AI's risk to humanity: If we can't understand it, can we man