In current software parlance, this is often used in stupidly trivial ways, but digital twins have a long and important history and function
I wholeheartedly agree that it's a safe bet that very very few digital twin projects achieve anything close to what they propose. More typically it is a flashy label to stun management. I actually worked on digital twins -- I came into ML via physics -- and once you start digging, you find the rabbit hole is deep enough that absent massive government grants or angel corporate investors you will never get anywhere close to what could be a called a digital twin.
Not that there's any issue with lofty goals.
Note that I'm not directing this comment specificaly at the authors of these papers (I haven't read it, just skimmed through it). Just observations from experience.
E.g. you can and will sync properties to the digital twin, and back. Obviously this does not work for animals (yet?...) but the digital twin of a car can have changes that are then propagated back to the original.
I don't get any of this in any definition of "model" or "simulation" I know.
AI overloard controls your every existence until you are killed.