I agree with the skepticism in the comments but one area where I think ML does have a lot of potential in scientific discovery is the field of SciML and Symbolic Regression.

The first uses physics-informed neural networks in systems of ODEs to solve complex problems faster and even helps derive DEs that fit some data.

The second uses a fixed set of symbols to fit a tree of them to data. This gives a closed-form solution that best fits the data. Both are pretty hot in the Julia community and have results published