Not a bad effort from The Guardian on a somewhat deep topic. It's sort of fabricating drama of course: if you only need predictions, then you use a neural network. If you want to understand mechanism, then you develop theory and use explicit stochastic models in a standard statistical framework. There's not actually any conflict there, but obviously lack of conflict doesn't sell newspapers.
Interesting. But a further objection is that larger datasets increase the need for hypotheses. In large datasets, nothing is obvious and researchers need theoretical starting points!