I think your expectations are a bit high! You are expecting the camera to:
a) identify a face -- yep, fair enough that's what face recognition is for
b) identify specific features, of which most faces have two
c) work out which one you want to focus on
d) shrink the focus spot to small enough to only cover the area of one eye
e) centre the focus spot on that eye -- which might very well not lie on one of the manually selectable spots.
All of that with wafer-thin DOF due to the selected aperture.
Sounds a bit demanding to me.