Actually, that experiment is measuring if people direct attention toward where OTHER people are looking. A face on a computer screen glanced in a particular direction, and then there was also an asterisk ("target") that the subject was supposed to stare at.
When the direction of gaze of the computer's "face" predicted the location of the target (i.e the target appeared on the gazed-at side most of the time), both autistics and typical subjects followed the gaze to help them find the target. On the other hand, when the gaze direction of the face was random with respect to the target's location, the autistics disregarded the face, whereas the typical subjects still followed it.
However, the other phenomenon (not thinking about what you're looking at) is also very true for me sometimes (particularly when what I'm looking at is not interesting).