The idea of A.I. monitoring other A.I. for safety and alignment may feel uncomfortable, but “we’re going to have to go there,” argues the computer science professor Arvind Narayanan on “The Ezra Klein Show.”