I just want to state this as context, not as an invitation to tangent into an argument: I think general purpose AI is going to fade from consumer software again, as it has so many times before. But I suspect that some of the tools may find a home in areas where all of the problems are multivariate, and things like advanced techniques in linear algebra can help find signals in the noise when you can't control an environment.
I recall years ago someone discovering that chemo works better on an empty stomach, and not just for the obvious reason of not having anything to throw up. Normal cells in "starvation mode" absorb toxins slower, while many tumors ignore this signal. If we can nail down things like "If you have these genes and your serum vitamin D is > 120 and you fast for >8 hours and ingest 20-40 mg of caffeine an hour before infusion, tumor shrinking is increased by 40%" by mining through mountains of telemetry and then working backward from there to find the causal link.