The Difficulty of Surveillance Crowdsourcing
Internet Eyes is a U.K. startup designed to crowdsource digital surveillance. People pay a small fee to become a “Viewer.” Once they do, they can log onto the site and view live anonymous feeds from surveillance cameras at retail stores. If they notice someone shoplifting, they can alert the store owner. Viewers get rated on their ability to differentiate real shoplifting from false alarms, can win 1000 pounds if they detect the most shoplifting in some time interval, and otherwise get paid a wage that most likely won’t cover their initial fee.
Although the system has some nod towards privacy, groups like Privacy International oppose the system for fostering a culture of citizen spies. More fundamentally, though, I don’t think the system will work. Internet Eyes is primarily relying on voyeurism to compensate its Viewers. But most of what goes on in a retail store is incredibly boring. Some of it is actually voyeuristic, and very little of it is criminal. The incentives just aren’t there for Viewers to do more than peek, and there’s no obvious way to discouraging them from siding with the shoplifter and just watch the scenario unfold.
This isn’t the first time groups have tried to crowdsource surveillance camera monitoring. Texas’s Virtual Border Patrol tried the same thing: deputizing the general public to monitor the Texas-Mexico border. It ran out of money last year, and was widely criticized as a joke.
This system suffered the same problems as Internet Eyes—not enough incentive to do a good job, boredom because crime is the rare exception—as well as the fact that false alarms were very expensive to deal with.
Both of these systems remind me of the one time this idea was conceptualized correctly. Invented in 2003 by my friend and colleague Jay Walker, US HomeGuard also tried to crowdsource surveillance camera monitoring. But this system focused on one very specific security concern: people in no-mans areas. These are areas between fences at nuclear power plants or oil refineries, border zones, areas around dams and reservoirs, and so on: areas where there should never be anyone.
The idea is that people would register to become “spotters.” They would get paid a decent wage (that and patriotism was the incentive), receive a stream of still photos, and be asked a very simple question: “Is there a person or a vehicle in this picture?” If a monitor clicked “yes,” the photo—and the camera—would be referred to whatever professional response the camera owner had set up.
HomeGuard would monitor the monitors in two ways. One, by sending stored, known, photos to people regularly to verify that they were paying attention. And two, by sending live photos to multiple spotters and correlating the results, to many more monitors if a spotter claimed to have spotted a person or vehicle.
Just knowing that there’s a person or a vehicle in a no-mans area is only the first step in a useful response, and HomeGuard envisioned a bunch of enhancements to the rest of that system. Flagged photos could be sent to the digital phones of patrolling guards, cameras could be controlled remotely by those guards, and speakers in the cameras could issue warnings. Remote citizen spotters were only useful for that first step, looking for a person or a vehicle in a photo that shouldn’t contain any. Only real guards at the site itself could tell an intruder from the occasional maintenance person.
Of course the system isn’t perfect. A would-be infiltrator could sneak past the spotters by holding a bush in front of him, or disguising himself as a vending machine. But it does fill in a gap in what fully automated systems can do, at least until image processing and artificial intelligence get significantly better.
HomeGuard never got off the ground. There was never any good data about whether spotters were more effective than motion sensors as a first level of defense. But more importantly, Walker says that the politics surrounding homeland security money post-9/11 was just too great to penetrate, and that as an outsider he couldn’t get his ideas heard. Today, probably, the patriotic fervor that gripped so many people post-9/11 has dampened, and he’d probably have to pay his spotters more than he envisioned seven years ago. Still, I thought it was a clever idea then and I still think it’s a clever idea—and it’s an example of how to do surveillance crowdsourcing correctly.
Making the system more general runs into all sorts of problems. An amateur can spot a person or vehicle pretty easily, but is much harder pressed to notice a shoplifter. The privacy implications of showing random people pictures of no-mans lands is minimal, while a busy store is another matter—stores have enough individuality to be identifiable, as do people. Public photo tagging will even allow the process to be automated. And, of course, the normalization of a spy-on-your-neighbor surveillance society where it’s perfectly reasonable to watch each other on cameras just in case one of us does something wrong.
Categories: Privacy and Surveillance