The salient question is why anyone would ever think a system producing output based on existing statistic distributions of fragments of text would be a reliable source of truth. That's the real enigma.