Machine Learning is being used to drive scientific process. Is it changing how science works?
The classical view is that science proceeds by three steps: hypothesise, predict and test. A recent article in the Guardian questions if that process is being superseded by AI. The main examples it cites are AlphaFold (the recent solution to the protein folding problem) and some examples from psychology. Big Data can now be plugged into black box algorithms that produce results without understanding how we got the answers.
Generally, I don't buy it. I see ML as simply another tool, much like a microscope or a large hadron collier, that allows scientists to have greater capabilities and to test larger theories. AlphaFold didn't provide a set of new theory free predictions, it just meant that scientists could avoid doing thousands of laborious slow experiments to determine the structure of each protein. Scientists still have to work out what to do with those structures. There is still very much the need for humans in the loop to decide where to point those tools, what sort of data to give them to train them on and how to interpret the results.