Not only do I know how it works, but I'm also a photographer (who has edited using similar techniques manually) and work as a developer and read a lot about computational photography so I probably understand how it works on a deeper level than most people.
Apparently you don't know how it works. I don't usually call people out like this, but you called me out on this without even knowing the facts yourself. I'll allow Apple to explain:
Emphasis is mine. They use ML for everything. I'm sure that in the parts that have movement, which they discard, they are applying ML to fill in the details and de-noise so those parts don't look too weird from being boosted. Basically the neural network has enough good data from the parts that aren't moving that it can bring up the parts that are moving (and therefore don't have enough usable exposure info to do noise stacking or expose for shadow detail) to make it look natural with the rest of the scene. But I wouldn't be surprised if you pixel peeped moving objects if they don't look quite as good up close. On Instagram though? Perfect.
People truly underestimate the power of computational photography. It's a game changer to the fullest extent of that phrase. They are doing things on the fly that might take an expert dozens of minutes to hours to get right with careful preparation and editing. And they do it in a fraction of a second. It's nuts. I can't believe how many people I saw who were upset that they weren't adding this for the XS. It's a lot of processing power! Good thing the iPhone 11 Pro has much longer battery life and faster CPU and neural engine.
