SURVEIILANCEDetroit Takes Important Step in Curbing the Harms of Face Recognition Technology
In a first-of-its-kind agreement, the Detroit Police Department recently agreed to adopt strict limits on its officers’ use of face recognition technology as part of a settlement in a lawsuit brought by a victim of this faulty technology.
In a first-of-its-kind agreement, the Detroit Police Department recently agreed to adopt strict limits on its officers’ use of face recognition technology as part of a settlement in a lawsuit brought by a victim of this faulty technology.
Robert Williams, a Black resident of a Detroit suburb, filed suit against the Detroit Police Department after officers arrested him at his home in front of his wife, daughters, and neighbors for a crime he did not commit. After a shoplifting incident at a watch store, police used a blurry still taken from surveillance footage and ran it through face recognition technology—which incorrectly identified Williams as the perpetrator.
Under the terms of the agreement, the Detroit Police can no longer substitute face recognition technology (FRT) for reliable policework. Simply put: Face recognition matches can no longer be the only evidence police use to justify an arrest.
FRT creates an “imprint” from an image of a face, then compares that imprint to other images—often a law enforcement database made up of mugshots, driver’s license images, or even images scraped from the internet. The technology itself is fraught with issues, including that it is highly inaccurate for certain demographics, particularly Black men and women. The Detroit Police Department makes face recognition queries using DataWorks Plus software to the Statewide Network of Agency Photos, or (SNAP), a database operated by the Michigan State Police. According to data obtained by EFF through a public records request, roughly 580 local, state, and federal agencies and their sub-divisions have desktop access to SNAP.
Among other achievements, the settlement agreement’s new rules bar arrests based solely on face recognition results, or the results of the ensuing photo lineup—a common police procedure in which a witness is asked to identify the perpetrator from a “lineup” of images—conducted immediately after FRT identifies a suspect. This dangerous simplification has meant that on partial matches—combined with other unreliable evidence, such as eyewitness identifications—police have ended up arresting people who clearly could not have committed the crime. Such was the case with Robert Williams, who had been out of the state on the day the crime occurred. Because face recognition finds people who look similar to the suspect, putting that person directly into a police lineup will likely result in the witness picking the person who looks most like the suspect they saw—all but ensuring the person falsely accused by technology will receive the bulk of the suspicion.
