Detroit: Police issue new guidelines for facial recognition after unlawful arrests
![Detroit Police Department](https://cdn.posteo.de/news/2024/07031700/DPD.jpg)
The city of Detroit has imposed stricter regulations on the Detroit Police Department (DPD) for how automatic facial recognition can be utilized. The regulations were revised as part of a court settlement in which a man was wrongfully arrested and detained in the Michigan District Court for 30 hours as a result of faulty automatic facial recognition in 2020, the authorities stated last Friday. In response, the civil rights organization, American Civil Liberties Union (ACLU), sued the Detroit Police Department on behalf of the victim in 2021.
Stephen Lamoreaux, head of informatics at Detroit’s Crime Intelligence Unit said the city now has the strictest regulations in the country. The police stated on Friday that they were satisfied with the changes and are convinced that it will be recognized nationally as an example for how facial recognition should be ideally handled. The aim was to ensure – much like the plaintiff ACLU – that police work would be fair, just and conducted in compliance with the constitution. In the future, automatic facial recognition should only be used for serious criminal offences like assault and battery, murder and burglaries.
For instance, the new regulations require that an investigator must first receive approval from at least two additional people before the technology can be implemented and it is no longer permitted to arrest suspects based solely on results from facial recognition software or from photos. There must be additional “physical proof” such as traces of DNA at the crime scene or a registered cell phone of the accused within the cell site of a crime scene, for example. A line-up of suspects – also in the form of displayed photos – with evidence based solely on automatic facial recognition also no longer allowed and requires further evidence. If these requirements are met, the officer who shows potential witnesses the photos is not allowed to know who the suspect is themself.
New guidelines and requirements for documentation
In an effort to promote transparency, the police are now required to provide information about the weaknesses of facial recognition technology and to disclose when it has been used during an arrest. Additionally, officers must disclose if facial recognition did not determine any suspects or if the results suggested other suspects. All cases since 2017 are to be investigated, audited and reported to the District Attorney in which arrest warrants were issued based on automatic facial recognition.
Additionally, police departments must organize training courses that explain the risks and dangers of the technology and the high rate of misidentifications with people of color.
According to the New York Times (NYT), the guidelines are going into effect this month – a federal court is to enforce it for four years. Phil Mayor, an attorney at ACLU, explained to the NYT that the court settlement makes the DPD a leader nationwide for adhering to strict regulations – after he previously spoke of the best-documented abuse of facial recognition technology.
Part of the closed settlement also includes a payment of 300,000 US dollars in damages to the victims. The Detroit City Council already approved the settlement in May.
The case
The plaintiff was arrested in front of his family as a result of misidentification and detained for 30 hours. Facial recognition software identified the Black man, Robert Williams, as a potential suspect of a robbery at a jewellery store in 2018. The system compared drivers license photos and mug shots with blurry recordings of the perpetrator from a private security camera.
The program compared the face of the perpetrator from the security camera footage with the existing data and found 243 supposed matches that showed similarities with the person they were looking for. The software sorted the presumably matching photos by probability to reduce the amount of suspects. According to police documents that were disclosed to Williams as part of the lawsuit, a photo of the plaintiff from an expired drivers licence was in ninth place on a ranked list. The investigator in charge found Williams to be the most matching fit and sent a report to the Detroit Police.
Apparently Williams’ photo was included in a selection of six different portraits of people that were shown to the owner of the security camera. She confirmed that Williams looked most similar to the perpetrator. Afterwards, an arrest warrant was issued.
Other victims also pressed charges
To date, there have been three cases made public of arrests based faulty automatic facial recognition in Detroit.
In one case, a Black woman was accused of taking part in a robbery and robbing a man in February 2023. Automatic facial recognition software from the police identified her as a potential suspect. After a witness indicated that she recognized her on a photo, Porcha Woodruf was arrested in front of her children while in her eighth month of pregnancy.
The police detained her for eleven hours and interrogated her. Woodruff was also charged in court with robbery and theft. The mother of several children was only set free after paying bail of 100,000 US dollars. A month later, the public prosecutor dismissed the case against her and discontinued it.
Afterwards, Woodruff sued the city of Detroit and a police officer. She accused the officer of violating her fundamental rights and discriminating her based on skin color – because they implemented facial recognition software that has been proven to frequently misidentify Black citizens more than others.
Discriminating technology
For a long time, civil rights activists have criticized the use of automatic facial recognition by law enforcement agencies for its inherent flaws and unreliability – especially when attempting to identify Black people. The likelihood that innocent people coming under suspicion is considered high.
At least seven people have been unlawfully imprisoned in the USA because the police relied on incorrect facial recognition results, as explained by the American Civil Liberties Union (ACLU) in April. In almost all cases, Black people were the victims.
Among other things, a study commissioned by the US government and conducted by the National Institute of Standards and Technology at the end of 2019 found that the error rate for people with dark skin was 10 to 100 times higher than with people with lighter skin. Women with dark skin color had the highest rate of misidentification. Indigenous people and people of Asian descent were also frequently affected.
Earlier investigations also produced similar results. As a consequence, US cities like San Francisco and Somerville officially forbade the use of facial recognition.
A warning to stop
Nevertheless, the DPD is convinced of the usefulness of automatic facial recognition. Last year, James White, the Police Chief of Detroit reported to the NYT that the technology helped remove 16 murderers from the streets. However, police officials did not provide specific details or further information regarding the cases when asked. White blames the unlawful arrests on human error. His officers relied too much on suggestions made by the facial recognition software and too little on their own judgment.
For Molly Kleinman, Director of the Science, Technology, and Public Policy program at the University of Michigan, the new protective measures sound initially very promising but she remains skeptical, according to the NYT. Surveillance cameras are everywhere in Detroit if the technology really did what it promised, Detroit would be one of the safest cities in the country.
Willie Burton, Police Commissioner and member of the supervisory authority that approved the new regulations, also considers the new guidelines a step in right direction. However, as in the past, he rejects the use of automatic facial recognition technologies. The technology is simply not advanced enough, Burton explains. A false arrest is one too many and having three in Detroit should be a warning to stop. (hcz)