Imagine sitting in a coffee shop, having a nice conversation with your friend Dave. If Dave looks at something, your eyes will reflexively move to look at the same item. This is actually quite convenient, because it may help you figure out what Dave is talking about, or what he might comment on next. How much of this joint attention reflex depends on Dave's face? You'll do this even if he only moves his eyes, without turning his head, so it might be that what you're reacting to isn't so much his face, but the movement of his pupils.
What we know about certain brain mechanisms makes this question even more interesting: different regions of our brain are involved with processing faces (areas in the temporal lobe) and directing spatial attention (areas in the parietal lobe). In fact, face processing is so specialized there are specific sites within the right temporal lobe just for recognizing and identifying upright faces. It's much harder to recognize an upside down face, in part because the specialized area of the right hemisphere isn't used the same way.
Just a quick note about the hemispheres: you have two, and any time we say things like "language is in the left hemisphere" or "face processing is in the right hemisphere" we are talking about the brains of typical right-handed people. This is not because we have anything against left-handed people! It's just that most people are right-handed and so most of the work has been done with right-handed people. As far as the brains of left-handed people go, it does not seem to be the case that it's simply reversed, and there seem to be different kinds of left-handedness (to make things even more complicated!)
To figure out if joint attention depends on faces and what parts of the brain are involved, Alan Kingstone, Chris Kelland Friesen and Michael Gazzaniga conducted a series of experiments on a very special individual, "J.W." Because of severe epilepsy that was not responsive to drugs, J.W. had surgery to cut his corpus callosum, the large bundle of neurons that connect the two hemispheres. This greatly reduces the seizures of epilepsy, but it also means that information cannot be shared between J.W.'s hemispheres: what the right hemisphere does is completely isolated from what the left hemisphere does.
By using stimuli that are briefly presented, researchers can know which hemisphere has what information. When J.W. is looking at center of the computer screen, images presented on the left half of the computer screen will be processed by the right hemisphere; the left hemisphere will process images presented on the right half of the computer screen. This is true for everyone (not just J.W.) and is called contralateral projection--the left hemisphere receives input from the right side of the world (vision, sound and touch) and controls the right side of your body; the right hemisphere does the same for the left. When J.W. keeps his eyes focused on the center of the screen, we know what each hemisphere is seeing (we know this for other people, too, but their hemispheres are connected and so information passes from one to the other). Using two very simple faces and a target detection task, Kingstone et al. can find out how much the joint attention reflex depends on the face by manipulating whether or not the right hemisphere (the one with specialized face processing) is presented with the target.
The task is really simple: keep your eyes on the cross, and quickly identify where the asterisk appears. Half the time it appears on the left, and half on the right. To see if joint attention matters, two simple faces are presented, and right before the target appears, pupils are added to both faces. Half of the time the pupils mean that the faces are looking where the target is about to appear, but half of the time they look the wrong way.
In three different experiments, Kingstone and his team presented J.W. with faces, inverted faces and only the eyes. When the target was on the right (and so processed by the left hemisphere), J.W. was equally fast at detecting any target for all three kinds of pictures. However, when the target was on the left (and so processed by the right hemisphere) the faces mattered. Here's a graph depicting how quickly targets on the left were detected in congruent and incongruent trials.
Targets appearing on the left, and so processed by the right hemisphere, take longer to find when the face looks the wrong way. It takes longer because attention is reflexively directed where the face is looking, and that's not where the target appears. The face didn't matter at all when the left hemisphere was detecting the target (times were equally fast, around 450 ms), so this gaze effect has something to do with face processing of the right hemisphere. You can see that in the graph when you consider the inverted face condition: now it doesn't matter where the face is looking, the right hemisphere is fast for both congruent and incongruent trials. Face processing is best with upright faces, and the specialized areas of the right hemisphere aren't used the same way when the face is inverted (they might not be used at all). How much of the face do we need for a gaze effect to occur? In fact, the eyes alone are enough to redirect attention for the right hemisphere, as you can see in the graph.
Now, back to your conversation with Dave in the coffee shop--the fact that your eyes reflexively follow his does have to do with his face and eyes, not just the fact that some movement was occurring. If it was just movement, both hemispheres would show the gaze effect, not just the one with specialized face processing.
Kingstone, A., Friesen, C. K. & Gazzaniga, M. S. (2000). Reflexive joint attention depends on lateralized cortical connections. Psychological Science, 11, 159-166.
Learning how to do magic professionally requires theater training, and special attention given to how the magician uses face and eyes to convey (often false) information. The artful control of attention (misdirection) and thought exhibited by a professional magician utilizes facial expression. This does suggest however, that those with right hemisphere damage may not "get" the magic as well. Hmmm, a possible fun research project if we could locate enough of the right (ahem) subjects.
About hemispheres—I've noticed many "identical" twins are left-right mirror images (asymmetry of face, in particular, is easy to notice).
The few pairs I've known well also had an asymmetry of perception or practice (I don't know exactly what it is) that has always made me wonder if the underlying brain structure is also mirror-image _physically_ and then the normal human hemisphere differences are developed on top of this.
Example: Identical twins, both painters. One paints extremely detailed photo-realistic canvases (huge, 6 or 8' across, with exquisite detail). The other paints extremely detailed realistic canvases, comparably large, of greatly enlarged objects, like a single eggplant.
I wonder if anyone else looking at pictures of twins has commented on the "mirror image" faces, or studied this at all.
Oh, and just to clarify—the identical twins I"ve known have been identical in hand dominance, and for that matter in eye dominance as near as I could tell.
I think that the lateralization (left and right brain, handedness, etc) develops identically, as the twins grow, but maybe there was some handedness at the point they became twins that shows up as a mirror image in the underlying bone structure (at least, for facial asymmetry) and perhaps, I wonder, in other ways.