Could we reasonably expect an AI to something right 100% if a human could do it with 100%?

Could you tell if someone has down syndrome pretty obviously?

Maybe some kind of feature exists that we aren't aware of