[From Dr. Dobb's AI Expert Newsletter]
Evopedia is an online encyclopedia with a difference. It evolved.
Evopedia's home page explains how. In evolutionary programming, you need a problem, a way to generate random solutions, a way to evaluate solutions (a fitness function), and a way to mutate solutions. For Evopedia, the problem is the need for an informative Web page about some topic. To start, Evopedia creates some Web pages at random from other pages, using Markov text-generation. It copies them to www.evopedia.com under their topic headings, and waits for several weeks while gathering stats on their usage. The worst pages (presumably the least read) are discarded; the best are kept and mutated to make new pages. These get copied to the site in their turn, and so the cycle repeats, converging eventually (so Evopedia hopes) to sense.
To mutate pages, Evopedia swaps sections within them, replaces sections with new sections, or adds or removes sections. Its home page is weak on detail, but as common in evolutionary programming, it appears to model these mutations on the behaviour of DNA. Those unfamiliar with genetic algorithms will find copious information on the Web; a good intro with clear diagrams is the Evolutionary Computing chapter of Shahab Mohaghegh's Virtual Intelligence and its Applications in Petroleum Engineering.
Does Evopedia work? Here's the entry for Cake Recipe
Now Isn't this cool!, what do you think Mr. Turing?