I think it applies to most STEM fields. I'm a reviewer for several journals in a STEM field (not AI/ML specifically, but some manuscripts do try to apply AI/ML to this field) and the vast majority of authors seem to upload their preprints to arxiv etc.
Social sciences may be behind though as you say, I do not know as I'm not in that field.
As someone with graduate degrees in both STEM (math) and social science (psychology) fields, it's true that social science is way behind STEM in terms of preprints to digital archives. It's possible there's momentum here in the last 5 or so years that I'm unaware of though.
That matters for a few reasons:
(1) the average person more frequently encounters psychological, social and medical issues more than they do math problems. And since the research in those fields tends to be pay walled, people are at the mercy of things like SEO spam medical and health sites.
(2) wrong ideas in medicine or psychology can (and have in the past) damaged entire generations of people. So in that sense their blast radius can be very large. This means that peer review is especially important and that there's a potential negative externality to posting preprints and drafts before they're finalized. I suspect we'd have to solve the peer review and quality problem before STEM-style preprint archives become the norms in all fields.
For Bioengineering it definetly is, but a lot of Medicine is still locked behind high impact Wiley and Sage publications, and for a lot of that research it's fairly easy to pay the $3-4k to make the article open access.
