For a sufficiently complex system, working out whether it is conscious can be difficult.

This is not at all true with current LLMs. The transformers architecture doesn’t even have real memory - the neural networks are all feed forward.