... yet. The next AI Winter is right around the corner. While there has been seriously amazing work coming out of ML, in industry there is an insane amount of resources being put into completely BS ML systems. Outside of FAANG there are many "fancy" ML projects in industry that are just continually accruing massive technical debt, and for no good reason. The number of deep neural networks being used for problems that can be solved with radically much more simple (and easier to debug and maintain) techniques is astounding. Right now there is a boom in complex and fragile a systems made by teams of PhD in non-engineering fields that don't know anything about system or software design.
In a few years these complex systems will need to be maintained and fixed and it will quickly become apparent that the vast majority of these projects are a huge business liability with minimal financial or engineering benefit. Expect a huge back lash against predictions coming from complex, non-interpretable linear algebra heaps.
The pendulum will swing back, and likely pretty hard (as always probably too hard). I'm not sure theorem provers will be the winner, but the general class of understandable, predicable, and explainable things will become more popular. It is quite possible that "my tool proves things are stable and work" will be a big sell in the next wave of tech.