The New York Times has a brief article on why we have a tendency to see faces in chaotic or almost random visual scenes.
The tendency to see meaning in essentially random data is variously known as apophenia or pareidolia, and statistically would be known as a Type I error – a false positive.
Although it is controversial as to whether it is specifically dedicated to recognising faces, an area of the brain known as the fusiform gyrus is certainly heavily involved in perceiving faces.
The fact that this area is so specialised for faces might lead us to detect faces even when they are only suggested by a few dots, the position of clouds or the markings on just about anything.
“The information faces convey is so rich ‚Äî not just regarding another person’s identity, but also their mental state, health and other factors,” he said. “It’s extremely beneficial for the brain to become good at the task of face recognition and not to be very strict in its inclusion criteria. The cost of missing a face is higher than the cost of declaring a nonface to be a face.”
There’s a great web page with pictures of ‘cloud faces‘ if you want to see how spectacular some of these effects can be.
Link to NYT article ‘Faces, faces everywhere’.
Mind Hacks 
Penn & Teller’s show (Bulls**t) did an episode that partially addressed these phenomena. See (http://www.sho.com/site/ptbs/prevepisodes.do?episodeid=s3/signs) for a description. Basically, it covered various religious phenomena that have been reported in the press. (Anybody remember the woman who saw the Virgin Mary’s image in her grilled cheese sandwich?)
I have always wondered whether feral children have these same tendencies. (Feral children are just one example, I mean any other people who have not necessarily been brought up with their wiring heavily geared towards social interaction with other people)
Sure, it’s good for brains to be able to accurately identify faces. But do we do this because of inherent wiring, or because that’s the way the brain wires itself after birth as a result of copious human interaction?