The study of morality for self-driving cars, using the trolly problem.

Paperclip maximizer would suggest the ai will purposely kill undesirables to “save” others because no option to avoid harm in the situation was presented in the training data.