I don't think there's a fundamental reason why you couldn't program AI digitally. Maybe there's some high level reason why it needs analog processing but I doubt it.
ML models use floating point numbers which approximate continuous values. An analog computer like you are describing could maybe speed up those calculations but it wouldn't fix the fact that ML models just can't be intelligent because of how they work (in my opinion).
I agree with the author of the article but I'm curious if there is any well understood model of biological intelligence that we could use to say whether an artifical intelligence system has parallels to it or not.