Not picking on any of the people in this thread because this is the default attitude I've encountered, but it's still bonkers. It's fairly easy to convince people that all our selection processes (hiring, grad school etc.) have ~0 predictive power. But then they insist that we should nevertheless put on an elaborate, costly, and harmful ritual of pretending to identify superior candidates, because they find the alternative of switching to a partial lottery to be simply unthinkable. What?! Explicitly randomized admissions are vastly preferable to processes that ultimately result in arbitrary decisions (at best).
This is a key point we make in the AI Snake Oil book β the demand for predictive AI in areas like hiring arises because of our deep, seemingly hardwired discomfort with randomness. When I'm asked about the impact I hope our work will have, moving the needle even an iota on our attitudes towards prediction and randomization is at the top of my list.