Probably assigns confidence scores and then rotates with least and most confident answers to the same question to try to identify an object

Could be, in my experience it's always a traffic light or a fucking bus, and the images many times are similar, so not sure it's doing anything too advanced