Actually, I would argue that they aren’t.
First of all, all of the discussions of AI art are specifically talking about a particular type of AI: generative AI.
There are other types of AI in development and in current use which could make your 3rd statement come true (if only partially).
I don’t think AI will bootstrap itself, as all current AI isn’t really intelligent at all. But I can see other ways we can destroy ourselves by relying too heavily on AI. In fact I think one of the dangers is that AI can appear to be intelligent, and that leads people to trust it when they should not.
Note that there are, already, AI driven Kamakazi drones being used on the battle fields of Eastern Europe.
Then, regarding generative AI in particular, I see the possibility of the technology reaching a functional asymptote and the potential degradation of output quality as being two sides of the same coin.
In a way, they are different vectors: the quality of generated content is largely driven by the quality of the data being used for training, whilst the ability to use that training data is a function of the development of the technology…
…and, with current LLM technology, how much computing power, electrical power, and clean water you are prepared to throw at it, and how much pollution you are prepared to create.
The development can plateau as it has with LLM, and the quality of the training data can degrade at the same time. I don’t see these as mutually exclusive.
Cheers,
Keith