The problems with Amazon’s facial recognition technology
Amazon’s problems with Rekognition began last June, when it transpired that the company was selling the facial recognition technology to some US government agencies, among them the Immigration and Customs Enforcement Service (ICE), responsible for separating children from their families and other barbaric acts in the name of protecting the country’s borders. Amazon faced protests from its employees, who demanded that the company take an ethical stance and cease working with ICE.
In fact, Amazon’s problems began a month earlier: the American Civil Liberties Union (ACLU) launched an awareness campaign after an Amazon AWS blog post mentioned that Rekognition was being used by the Washington County Sheriff’s office. The ACLU warned of the dangers its use by the authorities posed, in large part because of the risk of mistaken identity.
The ACLU decided to take the issue to Congress, doing so in an intelligent way that has allowed it not only to further raise the visibility of the issue, but also to guarantee a response: it applied the technology to the official photographs of members, which was wrong in 28 out of 535, identifying them as criminals. Republicans and Democrats, men and women of different ages were among those mistakenly identified as felons, and notably 40% of them were people of color, despite the fact that 80% of members of Congress are white.
The ACLU’s approach is brilliant, proving to representatives that they could be falsely identified as criminals if Rekognition is routinely used by the authorities. In response, the members of Congress concerned have called on Amazon to explain. A “Jeff, we need to talk” call should mean changes to how the technology is adopted and hopefully will spark a much-needed debate on the subject.
Amazon has defended its technology, blaming the disappointing results to a poorly calibrated algorithm. The ACLU used the algorithm through AWS, which Amazon recommends, at a cost of $12.33. The claims Rekognition is right 95% of the time, but the ACLU says 80% is closer to the mark. The point here is that its demonstration has worked, meaning greater scrutiny of the use of Rekognition and similar technologies from now on, whilst others, such as Microsoft, strives for greater perfection.
We should all be grateful for the work of organizations like the ACLU in defending our civil rights. Debate on the use of technologies such as facial recognition is extremely important and in many ways defines the type of society we want to live in. This is not a “to use or not to use” question: advanced technologies like this cannot be ignored, and much less when it can be improved by being fed more data. That said, it’s pretty clear that facial recognition technology is not yet ready for use by law enforcement, given the dangers of discrimination. But once the technology does work, then we certainly need to discuss just what we’re going to use it for, under what circumstances, and what kind of society we’re heading toward if we do.
(En español, aquí)
