Failed: Surveillance Tech Predictions

Days before 9/11, MIT's Technology Review released a tour de force review of surveillance technologies that would soon change the world. Now, nearly a dozen years later, nearly none of them are reality. In this note, we examine these projections.

Their warning is especially ominous:

"Many observers argue it’s no longer a question of whether ubiquitous surveillance will be applied, but under what guidelines it will operate-and to what end."

And with the recent uproar over the NSA and Prism, ubiquitous Internet surveillance is certainly a risk but in the physical world, we are still far from it. Here is a recap and commentary of the claims MIT TR made in 2011:

• "The job of spotting suspicious people and behavior in this stream of electronic imagery is becoming automatic, with computers programmed with special algorithms for matching video pixel patterns to stored patterns associated with criminals or criminal actions-and the machines themselves passing initial judgment on whether a behavior is normal." Remains science fiction, or BRS Labs claims.
• "Identifying specific human beings by measuring the color spectrum emitted by their skin." Science fiction.
• "Automatically measure such characteristics as leg length and waist width to provide ... 'the measurements you give to a tailor.' The idea here, he says, is that those numbers should be able to serve as a kind of body fingerprint for identifying specific individuals." Science fiction.
• "Developing sensor-riddled “smart floors” that can identify people by the “force profiles” of their walking feet." Science fiction.
• "An antiterrorist technique that uses a special camera to identify individuals from a hundred meters off by the patterns of color, striation and speckles in their irises." There is Sarnoff's Iris or the move but that's a far simpler and more limited implementation.
• "Record customers’ facial expressions and eye movements, tracking the effectiveness of in-store promotions." There are some tracking based on movement but little more sophisticated than that.
• "System detects and tracks both invariant aspects of a face, such as the distance between the eyes, and transient ones, like skin furrows and smile wrinkles. This raw data is then reclassified as representing elemental actions of the face ... Many facial expressions reflect human emotions, such as fear, happiness or rage, which, in turn, often serve as visible signs of intentions." Science fiction.
• "Automatically scan collective human activity for signs of anything from heart-attack-inducing Type-A behavior to sexual harassment to daydreaming at the wheel to homicidal rage." Science fiction.

It says quite a lot that a prestigious, respected publication could be that dramatically off, even given a decade for their projections to come through. 

The theme in these claims was summarized in the article as the “goal of developing computer systems that can detect human activity, recognize the people involved, understand their behavior, and respond appropriately.” While we have some basic video analytics, like tripwire detection, nearly none of this goal is close to being met even now.

Computer vision problems are incredibly hard. Even with ~8 generations of Moore's law, and computing power 2 orders of magnitude greater than a decade ago, they remain hard, and mostly out of reach. This will change over time (it remains the 'next big thing') but we have clearly and vastly underestimated the difficulty of delivering production physical surveillance systems.

Despite this, the ACLU warned then, in 2001, "The technology is developing at the speed of light." More like diamonds.