Some ideas about reinforcement learning
Stuart Russell              

Reinforcement learning is the process of modifying behaviour to optimize
a reward signal. Research on this topic has the potential to increase
our understanding of how real organisms learn. I will present three
ideas that may help in this regard: 1) that partial,
structural constraints on behaviour may help in scaling reinforcement
learning to more complex tasks; 2) that the reward signals optimized by
real organisms may be inferred from observation of behaviour; and 3)
that noise in biological control systems may have an important
influence on the control strategies employed.