Machine Learning
XKCD has a wonderful new comic:
These days there also was an article in MIT technology review criticising the same issue.
For me this is a daily issue. My colleagues usually are happy to just get a quick test procedure; usually on smallish data. Chances are it pattern matches some noise. But nobody will find out as reading what a classifier found is near impossible. I mean sure we weed out all the nonsensical predictor variables – but even on the sensibel there is additional information mixed in.