Slashdot Mirror
When Metadata Analytics Goes Awry
jfruh writes "When blogger Dan Tynan started seeing lots of Latvians in his LinkedIn People You May Know list, it was pretty funny, considering he'd never been to Latvia or ever met anyone from there. But now that shadowy spy agencies are using algorithms similar to LinkedIn's to see if we're terrorists, mistakes like this are a lot scarier. From the article: 'More than ever -- and online in particular -- who you know can be more important than who you are. In fact, who somebody thinks you know may be more important than who you are, especially if that somebody is a faceless government bureaucracy with limitless power to izjaukt savu dzvi (mess up your life).'"
Then I have to prove a negative, that I do not know this person. All their evidence points to the opposite. "He was in New York at the same time!" (BUT I LIVE THERE) "Doesn't matter". "Your fathe'rs, cousin's, uncle's former roomate went to Iran as an exchange student", etc, etc.
That's an excellent point - its the classic example of the Prosecutor's fallacy. By definition anyone found through a meta-data search will have strong evidence against them. If someone was caught independently plotting terrorist activities then it would be valid to say that it would be very unlikely for them to have a lot of connections with known terrorists. Trawl through databases and find someone who has a lot of known connections and it doesn't say a lot. Its like if you have evidence that someone tampered with a lottery and won you could say the chances of winning are one in 14 million (or whatever), but if you look for people who have won then it is not valid to say that they must have cheated as the odds against winning are so low - because someone will through chance!