Why Do Keynote Speakers Keep Suggesting That Improving Security Is Possible?

TL;DR: an in-depth talk by tenured Harvard professor James Mickens describing why unexplainable ML models will always be a security risk.