As widely reported, the Manifesto of Anders Behring Breivik contains coded information which may lead to geographic locations in major European cities. Further, the Manifesto may contain other coded information, which by speculation could identify conspirators or possibly the existence of armed and timed explosive devices. The speculation, if true, would have the obvious dire consequences.
At present, Norwegian security analysts, assisted on invitation by the American Federal Bureau of Investigation, and others, are intensively investigating the Manifesto for coded expressions and their translations to plain language. In this endeavor knowledge of speculated content is important, though providing an obvious bias to the decoding effort. For a simple example, consider that one suspects that a code says, “The sky is blue.” One then looks for patterns containing the coded words for “sky” and “blue.” Upon supposedly finding them the decoding soon follows. However, suppose the true coded message is, “Tomorrow it will rain.” That message will be missed completely. If the analyst had been looking instead for the words “tomorrow” and “rain,” then perhaps he would have found the true message. The failure, in the first instance, is to accept a false hypothesis, an error known as a Type II error in the world of hypothesis testing. (A Type I error would have been to reject, “The sky is blue,” assuming that had been the true coded message.)
All this is well and good, standard stuff. The issue in the Behring Breivik case arises because of the extremely severe loss function involved. Assume the hypothesis is that the coded information in the Manifesto contains irrelevant babble, and that no public danger exists, no matter what the decoded statements imply. This hypothesis is widely believed by investigators and specialists to be true, with high probability. But what if that hypothesis is false, considered the remote possibility. Then obviously the danger to the public is extreme. By contrast, the cost of determining that the Manifesto language is in fact babble, i.e., rejecting the alternate hypothesis that the man has remaining widespread criminal capabilities, has modest cost. That result only depends only on the diligent investigation of the Manifesto by qualified professional computer security cryptanalysts, taking days, perhaps weeks, but no more. Law enforcement and military authorities could then take any information gleaned, and proceed to confirm that no threats exist.
The technical language here is that one needs to devise a test with high power, that is, one with a very low probability of the Type II error. This is not at all an easy task, for the variables herein are those of human nature and psychology—very difficult to quantify. So, any result of these investigations will only be expressible in general, qualitative terms, giving scant comfort to all of us who realize that the errors of all types—Type I or Type II, or just human—can always happen, with whatever small chance.