Facial recognition reboot

The Detroit Police Department has agreed to restrict officers’ use of facial recognition technology to solve crimes. The revision is part of a federal settlement with Robert Williams, who was wrongfully arrested in 2020. Officers had detained Williams after using the software to search for a shoplifter who robbed a store in 2018.

Detroit detectives ran a low-quality photo from the store’s surveillance system through the department’s facial recognition software, according to a lawsuit filed by the American Civil Liberties Union on Williams’ behalf. Michigan State Police tentatively matched the photo to an expired driver’s license photo of Williams, the lawsuit claimed.

Authorities held Williams for 30 hours before releasing him on a $1,000 bond. The prosecutor dropped the case during a hearing two weeks later, and city officials cleared Williams’ record and agreed to pay him $300,000 in damages.

Under the new department policy made public June 28, police cannot make arrests based only on facial recognition results. They also cannot conduct lineups based on the technology alone without other reliable evidence.

The department must also conduct training on the software and its potential to misidentify suspects. It will conduct an audit on all cases after 2017 in which officers used the technology to secure an arrest warrant.

Some cities, including Boston and San Francisco, already restrict or ban the use of facial recognition technology. But the Detroit City Council voted in 2020 to continue using it.

At least seven people across the country have been wrongfully arrested based on facial recognition software, according to the ACLU. Last year, the U.S. Government Accountability Office found that seven federal agencies used the technology during criminal investigations between 2019 and 2022. Three agencies had stopped using the software by April 2023, and two implemented training requirements for the programs.

No federal laws currently govern the use of facial recognition programs by law enforcement.

Say what?

Google is nearly doubling the number of languages its translation software can ­recognize with the help of an artificial intelligence language model called PaLM 2.

Users will be able to translate 110 more languages on top of the 133 already available. According to Google, some of the languages are spoken by over 100 ­million people, while others are used by small indigenous communities.

Cantonese is one of the most requested languages for Google Translate, according to a company blog post. Manx is a Celtic language from the Isle of Man that almost went extinct before locals revived it. Both will be included in the update along with Punjabi, the most spoken language in Pakistan.

About a quarter of the newly supported languages come from Africa, according to Google. The update is part of Google’s 1,000 Languages Initiative that aims to support the most spoken languages around the world. —L.C.