The simplest case is just to pick the most central square, and then you could probably improve that by picking a standard square according the rule of thirds. Those are the naive algorithms - this choice of alternative algorithm is deliberate and isn't as naive or simplistic as you're claiming.

The algorithm is only considered useful because it appears to do better than that, on whatever examples that the developers tried (i.e. there was a business case for using it), and against other possible code.

Including, likely, pictures of their own selves. That's certainly what I would test it on, until it vaguely worked.

What 'awful lot of assumptions' do you think I am making? I don't imagine we are in disagreement about this.