Before handing decisions (even about how to treat individual humans) to AIs, the previous step is to tell the human officials or service representatives to follow rules like AIs. This is already common and we can see how badly it malfunctions.