The article provides a testable protocol for models. Yet we're fundamentally limited to behavioral observation - consciousness cannot be proven from the outside by definition.

> consciousness cannot be proven from the outside by definition.

If it invents a concept of consciousness, I don't think a being without consciousness would invent a concept like consciousness to describe itself.

So when we have LLM zero that doesn't have the word "consciousness" and no human inputs and it still starts talking about consciousness then I think we can say it is conscious. Humans did this, so I'd expect an AGI to be able to do it as well if it is conscious.

In any case, AGI or not, the external observer still faces a choice - algorithmic artistry or genuine subjectivity.

Yeah, but it is very hard to argue if the machine figures out consciousness exists from first principles rather than from human input.

Since consciousness isn't observable the only way to realize the concept exists is to be conscious yourself, or have the concept explained to you.

This article is full of those, but some of them is a good food for thought.

Even more so the end of that abstract:

  "and then it can be diagnosed using specialized methods of subjectivity provocation."

Rather than debating terminology, why not test the diagnostic methods directly?