AI Ethics: Machines That Do Not Decide

A model does not 'choose' — it draws samples from a probability distribution. When we hold it responsible like a moral subject, what exactly are we doing?