It was actually worse than that in the Air France case -- if the rudder inputs from the two pilots were different, they were averaged before being applied to the physical control surfaces.
It could work both ways: either with one pilot in control or with both.
The averaging thing was not as stupid as it sounds when values are close.
It's completely crazy when both rudders were in the opposite direction at the maximum (average of -100,+100 = 0), but it must have seemed an impossible case to the designers...
