It's not impossible, and it might even be plausible... But it's a really, really difficult question.
People outside of science: don't make the mistake of thinking that scientists haven't thought about this before! haha
People outside of science: don't make the mistake of thinking that scientists haven't thought about this before! haha
If you thought ML was difficult, try modeling a human being's digestive tube from mouth to anus down to the cell: welcome to a category of problems where climate and gravity are the "simple" ones.
On the other side of that space though, potentially the promise to increase by orders of magnitude our mastery of human condition both biologically and experientially.
The simplified concept of humans being one organism will continue to be taught nonetheless, because it's extremely useful, and not even wrong in most contexts in which it's applied.
There's nothing special about this, you can say the same thing about any simple model in biology: that the brain is an organ inside the head is a simplification, any diagram of a homeostatic system or metabolic pathway that fits on a single page is a simplification, the "central dogma" of DNA -> RNA -> protein is a simplification...
All of these things are well-known. They continue to be used as models, because, well... they're useful models. And there's nothing wrong with that!
I think to make an "AI" on par with simulating the brain, you would need to simulate the gut first. And probably vice versa!
You're right that it's unnecessary to emulate the brain down to its finest implementation details, down to the molecule or even down to the cell.
However, I contend that it's essentially impossible to create a "relatable AI" (an AI that behaves and thinks like humans do) without proper consideration of embodiment. A large part of why the brain works the way it does, at a macro level, emerges from the vehicle it's in, and broadly speaking both its afferents/inputs and efferents/outputs.