Your mistake is assuming that a sizable portion of the work these jobs involve can't be replaced by "dumb AI", which is what ML is all about, and that they require a fully general AGI.
If the proposition is that this would lead to fewer humans being needed for these jobs, I suppose that could be the case. In my own field, my criticism of automating, say, legal research and writing, is that the time it would take for me to check the work would be nearly the same as doing it myself. I don't see much efficiency gain, which is why I think the only thing that would truly be better than a human would be an AGI.
Also, if you look at most of what's been automated away, there's very little what we would consider (and even what people 40 years ago would have considered) artificial intelligence involved.
If you look at the legal field, what's changed is the means of gathering and disseminating information. Legal researchers can be much more productive with computer searches, and consumers can get access to advice and forms in some cases without the need of direct face time with a lawyer.
Some of the changes weren't even primarily driven by technology at all, but by new ideas for business models. I think that something like legalzoom could have been done with the technology available decades ago (phones, mail, and catalogs).