Forbes: Congress Gets A Demonstration of Facial Recognition's Biases
It’s hard to get lawmakers to understand privacy issues, let alone commit to taking action. Yet in one day, the American Civil Liberties Union (ACLU) pulled off an incredible feat. They showed the public how facial recognition technology can lead to false and biased accusations. By the afternoon, two prominent members of Congress sent a letter to Amazon CEO Jeff Bezos demanding an immediate meeting.
The Congressmen were angry at the results of a report which showed that Amazon’s facial recognition software, Rekognition, had identified 28 legislators as potential criminals. The ACLU conducted the study by comparing the photos of each member of Congress to a collection of 25,000 arrest photos. Most of the wrongly identified were non-white – creating an enormous controversy.
The ACLU’s intention was to put a spotlight on news that Amazon was selling this facial recognition technology to law enforcement agencies around the country.
The methodology used was far from perfect. The dataset had its own flaws, and as Amazon pointed out in its response, the confidence threshold selected was not appropriate for the situation. However, it got the point across. For only $12.33, the civil liberties group made a compelling case for why facial recognition technology isn’t ready to be used in our criminal justice system.
There’s three reasons why this incident struck a nerve: it approached the issue of bias in facial recognition technology in a novel way, made an abstract concept more tangible, and personally targeted the people who have agency over these matters.
Amazon poses a unique challenge for privacy activists. The company has excellent operational skills, an enormous cloud infrastructure, and an extensive track record commercializing technologies at a large scale. Despite warnings from investors and employees, there doesn't appear to be enough of an effort devoted to mitigating bias. Amazon isn’t the only company selling facial recognition technology -- but it is the most prominent. And it’s already facing scrutiny from the Trump Administration. Instead of choosing one of the many lesser known competitors, the ACLU picked a target that is easy to criticize openly.
Additionally, it becomes a lot easier to understand how something can go wrong after it actually happens. For years researchers have been cautioning the industry to pay attention to the inability of algorithms to properly differentiate between people with darker skin and the racial disparities in training data. But facial recognition technology is being sold and deployed even though these issues haven’t been addressed properly. The danger is that we are automating our worst prejudices. This report proves that the rank and status of a person does not matter, and it will not get better until we address these problems in a comprehensive way.
Finally, politicians have historically worked to protect privacy when it’s their own information at stake. We’ve had a situation like this once before: during the 1987 Senate Supreme Court confirmation hearings for Robert Bork, a reporter was able to obtain and publicize the judge’s video rental history. The revelations were somewhat boring and didn’t reveal anything salacious, but lawmakers worried that they might be next. The following year, Congress passed the Video Privacy Protection Act which criminalizes improper disclosure by “video tape service providers.”
There’s an opening for history to repeat itself. Republicans and Democrats were both impacted by this incident. The time is now to take bipartisan action to make sure this technology does not lead to widespread discrimination.