UnBlinking: Symposium: Recognition Markets and Visual Privacy

Abstract

Image capture technologies--tiny cameras, zoom lenses, and cheap sensors--have become incredibly sophisticated and ubiquitous. But despite the rapid development and proliferation of image capture devices, there still seems to be a sense that we can "hide in plain sight" of the unblinking eyes that surround us. Like teenagers who bare their souls on MySpace pages but feel confident that their parents will never find them, we are comforted by the sheer volume of images being captured. In other words, we don't mind being seen, as long as we aren't recognized.

This suggests that the discussion about visual privacy should focus on not image capture technologies, but image recognition technologies. Recognition links images to specific people, places, and things. Without recognition, surveillance images are just so much hard drive space. With recognition, images gain tremendous force, becoming memory, evidence, proof, or propaganda. Recognition technologies have not developed at nearly the rate of capture technologies. Even the purchasing power of the United States Department of Homeland Security has not yet resulted in recognition systems with acceptable error rates. But this is changing, in a somewhat unexpected way.

Every year on April 20th, University of Colorado students gather in Farrand Field for an act of civil disobedience and hedonism: the mass consumption of marijuana. In 2006, the University of Colorado Police Department prepared for this event by placing cameras around the field and posting signs notifying visitors that all activity there would be photographed and videotaped. Confident in their numbers, the students ignored the signs, perhaps secure in their belief that though they might be seen, they wouldn't be recognized. The UCPD duly captured images of the ensuing conflagration, and then made a novel move: they posted the images on a public web server, with the explanation that anyone who successfully identified a pot-smoking miscreant would receive fifty dollars. Within days a substantial portion of those depicted had been recognized.

The UCPD used the internet to enroll a mass audience in the task of identification, in effect creating a human-machine hybrid technology of image recognition. Throwing images against a wall of eyeballs to see what sticks is not entirely new: consider the Most Wanted posters on the wall at the post office. What is new is the scale and speed at which the internet can bring together those willing to sell their powers of recognition with those willing to buy. This realization has resulted in an explosion of proposals for, and some implementations of, new socio-technical systems which use "human-in-the-loop" approaches to solve image recognition problems.

What are the implications of these "recognition markets" for visual privacy? How should they be regulated? What ethics should govern their use? Should sellers of recognition be made aware of how the images they empower will be used? Who should be held responsible for cases of misrecognition, especially in systems which divide up the task of recognition among multiple parties? How might these systems be exploited or attacked? This paper will discuss the development of recognition markets thus far as well possible developments in the future, and will suggest some guidelines for their ethical design and implementation.

UnBlinking: SymposiumNov 3/4, 2006

Abstract

Biography