Interesting article, although the basic thesis is not new. One way to view the application of machine learning to science is simply that ML automates part of t

Interesting article, although the basic thesis is not new. One way to view the application of machine learning to science is simply that ML automates part of the scientific method: inducing plausible hypotheses from the data. I don't see how the author concludes that "correlation is enough" or "we don't need models": the whole point of machine learning is to develop a model that explains the available observations and is useful for predicting future observations. Just because those models are expressed mathematically and derived algorithmically doesn't mean we don't need a model at all.