The use of automated face recognition in law enforcement is not a binary discussion point. Yet, the political leaders of San Francisco have turned it into just that with recent legislation that bans local government agencies use of the technology. In turn, the safety of their citizens is being put at risk without any rational justification.
How law enforcement uses face recognition
At the core of the issue is a misunderstanding regarding how face recognition technology is used in law enforcement.
Heavy-handed authoritarian regimes in China, Russia, and elsewhere may seek to use face recognition as a surveillance tool that would suppress the same civil liberties guaranteed to United States citizens by the constitution. However, in the U.S. the overwhelming majority use of face recognition technology by law enforcement is as a forensic identification tool used to help identify perpetrators of violent and harmful crimes.
When used properly, face recognition technology is one of the most powerful tools available in today’s law enforcement investigations. For example, if someone robbed a bank, and their face was captured on camera, trained facial examiners using face recognition technology could search a surveillance video frame against a database of mugshot arrest photos in order to generate an investigative lead. Or, as happened recently, a high quality face image was captured of a person in an elevator moments before he broke into a young woman’s apartment and assaulted her. Or, when police collect child pornagraphy evidence, face recognition technology is used to identify and rescue the victims, as well as identify any perpetrators in the imagery. There are thousands of other cases over the years where grotesque or systematic crimes have occurred and face recognition was invaluable tool for identifying the culprit.
As a frame of reference, here are the 2017 Crime Statistics in the City of San Francisco:
- Population: 881,255
- Cases of Violent Crime: 6,301
- Cases of Murder: 56
- Cases of Rape: 367
- Cases of Robbery: 3,220
- Cases of Aggravated Assault: 2,658
- Cases of Property Crime: 54,356
- Cases of Burglary: 4,935
- Case of Larceny-Theft: 44,587
- Cases of Motor Vehicle Theft: 4,834
Unfortunately, due to a blanket overreaction by government leaders, face recognition technology cannot be used when investigating any of these harmful crimes in San Francisco, even if a high quality face image of a perpetrator is available.
The Board of Supervisors in San Francisco do not seem to understand these clearly positive and potentially life-saving uses. As the bill states:
The propensity for facial recognition technology to endanger civil rights and civil liberties substantially outweighs its purported benefits
That is, they believe their law enforcement officers are more likely to use face recognition technology to violate the constitution than to help the lives of their citizens. This is despite the sworn oath every officer takes to uphold the constitution and lack of precedence of face recognition misuse. Why does San Francisco even have law enforcement agencies if they cannot be trusted to protect and uphold constitutional rights?
Face recognition is an accurate identification method
As stated in the bill, there is a belief that face recognition is an inaccurate technology. Aside from the obvious contradiction that it cannot both be inaccurate and also used to track, surveil, and oppress citizens, automated face recognition technology is often more accurate than humans.
Automated face recognition accuracy is also improving at staggering rates. Over the last several years there has been over a 1,000x reduction in error rates by face recognition algorithms. And, despite these tremendous improvements, several years ago automated face recognition was already an invaluable law enforcement tool.
With over a decade of use in policing in the U.S., there have been zero documented cases of automated face recognition technology resulting in wrongful arrests. One reason is that face recognition search results are not probable cause for arrest. Another is that law enforcement agencies do not use face recognition as a purely automated technology; human analysts examine the search results in forensic identification. And there are generally strict standards and guidelines followed by law enforcement officials.
There are inaccuracies in every form of identification technology, whether it is latent fingerprint analysis, DNA, ballistic identification, witness testimony, or the countless other tools used in law enforcement investigations. The key is to understand the limits of the technology, and employ workflows that are informed by the strengths and weaknesses of various identification methods. This is already the case with automated face recognition as extensive guidelines exist for proper usage.
Factors that influence face recognition accuracy
There is a belief that face recognition algorithms are “racist” and “sexist”. This is misleading and amplified by non-scientific research articles that generate inaccurate media headlines.
While the accuracy of face recognition can vary due to race, different algorithms and camera types influence accuracy more than race (or skin phenotype). Differences in the angle of the face relative to the camera also have a stronger influence on accuracy, as do many other environmental factors. Face recognition accuracy has been shown to be lower for women than men, but this clearly seems due to the cultural use of makeup.
Biases in the way humans recognize faces
While there are various biases in face recognition algorithms, humans are unfortunately rather terrible at face recognition when they lack familiarity with the person they are viewing. Humans are also quite poor at recognizing persons of races other than their own.
The cognitive limitations of human facial identification have resulted in a wide number of false arrests and convictions due to mistaken witness testimony. The issue is so bad that in roughly 70% of DNA exonerations, eyewitness testimony was one of the pieces of evidence that resulted in the false conviction.
The importance of facial identification
Our facial appearance is arguably the single most public piece of information about us. We will hesitate to provide our name to a stranger when asked, but we will readily let everyone in public view our face. Socially progressive countries such as France and Denmark have even banned the concealment of one’s face when in public.
While the linking of our facial appearance to other personal information needs to be regulated, our facial appearance is the first piece of information provided in nearly every public engagement we have. We simply cannot conceal our face and have acceptance in society.
This is not by mistake either. Facial identification is so important to the human race that we have evolved an entire region of our brain dedicated to the task.
Being misinformed is not an excuse
Face recognition has demonstrated clear benefits in enabling law enforcement to solve thousands of serious crimes over the last decade without a single example of an innocent person being falsely arrested due to a misidentification. This is an astounding improvement over the legacy method of eyewitness identification.
The Board of Supervisors of San Francisco seem to be unaware of these law enforcement use cases, and have instead determined that the propensity for facial recognition to endanger civil rights and civil liberties (which would already be constitutional violations) substantially outweighs its purported benefits.
The bipartisan legislation recently proposed in the U.S. Senate is a great example of an effective approach to regulating face recognition technology. While there are a few areas where the bill needs stronger regulations and restrictions, as we will highlight in a forthcoming article, it is crafted in a manner that will limit the technology’s use for harmful purposes while still allowing all the overwhelmingly positive uses.
Being misinformed is not an excuse for endangering citizens, and the San Francisco city lawmakers need to properly justify their decisions. It is hoped that this article can serve as a rational discussion point moving forward amid misinformation that has now confused a growing number of elected officials.
- How Automated Face Recognition Works
- Emerging Trends in Commercial Face Recognition
- 10 Steps for Selecting a Face Recognition Algorithm