Dr. Herbert Lin has an excellent article in Science Magazine this month that teases out the technological fault lines that have emerged over the last year in the debate over digital surveillance. I want to call attention to the section on necessary verse useful data. Herb contends:
“Critics of U.S. collection efforts have argued that information derived from many surveillance programs has not been necessary in thwarting any terrorist plot. Less has been voiced about whether a given piece of information, even if not absolutely necessary, may still be useful. Intelligence analysis does not seek to reach a conclusion that might be possible from a minimal set of information but rather conclusions that are supported as much as possible.”
I have a few reactions to the distinction Herb is making. First, I would actually distinguish between necessary, useful, and useless data. NSA has actually been collecting a large amount of information that it knows it does not need. This is the case with its bulk phone records database. If NSA had focused on the data it had a good reason to believe was useful and had taken pains to avoid collecting useless data, it would have preempted a lot of the debate over the last year.
Second, Herb is exactly right about the value of useful but not necessary data. We don’t want to create a collection regime that limits intelligence agencies to the data we expect to be minimally necessary. Such a regime would ensure that those agencies actually miss some necessary information and, as Herb argues, it would ignore the value of corroborating information.
Third, while corroboration is valuable, we also should not ignore the diminishing returns of more data and the costs associated with information overload. It isn’t always useful to have more data. With most successful terrorist attacks against the United States over the last decade, the intelligence community has not suffered from lack of reporting. Rather, the problem in many cases has been too much reporting. This made it difficult to find the credible intel that would have helped disrupt attacks.
The points above leave open the fundamental question of Herb’s article. How do we draw a line between an overly restrictive regime focused on necessary data and an overly broad regime focused on useful data? That is a challenging question that we are all now trying to answer.