Google puts a lot of work into creating a virtual map of the world with Street View, sending cars and backpackers everywhere with huge cameras. But what if a computer program could do all that automatically? Well, there's one that can. All it needs is Wikipedia and Google Images.
Developed by Bryan Russell at Intel Labs and some colleagues at the University of Washington in Seattle, the program is almost deceptively simple. First, it trawls the internet (mainly Flickr) for a wide variety of pictures of a location. By looking at them from different angles, it's able to piece together a pretty good idea of what it looks like in 3D space from the outside. Like this:
Then, for the interior of that 3D shell, the program cruises through Wikipedia, making note of every single noun-phrase, since its dumb robot brain can't tell what is important and what is not. Finally, it searches Google Images for its big stack of phrases, pulls the relevant pictures (if it can find any), and plasters them roughly where they belong in the model's interior. When that's all said and done, it can then behave as a procedurally generated 3D tour that guides you through a recreation of whatever you're reading about on Wikipedia. Awesome!
It's a crazy idea, but it seems to work pretty well for popular landmarks, at least. The operating logic here is that if a thing is important, there will be pictures of it on the internet, and text describing the pictures. So long as that's true, it's possible to start piecing things together.
For the moment, the program has only compiled complete virtual representations of exceedingly popular and well-documented sites like the Sistine Chapel. With less common places, there's less data to pull. But with the advent of tech like Google Glass, and the wide proliferation of smartphones that can take a half-decent picture, the data-voids are slowly getting filled in -- and they'll only fill in faster as time goes on.
You can watch a video explaining the process in more detail below: