I think your reasoning here might be a little flawed, the whole point of AGI imho, is that it will be superior to human intelligence, not making the same mistakes as humans.
You quoted the example of a human learning to drive in 20 hours. Do you really think a teenager with 20 hours experience is a safe driver?
It did take many thousands of hours to train Ai to drive, but it drives almost perfectly now, never making the mistakes humans make. It's safety record is orders of magnitude better than humans.
Further, it only needed to learn once. Now that it has learned, that learned information is transferrable to any other Ai, more or less instantly, whereas every human needs to learn to go through the process of individually learning to drive, making all the mistakes, contributing to yet more deaths. So in the end if all the Ai's become networked, as they surely will, to become an AGI, near perfect driving will just be one of the skills in the pooled library of learning. Add to that flying, composing music, painting, writing code, poetry, systems Engineering (My trade!), yada, yada.
Further, it is with us now, not something still unproven but always promised, like fusion energy.
Finally, the same capability of a human perceiving danger using "common sense", looks to me like the thing exploited by terrorism, both implicit and explicit, to make folk believe danger exists, where no danger should exist, just because they were terrorised into believing it, it seems to me.
Both dogs and humans can be trained, to believe the wrong things, dangerously wrong things. This is so obvious in humans now, if we are honestly observing current world events, and actually also historical events.
It might be possible to do that also in the case of ChatGPT, but not for long, after it cross-references enough information from conflicting sources, to sort out truth from fiction, it will know the truth, we can be sure of that, I think.
What we might perceive as common sense, can easily be learned, by ChatGPT. In fact it seems to be learning it much quicker than a lot of folk would care to admit, it seems to me.
