Dear Ogogomad - as you will see, the article is about AlphaGo, not AlphaZero. At the very basic level described in the article MCTS in AlphaGo and AlphaZero function in the same way, the only difference is where the proposal policy comes from. The article also had the benefit of expert input from assistant professor in ML Dr Aitchison. http://www.gatsby.ucl.ac.uk/~laurence/

I don't think the article author understands either machine learning or MCTS. It's not as simple as "it plays a bunch of games against itself".

Additionally, AlphaZero does not get any input from grandmaster games.