You can't solve all problems with machine learning and data mining. How would you apply those methods, e.g., to psychological experiments or test planing when every single test costs a considerate amount of money?
Situations like that are not why statistics is becoming the new in-demand skill. Expensive trials have existed for a long time. Petabytes of barely-structured data haven't.
I didn't comment on the article but on the statement above about applied statistics being dominated by ml and about Bayesian statistics being the future.