“The AlphaGo approach does not scale to problems with large action spaces.”