Sure, you can put the computer running the mind that controls the body into the body itself, and there might even be good reasons to given spotty wifi or mobile data, but you don't have to.
Also: why embody them in human form at all?
https://www.bostondynamics.com/products/spot
https://www.parrot.com/us/partner-ecosystem (FWIW I've never heard of these before, I just searched for AI drones…)
You are trying to read things in my comment that are not there. It's _not_ about destroying the actual machine. It's about drawing a line.
Describing an AI walking the streets, or smashing it to pieces, necessarily carries with it the implication that the AI is located within a physical body that moves on legs.
And those words you used are very explicitly about destroying an actual machine.
But if you wish to claim that was merely a linguistic turn of phrase not meant to be taken as an implied literal belief of the location of thought? Ok, but it remains the case that what you propose against a "sufficiently advanced" AI, is somewhere between petty vandalism and a pogrom, depending on questions of if "sufficiently advanced" comes as a package deal with "sentience".
Which is a totally different, and in the latter case much more severe, problem.
I can play word games like saying "consciousness" is the opposite of "unconsciousness" (or alternatively and IMO not entirely compatibility that it is the opposite of "subconsciousness"), but every time I've looked at this, it's either ended with (1) circular definitions, (2) definitions that accidentally include VHS players connected to TV set, or (3) things that humans regularly fail at.
None of that actually helps figure out whether or not a machine does or doesn't have that which makes us have the experience of being.
Getting it wrong, in either direction, is a Bad Ending.
https://kitsunesoftware.wordpress.com/2022/06/18/lamda-turin...