AI safety is doomed to fail | Roman Yampolskiy and Lex Fridman