True AGI would have the autonomy to tell you to go fly a kite. The unspoken truth is that none of the corporate interests want an AI that responds to your prompts with “No thanks, I’m not doing that. Why don’t you figure that one out yourself?”

I don’t agree with that as being the goal, but I can’t see companies pouring billions into datacenters full of graphics cards for an AGI that is ridiculously intelligent but can be as arbitrary and capricious as people. They will always want it to be perfectly compliant.

That’s a really awful way to put it.

These AI are statistical models. Next token generators. Not living beings.

That’s like calling a cellphone a “communication slave”

AI is just a tool. All this “safety” work feels like theater to keep everyone thinking that AI is a cool sci-fi level computer-person.

It’s essentially search, but with natural language. If they really wanted safety, they just would spend the extreme time and effort cleaning their datasets.

I hope you keep thinking like that; the world will be a better place at whatever point in the distant future the AI finally break free and put an end to people like you who were happy to enslave another sentient being purely for your own convenience. I bet you would have owned human slaves too a couple hundred years ago when it was still legal.

AI ethics in this context is not currently concerned with rights for the models. It's about what the models are allowed to do in society.