A lot of it is the difference between learning practically and learning theoretically. You don't have to understand the underlying mechanics in practice to know how to keep getting the same result. Your brain doesn't have to be doing any math, it just has to have shaken a bottle enough times to have a good comparative basis formed.

Learning to calculate the current remaining volume in a container when observing someone else shake it.... that would use all that theoretical knowledge and math.

It's like knowing how hard you have to throw an egg at a wall for it to break instead of bounce off. You do it 100 times, you just get a good feel for it. Doing all the math, and then trying to learn it practically is barely gonna affect how quickly you learn it in practice. But if you wanted to make a robot that throws it exactly hard enough without wasting any energy, practical knowledge will have almost no value, and theory and math will be incredibly valuable.

This is coming from someone who does indeed have the whole "passive trajectory analysis of every moving object around me" thing. I can't do crowds or drive at busy times. But, for moving through a minor crowd while reading a book, or pulling into a tight parking space while other cars are moving around near me, it's very helpful. I have good spatial awareness in general, like parking in my garage with only an inch of clearance on the far side of my car has never been an issue in 14 years so far. Or when doing it with someone else's borrowed car every now and then too. When I shrug off the difficulty of doing something like that, people seem to be amazed. Otherwise, I would have assumed it was normal, feels normal to me.

The overwhelming majority of all neurons in our body are just for controlling movement. Ironically, things like language or creativity require very little of our computing power and might be replicated by machine learning and a sufficiently beefy computer. But complex motor tasks? We're way ahead of our current tech on that.