The age-detection part—as demoed in this Intel video—uses an optical sensor to consider the customer's face shape, along with distance measurements between the eyes, nose and ears. The idea of automating age-verification is hardly new, nor are some of the almost laughable failures in those efforts, such as this story from Japan about such a system that was fooled by a kid holding a picture ripped out of a magazine.
Of course, the importance of making the age determination accurately varies with the application. With this Kraft effort, no serious harm comes from an 8-year-old being given the dessert, nor from a 22-year-old who is told she looks too young. (Then again, even if the system works, it may be a well-thought-out strategy. Does Kraft really want to tempt kids with high-resolution dessert images and then deny them a taste? Yeah, that will establish a lot of great future goodwill.)
But when the restrictions are serious and legal—think cigarettes and alcohol—can a retailer rely on a kiosk's measuring skills? What if a 12-year-old has a very long face? There actually is a way such a system could be useful.
If used as the sole determination point, it's a terrible idea. What if it's merely a filter? What if it does most of the work of age verification, but leaves the final decision to an associate? Perhaps the associate would have to look at the customer and then hit a button, sort of the kiosk equivalent of buzzing someone into an apartment building.
Granted, that seems like a very expensive way to marginally accelerate operations. But if a store gets a lot of time-killing under-age requests, this could help. Also, it could be set up so the associate's involvement is barely noticeable, enabling the machine to take the heat for the denial. Purchase denials can be emotional and can get dangerous. This could have the psychological effect of deflating those situations.