Besides, Keen (and other postkeynesians) truly shows how the mathematical economics should be done, that is, by having a fully dynamic macro models with time and debt/money. I am afraid machine learning the article talks about is a similar distraction as econometrics. Unless mainstream economists will be willing to rework the theoretical foundations (give up on equilibrium, at least), and base it more on empiricism, mathematical approach, no matter how fancy, will always be a quandary.
But as other have said, the problem is that many economists are pretty much the prophets of the free market (or status quo in general). If you want to somehow claim a moral basis of society, then you can't have a model which allows multiple different solutions, and so you need to stick to equilibrium instead of allowing a dynamical solutions.