Unknowable Unknowns
Dan Geer · IEEE Security & Privacy · 2019
Addresses the concept of "knowable unknowns" and discusses this concept in the context of security models driven by artificial intelligence (AI). Kurt Gödel proved that there are problems for which it is impossible to construct an algorithm that always leads to a correct yes-or-no answer; those problems are undecidable. Alan Turing proved that the halting problem is undecidable in Turing machines. Alfred Tarski proved that truth in the standard model of a system cannot be defined within that system. Olav Lysne proved that it is not possible to verify electronic equipment procured from untrusted vendors, and that a vendor cannot build a system that supports verification by untrusting customers. Ben-David et al. proved that scenarios exist where learnability can neither be proved nor refuted. Finally, Donald Rumsfeld made commonplace the phrase unknown unknowns. And so we come to artificial intelligence