Can the Web Predict Future Attacks?
Shachtman has a lengthy post about a company which is hoping to use a software program to analyze trends from "chatter" and help the intelligence community predict events. Go read it.I'm skeptical about the intel value of "chatter". Not that it has zero value, but that given the number of false positives hit by so-called "chatter" analysis the value of such predictive models goes way down. And like any statistical model based in regression, it's easy to pick out the correct predictions post hoc and say, see, the model predicted this event would occur. While, at the same time, the models also predicted scores of other events that never came to pass. The goal of the data mining of "chatter" is to allocate limited resources more strategically. To get intelligence and law enforcement agents to spend their time in the most productive ways. But I've yet to be shown evidence that computer programs are actually doing this. Of course, I don't see what goes on inside the black box of intelligence agencies, but the modeling seems to be similar to other trend predicting programs such as Google Trends. Which, again, tend not to predict events very well. For instance, it would be difficult to find a causal stream going from "chatter" about a Jessica Alba nipple slip to that nipple slip. The cause and effect relationship is backwards. And again, given the viral nature of ideas being spread across online social networks, most of which turn out to be nothing, then it's hard to imagine the usefulness of such tools for prevention. But then again, those familiar with regression lines know that once you have post hoc knowledge of the models robustness and validity then you can always trace that line back to something close to origin. Such knowledge isn't very helpful in predicting events, but might be very useful in identifying individuals or networks which had prior knowledge of such an event occurring. And that is useful information, but might be as good as it gets.