Not to rain on the parade but merely to offer a few points to think upon...
1) Use of webcam with Hal will probably slow the program and its responses to you the user.
2) While head recognition is quite possible, facial recognition is still:
(a) difficult to incorporate with existing software
(b) difficult to streamline for performance
(c) costly to obtain / produce
(d) somewhat error prone
Compare facial recognition with voice recognition
Most voice rec apps are "speaker independent" meaning that the program will work with any voice as opposed to "speaker dependent" which is akin to recognizing only the current user/speaker.
A facial rec program would have to first, be trained to recognize your face similar to training a voice for dictation in a speech app.
Next, locate and isolate the user's head as to distinguish it from other items / images that might be present (background things).
Once isolated the program would have to do a "comparison" of all faces contained in its database (provided yours is not the only one).
Lastly, once a favorable match has been located, the program would most likely do a final comparison of features like brow ridge, chin, cheek bone placement, bridge of nose, eyes, etc. until it "knows" with certainty that it has a correct match.
While we're talking seconds and miliseconds the processing time to obtain the final match takes time away from the program's main instruction set.
One thing that might logically be done would be to have the program do it's facial search after startup is completed, find a match then drop out of its routine. It now knows who you are so it wouldn't need to keep "looking" at you. The program could then carry on as normal.
If you were to leave, log out, quit, screen saver (maybe) then another facial rec routine might be required.
Just some thoughts...don't get me wrong, I'm quite excited at the prospect of webcam usage for Hal. We'll just have to see how far Robert plans on pushing the limits.
Here's looking at you kid!