Visitors to the project are presented with the map so far, a crude bitmap build out of green and blue blocks. Green blocks represent land, blue blocks water. One of the blocks is red. The visitor is then asked whether that block should be land or water.
After a while, with enough visitors, the shapes of the continents start to appear and a first world map becomes visible. This is about where the project is now.
With every click, the ip number of the user is stored. In a second phase, this can be used to reconstruct worldmaps as pictured by people from a certain top-level domain, ie compare the world as seen by people from .org or .com. People will tend to make their own continent/country/region bigger and more far off places smaller.
The current approach is rather vulnerable to misuse. Anybody with the time and will could modify the map by clicking long enough. However, since all clicks are registered, clicks that come from IP numbers that contradict the common opinion too much could be filtered out. On the long run, these minority opinions could be projected on their own maps.
The position of the continents is another variable. Older American maps sometimes show the Americas in the middle. European and more modern American maps show Europe in the middle. Asian maps tend to show Asia in the middle. A google search for "world map" site:.jp returns about 50% maps with asia in the middle, while "world map" site:.us returns even some America centric maps.
In a similar project by Kevan Davis, The Smaller Picture. Here visitors develop together a font. The site has been around for a while and animation feature allows you to see how letters have developed. Most letters stabilized after 1000 iterations. After that, a lingering noise stays around. This could probably be fixed by the above described minority/majority opinion system