This is simply not true, and there are many tasks the architecture would fail with and the networks presented are architecturally different from the highest performing vision networks like GoogLeNet. The techniques behind AlphaGo have no memory component holding state between moves, such as a hidden state in an RNN. It is completely reactionary.

A simple game that this architecture would fail at is Simon [0], where you are presented a sequence and then are tasked to replay the sequence.

[0] https://en.wikipedia.org/wiki/Simon_(game)