[tensorturtle]

On the whole I agree, but having read "Reinforcement Learning: An Introduction" by him and Barto, this article comes across as a hardly nuanced self-endorsement of RL as "the inevitable future of AI". Without mentioning RL by name (but hinting at it with HMM, search&learning- like exploration&exploitation), I think he might be suggesting that any supervised learning is still too specific. Not to mention that he works for DeepMind, which has found fruitful applications of RL.