In other words stuff that requires some knowledge and physical dexterity to get the job done is too complicated or expensive to replace. Not so in a low product liability environments where AI or more aptly put "expert systems," will replace a human much like the menu tree on a telephone directory did years ago. Bad advice and steered outcomes will always be the purview of the folks who pay to have somebody write the code. Flipping burgers and taking drive through orders... soon to be a memory. So will elementary triage, configuring some widget for your home stereo etc., etc., etc. Just because somebody slaps an AI label on it doesn't mean it will be accurate or free from agenda driven "nudges." AI in high threat domains... now that's the scary part. Oopsies in that world have can have some fatal consequences. Maybe intentional, accidental or deniable, but none the less very scary given the ethics and morals of the modern world.
Where's the legal framework? Yeah, that's scary too.