17. Policy gradients are more effective than Q-learning in continuous action spaces.