The concept that GPT works by averaging what humans say (and write) and then representing that average on cue is not too far from the way 99% of humans think. There is some subjectivity as we filter the data to have our average, and that filter comes from education and parenting and anything we have been influenced by. Still, the principle stays very close to the GPT way, the so-called thinking is just what we say in our heads and sounds right (in harmony with our average).
The filter in the case of GPT is implemented by “fine-tuning”. That is a human intervention in order to remove the extreme positions and introduce civilized value into the system (GPT model). The process is, even unreliable, the only known way to filter data in order to mitigate biases naturally occurring in the bulk of human knowledge.
The point is, the deep understanding of the matter we are discussing is primarily an illusion. More than 99% of us have a very superficial (if any) understanding of what we are talking about. It is sufficient something to be intuitively right for us in order to declare to the world that this is our position and depending on the situation we are ready to make necessary sacrifices in order to defend it.
On the matter of GPT, who is to say if it is intelligent or only looks that way – the average Joe or Jane? For the mentioned 99%, GPT is intelligent because their intelligence works on a very similar principle. “Fake it ‘till you make it” is a surprisingly popular mode of interacting with the world, so if not now, the next iteration of GPT will fit the bill.
The upcoming GPT-4 (or any other, with similar power) will be intelligent to practically everybody except for a tiny (and vanishing) group of people (aka geniuses). Is there any danger? Yes, any new technology brings us new kinds of “atomic bombs” if maliciously applied, but we do it because of the many benefits of the same technologies. AI revolution will be the last human revolution. After the singularity, the torch of new knowledge (and eventual revolutions) will be carried by AI. Can we do something about this? Considering our history as species so far, – not really. Mitigation of some AI misuse, – maybe.
By the end of the century, the dominant species on earth will be no longer recognizable as humans from a contemporary perspective. A more optimistic scenario is some form of symbiosis between humans and AI, a less optimistic one would be: AI rules and humans (if still around) just observing in disbelief and anger.