… and fails at that goal by about the same margin, as you as you pointed out last year.
I attended an amusing private conference on AI in finance a few years ago. The commercial speaker gave a fluffy and meaningless talk, but the two high-profile professors who followed spent three hours talking almost exclusively about (high profile) failures and why you shouldn’t use “AI” for many problems. I think the organizer was suitably horrified; the next year’s conference was pure fluff.
I’m personally in the field, and making “AI” work with statistical rigour and on data which is assumed to be bad from the outset is fairly challenging. I spend a not inconsiderable amount of time deflecting inquiries into whether the approaches would work for a given application… usually it’s a “no”.
Although perhaps if I said “yes” more often I’d make more money 
…