The price of errors is too high, I'd imagine. I see this everywhere in my own domain - AI stuff is super easy to demo and prototype, but running in production i

The price of errors is too high, I'd imagine. I see this everywhere in my own domain - AI stuff is super easy to demo and prototype, but running in production in scenarios where errors are not cheap is a recipe for disasters. We are currently only pushing it where the existing process involving people is ridden with imperfections -- and AI can do no worse or better, but still not perfect.