IBM’s choice to end commercial sales of facial recognition may increase pressure on Amazon and Microsoft to halt contracts with law enforcement agencies. Algorithmic Justice League (AJL) founder Joy Buolamwini said she already sees signs pointing in that direction. On Tuesday, for example, the same day as George Floyd’s funereal, more than 250 Microsoft employees urged CEO Satya Nadella to end contracts with police.
“Google has stepped back from facial recognition, as has IBM, and we already see Microsoft workers mobilizing to demand change,” Buolamwini told VentureBeat Tuesday shortly after testifying in front of the Boston City Council in favor of a proposed facial recognition ban. Hours after this article was published, Amazon imposed a one-year moratorium on facial recognition use by police.
Amazon and Microsoft remain the only two major tech companies actively attempting to sell facial recognition software to governments and law enforcement agencies.
Teaming up with Google AI researcher Timnit Gebru in 2018 and AI Now Institute researcher Deborah Raji in 2019, Buolamwini published the Gender Shades project, audits of facial recognition from companies like Amazon, IBM, Face++, and Microsoft that found that they perform best for white men and worst for women with darker skin tones. Arguing that even perfect facial recognition can be misused, Gebru made the case for a facial recognition ban in an interview published this morning by the New York Times.
“We have reached a moment of reckoning where more people are re-evaluating law enforcement and the capacity for abuse that can be accelerated with high-tech tools,” Buolamwini said. “The moment we are in is a wake-up call for all tech companies working with police. At AJL, we support halting face surveillance and urge Microsoft, Amazon, and others to follow IBM’s lead in not equipping police with facial recognition technology for surveillance, racial profiling, and other harmful abuses.”
Thanks to the Gender Shapes project these three black women AI researchers coauthored, knowledge of race and gender bias is far more common today among lawmakers considering regulation in state legislatures and Congress. Earlier this year and in multiple hearings in 2019, facial recognition regulation received support from a bipartisan group of lawmakers in Congress.
On Tuesday, Buolamwini and the AJL called for major tech companies to commit $1 million to racial justice tech organizations like Data for Black Lives and Black in AI. AJL also urged facial recognition companies to sign the Safe Face pledge to not sell facial recognition to law enforcement agencies or other actors with power to use lethal force.
Earlier this week, IBM CEO Arvind Krishna said his company will no longer sell or research facial recognition and joined the calls for regulation other major tech companies, including Amazon, Google, and Microsoft, are making. Google previously stated it has no plans to sell facial recognition without additional regulation and earlier this year supported calls for moratorium.
IBM halted face detection for a publicly available API last year, and independent analysis found that IBM has no publicly known facial recognition contracts in the U.S. However, IBM’s decision carries some symbolic significance, given the company’s research with NYPD and history of working with Nazi Germany during World War II.
By contrast, Microsoft advocates for facial recognition in California and Washington legislatures, and AWS CEO Andy Jassy has previously stated that his company will sell facial recognition to any government so long as it’s legal. In May 2019, Amazon shareholders formally rejected a halt of facial recognition sale to governments.
Research published last year by the National Academy of Sciences found that police violence is a leading cause of death for black men in the United States. George Floyd, whose killing led to some of the longest and largest protests in recorded history, was buried in his hometown of Houston on Tuesday.
In December 2019, the Department of Commerce’s National Institute of Standards and Technology (NIST) conducted its first facial recognition vendor test based on race. In an analysis of algorithms from almost 100 companies, the test found systems misidentified the faces of people of Asian or African descent 10 to 100 times more often than white faces.
NIST Information Technology Laboratory director Dr. Charles Romine testified in January before the House Oversight and Reform Committee that Amazon was in talks with NIST for Rekognition to participate in its facial recognition vendor test. A NIST spokesperson today told VentureBeat that Amazon has not submitted any algorithm for evaluation under its Facial Recognition Vendor Test program.
Though CEO Jeff Bezos told an All Lives Matter customer they aren’t wanted this week, Amazon has a flawed past. Similar to criticisms made of an Amazon fulfillment center worker who executives attempted to describe as “not smart or articulate,” in January 2019 AWS VP of AI Matt Wood took the unusual step of criticizing the Gender Shades analysis in a blog post, an apparent attempt to discredit the accuracy of the work. Gender Shades was subsequently defended by a group of nearly 80 prominent AI researchers, including Yoshua Bengio, the deep learning pioneer and supporter of equality who is among the most-cited scholars in the world today.
For their work to reveal facial recognition performance disparities based on race and gender disparities in commercially available facial recognition systems, in July 2019, VentureBeat presented Buolamwini, Gebru, and Raji with its inaugural AI Innovation award in the AI for Good category. The trio are among a range of black women being mentioned in social media this week with the #CiteBlackWomen hashtag. To call more attention to the lack of citation and media coverage of women scholars as well as algorithmic bias, earlier this year the AJL, together with seven prominent women in tech and AI, launched the Voicing Erasure project.
Updated at 12:56 pm Pacific to include NIST spokesperson comment on Amazon participation in U.S. Department of Commerce’s National Institute of Standards and Technology (NIST) Facial Recognition Vendor Test program.
The audio problem: Learn how new cloud-based API solutions are solving imperfect, frustrating audio in video conferences. Access here