World 1: 40,000 people die a year from traffic accidents, but the driver is always a person. The courts may or may not find them liable, but the victims’ families at least have a name and face, and an explanation of events and motivations that they can rely on to find peace.
World 2: 4,000 people die a year from traffic accidents, but the driver is always an opaque ML model. Nobody is named, let alone found liable. And even if you had a name, there is nobody who can explain the reasoning of the model to you.
I am not sure World 2 is unquestionably preferable.
I think we’re missing something about justice when we only look at how many people die and not also how the survivors are able to make sense (or not) of their loss.
World 1: 40,000 people die a year from traffic accidents, but the driver is always a person. The courts may or may not find them liable, but the victims’ families at least have a name and face, and an explanation of events and motivations that they can rely on to find peace.
World 2: 4,000 people die a year from traffic accidents, but the driver is always an opaque ML model. Nobody is named, let alone found liable. And even if you had a name, there is nobody who can explain the reasoning of the model to you.
I am not sure World 2 is unquestionably preferable.
I think we’re missing something about justice when we only look at how many people die and not also how the survivors are able to make sense (or not) of their loss.