The unpopular, but more logical side of everything

Over-surveillance would be nice — If it only worked

February 28th, 2009 — Sir David, the former Whitehall Security and Intelligence Coordinator, said that the state should analyse data about individuals, such as travel information, tax and phone records, emails, etc., in order to find terrorist suspects.

The date may be quite far into the past, but it is still very much valid. And right now, in the age of technology, and in a time when everything is instantly connected, it is more important than ever.

Most people don’t like being constantly surveyed, but accept that this would help keep them safer; Omand accepts this, though, and says, “Finding out other people’s secrets is going to involve breaking everyday moral rules.” This will naturally be the case as we’ll need to screen everyone to find the small number of suspects. The main problem lies here — this small number becomes infinitesimally smaller with a larger pool of people. Your chances of false positives grow exponentially with an increase in the sample size.

Suppose you have an insanely accurate test that correctly identifies a true suspect eight out of ten times, and each time you use this test on an innocent person, it correctly identifies them as guilt-free nine out of ten times. Here you get one false positive, one chance of throwing an innocent in jail for life, and two false negatives — two chances for mass killing and widespread panic.

Now let’s look at this test from the other end — the end we can actually use it from. You have the test results, but you need to determine whether or not an individual is a terrorist or not. This, as mentioned before, depends entirely on the size of the population being tested.

With the sample size as ten, backed with the knowledge of the fact that one of them is a terrorist, you will get one positive, and one false positive on average. If you have a hundred people, your false positive frequency rises to ten, while your positive frequency stays stagnant at one. At this point and beyond (which is usually the case), the test becomes pretty useless (counter-productive, even).

With over sixty million people in the UK, even with your ridiculously, unrealistically accurate test, that gives you six million false positives, and you miss around two thousand true suspects. How, may I ask, is that of any help? But let’s be positive about this — why not suppose we could make the test even more accurate? After all, with all the development in AI going on, this is bound to happen ten, maybe twenty years from now. How about if the false positive count drops to only one in one hundred? You still get six hundred thousand false positives.

The accuracy levels of both these tests though, are unattainably high. It’s very difficult to identify slight abnormalities from the mundane things that everyone does. Throw the fact that these suspects lay false trails to mess with the analysts, and you’ve got yourself one heck of a problem.

As if all of that wasn’t enough, there’s also the problem of validating your tests. You’ll have to have a sample of at least ten thousand people, knowing for sure who’s a suspect and who isn’t, checking it against the results of the algorithm. It’s an impossibly hard task, and one that is equally hard to even fathom being done.