Why? They're flat and simple because (if you buy into the flat movement's sentiment) we don't need skeumorphism to distinguish a button anymore, but when you press it it does something 'because feedback'.

It doesn't really matter what it does, we just need to know we pressed it, so why shouldn't the thing it does be 'go a bit 3d'?

I noticed that as well... it's pretty visually awkward, going from no button-like affordances to suddenly having them appear during the click event.