I'm not sure if I get it. Data has always been used to generate observations about the world, which is the starting point for the scientific method. I'm reminded of how Kepler used volumes of data to figure out the planetary orbits.

http://csep10.phys.utk.edu/astr161/lect/history/kepler.html

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.

The comments show that the article author presents a false dichotomy. All the data collection and analysis depends on models. Then from that analysis you get new models and do new analyses. This is how science works, whether you have tons and tons of data or not.

Is there a hyperbole hall of fame?

Nice to see Wired dialing back up the naive techno-triumphalism. How long until the next bubble bursts?

The author of that article obviously doesn't understand the scientific method. Sad.

In what way?