Amazon Rekognition's face recognition system took 28 US congressmen for criminals

The American Civil Liberties Union (ACLU) continues its campaign against the use of face recognition systems by federal agencies and the police. ACLU insists that the quality of these systems is too low for real use. As a result, there will be a lot of false positives, which innocent people will suffer. Trying to convince the US Congress to ban the use of these systems, human rights activists launched a daring but effective action: they drove all American congressmen through the Amazon Rekognition face recognition system . The result was a bit predictable: the system recognized 28 congressmen as criminals . Photos of "heroes" in the screenshot above.

Amazon aggressively promotes its face recognition system. Recently it became known that the corporation has entered into contracts with police units of Washington County, Oregon and Florida. CEO Jeff Bezos is positioning Rekognition as an effective tool for face recognition in real time, including through video streaming, which comes from wearable video cameras on police uniform.

This is reminiscent of the system, which this year began to be tested in some cities of China, including in the city of Zhengzhou (Henan Province in the east of the central part of China). There, the police were given special glasses with video cameras, which are also connected to the software for face recognition.



Gadgets were given primarily to the traffic police officers who work at the busy Zhengzhou East Railway Station, because most people pass by before their eyes. The glasses are specially designed for the police and connected to a handheld computer. After scanning the passerby’s face, the computer connects to the central database where the search for matches is performed. The experiment proved to be effective: by February 6, the police managed to identify seven fugitives who are accused of hiding from the scene of an accident and human trafficking.

But in China, the protection of human rights is more condescending: 170 million surveillance cameras are already working on the streetsthat are connected to a single face recognition system. False positives are not particularly confused by anyone if the system helps to catch real criminals.

But American human rights activists believe that false positives are evidence that the system is not yet ready for use.

The results of the ACLU tests showed that Rekognition mistakenly took the images of 28 innocent congressmen as criminals. Interestingly, the false positives are distorted towards black-skinned people. 39% of the “affected” congressmen, although in reality black congressmen constitute only 20% of the lower house of parliament. These results confirm the concerns of the representatives of black groups, which are set out in the letterCongressional Black Caucus addressed to Amazon that using the face recognition system could have “deep negative unintended consequences” for the black population, illegal migrants and protesters. In fact, there are signs that the system is often mistaken when comparing black people, especially women.

Well, in general, everyone got under the distribution - both Democrats, Republicans, men and women of all ages from different regions of the country. They all fell into the number of system errors.

Representatives of the ACLU believe that every false positive act endangers a person’s freedom. A police officer may discriminate against such a person by subjecting him to an interrogation or search.



The scan was made on the public Amazon Rekognition service, the cost of the service was $ 12.33. For comparison with photos, 535 congressmen and senators selected 25,000 publicly available photos of people in handcuffs. The test was conducted with the default settings. 28 errors mean about 5% of false positives. Theoretically, this may be a good result, but if you apply it on thousands of people, it means a lot of innocent people who are searched because of a program error.

Of course, all the results of the program will be tested by people, but we all know that such systems can quickly get out of control and go into automatic mode, as happened with the system of issuing speeding tickets.

The ACLU called on Congress to “introduce a moratorium” on the use of this technology until it begins to produce acceptable accuracy.