@Fiddle
The true problem is, that this AI is being trained to act immorally, and think in "feel good" shit. But researchers won't know that by the time, it is used to grow food.
And when it will be deployed, "let them eat cake" will seem like a genius idea when compared to what the machine will cause.
I don't think AI will ever be functionally deployed, but what I do agree is that when it gets deployed it'll wreck shit by virtue of simply lacking the logic needed to accomplish its objective, like a profiling AI who suddenly looses the ability to distinguish between races, sexes and skill sets.
Another thing I never considered is the idea of an "immoral" AI because that's basically a weapon and nobody would know about it until it goes wrong, it never occurred to me but I'm hyped