerful machine learning systems have very limited capacities! As such, AlphaGo would have no chance at all against a 5-year-old without the intervention of an engineer. An engineer would need to thoroughly reprogram parts of it so that it could incorporate the rules and objective of the game! We are working on designing machines that are capable of creating their own objectives and learning how to achieve them autonomously by generating their own intrinsic rewards. ” And the first results speak for themselves: two years ago, Flowers reached second place in the Demonstration Awards at the Neural Information Processing Systems (NIPS) Conference with their robot Poppy Torso, which can set its own goals, organise its own learning and control the complexity of the tasks it performs. “We are currently applying these same principles to a project in the field of biophysics, enabling machines to automate the discovery of new biophysical structures.”

From humans to machines... and back to humans

For several years, Pierre-Yves Oudeyer's team has been involved in a third field of activity that aims to combine artificial intelligence and intrinsic motivation to serve human learning. With projects such as KidLearn currently being tested in primary schools in Nouvelle Aquitaine on over 1,000 children, Flowers team members are endeavouring to develop truly stimulating educational applications that could offer bespoke, scalable exercises in order to continuously optimise children’s progress and make them increasingly curious and ‘efficient’, without however depriving them of their own flexibility. “The research has been well received in the world of education and we are creating a project with a consortium of companies to conduct large-scale experiments.

At a crossroads

Looking back over his young career, Pierre-Yves Oudeyer is proud of the projects he has accomplished – but also, and above all, of having contributed to bringing together two worlds that coexisted without really meeting. “In my early days at the CSL, the idea of developing research projects at the interface between cognitive and digital science was still unusual. Similarly, when I arrived at Inria and started talking about algorithms for human learning and the possibility of modelling intrinsic motivation mechanisms using machine learning techniques, the subject was far from straight-forward. But today, through efforts in education and outreach within the two fields, truly interdisciplinary approaches are more and more common and fluent. This is a major step forward and I hope to continue to nurture this dynamic.

Keywords: Inria Awards Pierre-Yves Oudeyer AI Young researcher award Machine learning Artificial intelligence