
DanyJ wrote:
Someone told me a story once about a study of (I guess) a large organization's hiring procedures, trying to build an expert system that could automate the first rounds of hiring. Given prospective hires' resumes, and the list of those who had actually been hired, they asked the computer what the highest correlated factor was. It turned out to be skin colour. So they said "whoops, that wasn't the result we wanted" and they removed the photos from the corpus and asked the computer again, and it said "length of surname" -- presumably because honkies are more likely to be called Smith than Ramsoonajar.
Depending on geographcal location, there are probably more Singh's, Kaur's, Wong's, Ng's, Oh's, Poh's than Richardson's, Robertson's, Donaldson's ... So that presumptions may not be correct.
I daresay other parts of that anecdote are even less accurate :-)