1 Comment

Good post. But I think you'll lose most AI sceptics near the start. A lot of people don't think that an AI will ever be "like a human", and current AIs like ChatGPT don't do much to change that opinion. So I think you'll lose a bunch of people by starting off "imagine a person who doesn't need X or Y and who can do A and B".

I think your strongest point is the power-seeking argument. But even that is only scary if you think *one* AI is going to *successfully* acquire all the power too fast for anyone to do anything about it. If everyone has their own AI trying to get them coffee, then we just end up in a world that's different from now in the same way that "everybody has a computer" is different from "nobody has a computer".

You're skipping over various steps that are needed to convince people of the difference between AI as a useful tool and AI as a nefarious backstabbing schemer. And modern LLMs seem to be quite good at being realistic, and if the user asks for something unrealistic they're likely to point it out.

Expand full comment