Abstract
Reinforcement learning is one of the major strands of current computational intelligence: it is used to enable an agent to explore an environment in order to ascertain the best actions in that environment. Genetic programming is a method to evolve programs and given the similarity between genetic algorithms and reinforcement learning, it is perhaps surprising that so little attention has been given to using reinforcement learning to identify useful programs. This paper makes a start on this task by investigating using reinforcement learning methods for function approximation.