If you simply ask your local police department if it uses facial recognition technology, you may or may not get a complete answer. But if you make that inquiry in the appropriate legal language and pinpoint the phrasing of your query, you just might improve your odds of discovering the details of their facial recognition tactics.
A project launched in November by two transparent-government groups, Open the Government and MuckRock, aims to help by sharing lessons learned from using state public-records laws to quiz law-enforcement agencies about their use of this technology.
Even as facial recognition systems have begun expanding to airports, the automated systems used to match images of faces to records of identities have drawn extensive criticism for their potential invasion of privacy and possible bias, especially with nonwhite populations.
Specifically, a 13-page document provides a concise explanation of how the federal Freedom of Information Act (FOIA) and comparable state laws work; a checklist for drafting, submitting, and following up on a FOIA request; and sample text to use in that request. Making these queries and following up on them—”FOIA-ing” for short—can be a prolonged and frustrating process even for experts.
In a talk in December at the annual surveillance conference in Washington presented by the Cato Institute—a libertarian think tank—Open the Government policy analyst Freddy Martinez shared some early findings from the first 100 public-records requests sent by that group and MuckRock.
- Most police departments haven’t purchased facial recognition systems and instead use such services through data-sharing partnerships with other law-enforcement organizations or by renting them from third-party firms. That makes it necessary to phrase FOIA requests carefully to avoid accurate-but-incomplete responses such as “we don’t own this system and we don’t intend to purchase it.”
- FOIA requests for training documentation rarely pan out, suggesting that police who use facial recognition may not be sufficiently prepared for the task. “Training manuals are almost nonexistent,” Martinez told the audience at the Cato conference. One of the few documents to emerge from this round of queries reveals that Lubbock, Texas, sent investigators to an eight-hour course that concluded with 30 multiple-choice questions.
- While most police uses of facial recognition systems have relied on such image sources as driver’s license databases and booking photos, a few police departments are now querying footage from privately-owned Ring cameras. (I was going to call them “Ring security cameras,” but recent revelations of horrible security practices at that Amazon subsidiary make that description highly questionable.)
- The debate over facial recognition’s racial bias—highlighted most recently in a study released December 19 by the government’s National Institute for Standards and Technology—has yet to percolate through many policy departments.
In an email sent to me on December 30, Martinez said that “most/all” of the public-records filings recorded in this effort have come from either Open the Government or MuckRock. He added that he hasn’t seen other signs of attention to training and the potential for bias among police departments.
Asked for an example of a particularly responsive government, Martinez pointed to the Bay Area.
“San Francisco did a fairly lengthy release to us for all of their collection of biometrics,” he said. The San Francisco Police Department turned over 18 documents, most relating to the department’s use of fingerprints collected from arrest suspects.
But San Francisco is one of a handful of cities—others are Oakland and Berkeley across the bay, plus Somerville, Massachusetts—that have already banned police use of facial recognition systems. That’s not where the problem, to the extent that it exists, is going to be found. But even after such efforts as a map of government facial recognition applications released this summer by the activist group Fight for the Future, we’re still in the early stages of figuring out where to look.