Andrew Karpathy is the new leader of AI at Tesla. His specialization is in deep and reinforcement learning, especially the latter which is important for Tesla in connection with autonomous driving.
Reinforcement learning is a hot topic which is currently explored also by Google, Uber and Mobileye which was bought by Intel recently. With simulations cars shall learn by themselves how to react in difficult situations. An example is an intersection with many lanes: it is very difficult to formulate explicit rules which lane to chose. With reinforcement learning the cars will be trained to take the decision on their own based on the simulation results.
In this connection Karpathy mentions that simulations are quite expensive due to lack of data. Since Tesla actively collects data from the real world reinforcement learning can be made also economical.