The speaker is more concerned about the unintended consequences of poorly optimized objectives when using AI for safety critical systems like cars and aircraft, as opposed to the nefarious use of AI by humans, which has historically been the bigger problem.