We do not need to "replicate" human intelligence. It's enough to "simulate" it. The coming AI models will be entirely "alien" types of intelligence to us, and that's OK as long as they are useful. Most likely these AI models will be able to finally explain to us how our own brains work.
That’s not to say that it’s not useful in its current form, because it is, but that’s not what we are arguing here. You asserted that “Soon it will reach levels of complexity vastly exceeding any human brain.” while I’m asserting that the human brain is far, far more complex than you give it credit for and that the latest AI models are not even close to that yet. I am doubting that LLM’s and similar machine learning will exceed human brain complexity, and certainly not VASTLY, in any timeframe that I would call “soon”, unless there are fundamental discoveries made. It’s not even clear the current models can scale any further due to the lack of any new non-LLM-generated training data to train bigger models on, so the GPT3.5 to 4 shift is unlikely to reoccur.
It could be hubris to assume we know enough yet to replicate our 'kind' of intelligence. Just as it might be hubris to assert AI doesn't have some 'kind' of experience. A minimum (which will be raised once we have a better understanding) is still: change of state, to have an experience - we simply don't know what else exactly is required, so that threshold minimum remains to be raised.