Earlier quoted context omitted.
I sympathise with your perspective, but surely some of those needs (colour theory, contrast, etc) can be solved algorithmically and others (public perception and brand "feel") could be automatically tested by a site/script surveying users in the target segment.
You need someone creative to generate the good ideas to be used in the first place. Random mutation and A/B would surely take millenia, even assuming huge traffic in the first place. And even then, one of the internal criticisms of A/B at Amazon was that it helped refine local optima (e.g. a next button in a checkout cart) but is weak on a holistic view (whole page, user experience), and making larger innovative jump…
Do you think that what's out there already could serve as a fairly good starting place? Run them past users to work out which out of thousands of logos are effective. Then isolate components (marks, text, etc) and test those, plus random combinations of the same.
I can appreciate the value in very high level brand creation, but I also think you can get 90% of the way there at 1% of the budget for many businesses. A remotely competent designer can slap down a shape and choose a font for the text and beat half the small business logos out there in a few minutes.