Mass Surveillance and the Problem of Errors

In Wired, more reasons to worry about widespread government surveillance.

There are few, if any, studies demonstrating the effectiveness of mass surveillance. People with something to hide are adept at speaking in codes. Teenagers tell their parents they are “going to the movies” when they are going to drink beer. Attackers know to misspell the victim’s name, as journalist Daniel Pearl’s kidnappers and murderers did, to evade e-mail surveillance. Meanwhile, modern filtering technology can’t distinguish between breast cancer websites and pornographic ones.

Any search algorithm, whether public or not, is unlikely to be able to distinguish between innocent and criminal communications.

Even if the technology works, it fails. Even if a TMS was 99.9 percent accurate, it will produce a false positive one in 1,000 times. Whether it’s facial recognition at the Super Bowl, or sifting through e-mail communications, TMS will inevitably produce an unacceptably high number of false positives. Hundreds of thousands of innocent people will not be allowed to board their planes, will have their houses searched, their bank accounts frozen — at least until the mistake can be cleared up. At best, a “hit” will require someone to look more closely at the information, and we’ll need more agents to do it than we currently have, or could have.