Even in the 5% of "vaguely useful" outcomes: How many inefficient "common nonsense" decisions is AI going to make that cost human time and decimate profit, outcomes and workers? That measly 5% still isn't actually any good. Its 95% shit. 5% "technically edible" shit 100% wasteful
The most average of human design is already fairly problematic in terms of not wasting human time. Handing it to a machine can only be worse. It can only waste exponentially more untold billions of human hours.
We are undoing the computer revolution made possible by people like Douglas Engelbart and Ted Nelson. We are deciding instead of paying for studies which would improve the computer and produce enormous gains for everybody we are defunding users for a glorified coin-flipper worse than any human.
The only time these algorithms make sense is if the output must be completely indistinguishable from noise that "superficially resembles a thing" -- at which point a noise algorithm is exponentially cheaper and a bespoke procedural generator cheaper and better still.