Friday, June 14, 2013

Two points on PRISM

The discussion about PRISM has brought out a lot of commentary. Much of it, described in my two earlier posts on the subject, is silly. But two issues caught my attention in the last couple of days. The first was a WSJ article on whether the PRISM program suffers the statistical risk of being inaccurate. The WSJ has a regular column called the Numbers Guy which often offers remarkable insights about numbers related subjects in the news. This article presented some comments from Biostatisticians about whether the methodology to collect information in PRISM is likely to yield tons of false positives (identifying culprits where none are present). Two quotes will give you an idea about the issue. One PhD candidate in computational ecology wondered on his blog whether PRISM could (based on the algorithm) produce 10,000 false positives for every real hit. Peter F. Thall, a biostatistician at the University of Texas' M.D. Anderson Cancer Center commented that the even if the NSA methodology is "is terribly clever and has a very high sensitivity and specificity, it cannot avoid having an immense false-positive rate,"  That should give one pause, especially if this data is held in perpetuity (and with the rapidly declining cost of storage that is probably the default option).

There is a second issue which I began to think about while I was reading a book called Socialnomics.  The author makes a point about how instantaneous information is and also how important it is for purveyors of information to be genuine.   He cites an example of a college football player who made an inappropriate tweet about President Obama and was immediately sanctioned for his one comment.   The author does not mention the video that hit YouTube for a while of a couple at a campus in the west who were videoed on a rooftop of a campus building having sex.   In both of those cases the students might well have gotten away with something that was a bit outside the norm in earlier years, but no more.   But both are prime examples, of how 140 little characters can affect people's lives.

In earlier times, security agencies were able to get away with impinging on freedoms by simply claiming that they were protecting us.  I am pretty sure that we should be using all sorts of surveillance of potential terrorists, after all, I start with the notion that I have nothing to hide.   At the same time, in this era of instant communication, it would be good and prudent to establish a reliable review process to cross check whether the benefits outweigh the costs.  In the highly charged atmosphere of Washington politics that might be very tough, but it is still necessary.   From all of the commentary to date there is little evidence that this program received more than a scant attention from insiders about key questions about effectiveness and alternatives ways to achieve the results.    Irving Janis, in a book called Groupthink, argued a couple of decades ago that experts tend to fall into conforming errors.   In this age, we need to establish systems which will counterbalance those traps.

No comments: