As an artist, looking at a real 3d, flesh-and-blood person, you move around. They move around. Moving your head just a foot in one direction can complete obscure a feature that would otherwise get drawn (such as a hand that's now behind, rather than poking out from behind, their torso). So as you draw, you end up drawing from multiple different perspectives - oops.
From what I've been told, but what I also can vaguely confirm from my own experience, that's very much the point. Live drawing, and more specifically that of the human body, offers a greater (the greatest possible) bandwidth of information, not merely visual, but about the parameters of anatomical configurations if you will (preferably doing it more than once in one session with the same model but different povs and poses, as well as with different models over time), that are sucked up in the process as well. It's a sort of plastic understanding, the endorsement of which is more limited in 2-dimensional photographic referencing, and is more.. effective in the learning experience so to speak. In the courses I've attended they also emphasized completing volumetrically distinguishable segments/their contours in as little strokes as possible, so as to give the brain the opportunity to memorize the shape data as opposed to a series of pen-strokes rather (as well as clinging to blocking, i.e. starting out rough, going into detail iteratively). Which is very prone to unrewarding, deformed outcomes at first, but supposedly is a very useful step to make in getting gestalt on your side.
In exercise you're never under the obligation to deliver perfect results, in fact that's paradoxical. Using static references, on the other hand, supports this certain tendency to perfectionism and the fear of making "mistakes", which in learning and progression is not exactly a good thing. I believe.