Technology Plays Mixed Role in Marathon Bombing Investigation

I'm a resident of Cambridge, but I work from home (EE-Evaluation Engineering's headquarters is in Sarasota, FL), so I was able to stay on the job during the Friday lockdown as authorities pursued the Boston Marathon bombing suspects. Except, of course, I frequently checked the news, or non-news, as the case mostly was. The airwaves seemed to be filled with static views of the suspects' home in Cambridge or helicopters hovering over Watertown.

Over the weekend I've been reflecting on the role of technology in identifying the suspects in the Boston Marathon bombing and in finally locating and capturing the surviving one.

Perhaps most useful in this case was the widespread deployment of cameras (both surveillance cameras and consumer cameras) at the site of the blasts last Monday. That's a fairly low-tech technology, and it was accompanied by the usual problem with camera deployment: too many photos and videos and not enough people to evaluate them.

The Washington Post described the efforts to evaluate the photographic and video evidence: “The work was painstaking and mind-numbing: One agent watched the same segment of video 400 times. The goal was to construct a timeline of images, following possible suspects as they moved along the sidewalks, building a narrative out of a random jumble of pictures from thousands of different phones and cameras.”

Of course, technology also made a trove of multimedia data available to anyone with an Internet connection, and crowd-sourcing provided an assist in the aftermath of the tragedy. As the National Journal put it, “The Internet…aided thousands in coordinating an immediate relief effort after the bombing. Google set up a person-finder application to help victims, dispersed runners, and families find one another after the blast. A massive database was created to offer couches and beds to stranded marathon runners.”

Unfortunately, crowd-sourcing was not of much help to investigators. Here's the National Journal again: “While social media shined, it also faltered. Reddit users turned [into] vigilantes, leading many down false paths and accusing people innocent of the crime.”

The Washington Post article elaborated on this theme: “In addition to being almost universally wrong, the theories developed via social media complicated the official investigation, according to law enforcement officials. Those officials said Saturday that the decision on Thursday to release photos of the two men in baseball caps was meant in part to limit the damage being done to people who were wrongly being targeted as suspects in the news media and on the Internet.”

A higher tech initiative that would have been helpful is face-recognition technology. Unfortunately, the technology in this case wasn't up to the task—it's not clear whether that's due to problems in the existing government face-recognition system algorithms or due to the sheer volume of photographs and videos that needed to be processed. The Washington Post article quotes Boston Police Commissioner Edward Davis as saying that facial-recognition software did not identify the suspects even though both Tsarnaevs’ images exist in official databases: “Dzhokhar had a Massachusetts driver’s license; the brothers had legally immigrated; and Tamerlan had been the subject of some FBI investigation,” the Post reported.

Andrew Leonard in Salon has a follow-up on why face recognition failed in this case. He contacted Carnegie Mellon computer scientist Alessandro Acquisti, who has conducted successful experiments using “off-the-shelf facial recognition software to identify individuals by comparing photos from a dating site, or taken offline with camera phones, with photos uploaded to Facebook.” As Leonard put it, “…if Acquisti could do it, why couldn’t the FBI?”

Leonard quoted Acquisti as citing several reasons. First is image quality—the original bombing suspect images were taken from far away. The second reason is that government databases tend to have available one frontal shot of each individual. For facial recognition, a good frontal shot might not be as useful as several lesser quality shots from slightly different angles. The third issue Acquisti cited is computational costs, which he doesn't believe played a role in this case. But that leads to the fourth concern, which is the number of false positives that can result from processing tens or hundreds of millions of images.

Ultimately, the apprehension of the remaining living suspect relied not one bit on technology. When the governor ultimately lifted the lock-down (or the pathetically euphemistic “shelter at home”) order (or request), a Watertown resident wondered into his back yard, where he encountered blood near his boat, stored for the winter under a protective tarp in his back yard. The suspect lay near death inside the boat, but the boat was outside the search perimeter defined earlier by the government agents who had performed so heroically throughout the previous day and night.

Of course, the best outcome would be not in using technology to apprehend suspects but to prevent crimes in the first place. Years ago I saw a demonstration of using vision technology to automatically determine when a potential perpetrator would abandon a potentially dangerous package. Unfortunately, we are far away from being able to deploy such preventive measures at an event where one or two malefactors are outnumbered by hundreds or thousands of participants and spectators.

One additional note: some news operations seemed to criticize unknown organizations for apparently using government funds for educating the suspects at the “private, prestigious, elite Cambridge Rindge and Latin High School.”

OK, I can accept “prestigious” and maybe even “elite,” but private it's not. Cambridge Rindge and Latin is the public high school in Cambridge. If you are of high-school age, living in Cambridge, and your parents can't or won't send you to an expensive private school like Buckingham Brown & Nichols, then you go to Cambridge Rindge and Latin. My son got a great education there.

More in Instrumentation