Choice and saccadic latency in a concurrent random interval reinforcement program