“The potential of reinforcement learning (RL) in AI is not even close to being fully tapped.”