“I was locked up for no reason,” Nijeer Parks told The New York Times last month.
Parks was recounting his arrest for allegedly shoplifting from a New Jersey Hampton Inn gift shop, giving officers a false ID and then making a run for it in a rental car. Except that Parks had proof that he was 30 miles away when all of that happened. But police were convinced they had the right man — they used facial recognition to match the Nijeer Parks’ face to that of the Black man pictured on the phony ID.
Parks’ story backs up activists’ concerns about the critical flaws of artificial intelligence, specifically facial recognition. Studies have confirmed that facial recognition routinely misidentifies women and especially people of color. Which is why I’m alarmed that Gov. Charlie Baker insisted that it be included in the recently passed police reform bill. State lawmakers passed the bipartisan bill after months of intense negotiation and deliberately excluded facial recognition. But Gov. Baker made it clear he would only sign the bill with a facial recognition amendment — despite widespread vocal opposition from Black and brown lawmakers, the ACLU and members of the Celtics basketball team.
Police unions in Massachusetts lobbied for facial recognition to be included, claiming the methodology has been successful in identifying criminals. And that is true in some cases. But, what is also true is that this technology has a dangerous bias. As the National Institute of Standards and Technology reported, recognition technology misidentifies Asian and African American faces 10 to 100 times more than white faces — even more for Native American faces. In a test sample, 28 congressional members, mostly people of color, including the late civil rights icon Rep. John Lewis, were positively and erroneously identified with mugshots from a law enforcement database.
No one understands the limitations of this flawed technology more than MIT’s Joy Buolamwini, the founder of the Algorithmic Justice League and a top expert in artificial intelligence. Buolamwini has used her extensive research documenting the inherent bias in AI to persuade corporations like Microsoft, Amazon and IBM not to sell facial recognition technology to police departments. And last year, I highlighted the fourMassachusetts cities, including Cambridge, which banned facial recognition’s use.
Clearly the amendment included in the law goes against the tide of ever-widening acceptance that facial recognition — in its current form — can be harmful. Given its potential for harm, the now police-sanctioned use of the technology stands to undermine the other positive measures in the police reform law, one of the reasons state Sen. Sonia Chang Diaz called the compromises in the legislation “heartbreaking.”
The police reform law is absolutely a significant step toward addressing the concerns of citizens who’ve often suffered at the hands of violent police officers with little recourse. But Nijeer Parks' story is a sharp reminder of what’s at stake with facial recognition. The 33-year-old who was jailed before his case was finally dismissed notes how scary the misidentification was, saying, “I just never thought it would happen to me.”
