Depending on hiring filters, that might be true. If an application ever gets to an engineer though I'd bet on the last one or two winning.
I think a sensible organisation should skip the usual hiring filters in new fields, because [edit](my experience is in security) you can scoop up really good people who happen to have "unconventional" backgrounds if you have competent people evaluating them.
Regular HR people tend to do a really bad job with career switchers and the self taught since they mostly work off "signalling value". But in newer or very fast moving fields, the oddballs can be majority of decent applicants.
My limited experience is that very technical positions are not actually overwhelmed with applicants, and that it's not hard to evaluate if people have the right stuff because in these areas it's not difficult to devise quite objective challenges without resorting to shibboleths (guess what the interviewer is thinking or, do you come from the same technical culture as me).
Arguably ML positions should be the easiest to algorithmically hire for (at least for industry, not hard core research). Just put an automated judge with a fairly low bar on a business relevant objective function between your careers and your "submit job application" page :p
Personally I find this a little bit funny, because ML driven competence evaluation in "hard" (reasonably concrete objectives) fields should eventually render credentialism and signalling obsolete. But here we are, the $20k "certificate".
All these things said, a structured course of study is super useful for the undisciplined (certainly including me) and dropping "new car money" on something has a way of focusing the mind :)
Sometimes it's quite sensible to travel in the opposite direction to everybody else.
