Human drivers: people die due to reckless driving and human mistakes - and in most situations a good driver can anticipate a lot of dangers and avoid them. ML "assistants" getting control over cars: people die of unpredictable mis-assessments of inputs vs. outputs, and there's no way to anticipate any danger or try to avoid it.

People who advocate for non-deterministic systems taking control in cars are irresponsible and have blood on their hands.