Human operant learning under concurrent reinforcement of response variability