Title: Learning in the Development of Action

The central problem for motor control is adaptation to variable and  novel conditions. Movements cannot be performed in the same way over  and over because possibilities for action are always changing.  Behavioral flexibility is imperative so that motor decisions can be  geared to the current constraints of the body and the environment.  This presentation describes developmental changes in behavioral  flexibility as infants learn to sit, crawl, cruise, and walk. Each  posture operates like a separate balance control system. What infants  learn in an earlier developing posture does not transfer to a later  developing one. However, within postures, infants acquire learning  sets that promote tremendous flexibility in response to variable and  novel motor problems.

Title: Intrinsic Motivation, Cumulative Learning, and Computational  Reinforcement Learning

Motivation refers to processes that influence the arousal, strength,  and direction of behavior.  Psychologists distinguish between  extrinsic motivation, which means doing something because of some  specific rewarding outcome, and intrinsic motivation, which refers to  doing something because it is inherently enjoyable. Intrinsic  motivation leads organisms to engage in exploration, play, and other  behavior driven by curiosity in the absence of externally-supplied  rewards.  Intrinsically motivated learning has long been viewed as  essential for the cumulative development of an agent's competence in  interacting with the world.  In this talk, I review some of the  research directed toward developing intrinsically motivated  learning  systems, which is not at all a new idea though it is receiving  increasing attention. I focus in particular on how to design  intrinsically motivated reinforcement learning systems.  I discuss  five themes that stand out as being important: 1) the distinction  between a reinforcement learning agent and its environment at the  base of the computational reinforcement learning framework has to be  looked at in the right way;  2) internal state components that  influence intrinsic reward include a robot's memories, beliefs, value  function, and policy in addition to vegetative features like battery  and dust bin levels;  3) a guiding principle is that the learning and  behavior generation processes "don't care" if the reward signals are  intrinsic or extrinsic; the same processes can be used for both;  4)  the dividends paid by intrinsically motivated reinforcement learning  accrue over multiple specific tasks faced over extended periods of  time; and 5) intrinsically motivated reinforcement learning is a good  way to produce behavioral modularity that is essential for cumulative  learning.

Title: Developing Self-Consciousness and Values

As a species, we develop a special kind of self awareness, namely the evaluative sense of who we are in relation to others. In this presentation, I discuss the development of self-consciousness in children, viewing it as the by-product of the construction of shared values with others. I  present recent data on the emergence of negotiation and sharing propensities in young children growing up in various, highly contrasted cultural environments. My goal is to raise the question of what it would take to  build a developing robot, a robot that would simulate the child in its psychological growth, a machine that would be conscious of itself, ready to reciprocate and to give with no short-term reward or self-gratification.

Title: The typical brain, the autistic brain, and the behaviour of others.

How do we perceive and process the behaviour of others ? What are the pertinent informations that our brain needs to analyze accurately in order to trigger adequate socio-emotional behaviour? How is it possible to live if one is blind to these informations or if one is not able to use it properly? In this talk I will present a number of neurobiological data addressing those questions both in typical and high functioning autistic adult populations. I will then discuss how these data support theories about how our brain decodes the behaviour of others. The goal will be to appreciate the implications for research on robotics and show how complex social robots could be useful as a tool to help autistic children to develop alternative cognitive strategies.