So there's this company called NetBase that develops data-mining applications. And they decided to release a public demo of one of their applications under development, a health-oriented semantic search engine called HealthBase, promoting it with press releases aplenty.
Problem is, they didn't bother to actually [i]test[/i] their application thoroughly before deploying it. Which is a particularly troublesome issue when your data set includes a good deal of non-medical data... such as Wikipedia.
The result? As this TechCrunch article points out, one of the causes of AIDS was listed as "Jew"-- because the system mistook the verb "aids" in a Wikipedia article for a reference to the disease. Even worse, clicking on the link to "Jew" revealed that treatments for that condition include coarse salt and Dr. Pepper.
They've fixed some of the really glaring examples by now, but the service still offers up quite a bit of hilarity. For instance, one of the causes of ADHD is "campus", while one of the possible treatments for it is a "rat". And treatments for depression include the Great Depression, the gold standard, the Canadian Shield, regression analysis, and 3-tesla proton magnetic resonance spectroscopy.
More hilarious results can be found in the comments to this post at Language Log. Some of them have been fixed since the comments were posted; others remain as utterly absurd as ever.