Amazon's facial 'rekognition' software matches Congress to criminals
Amazon’s facial recognition software wrongly matched dozens of members of Congress to mugshots of other Americans with criminal records, in a test run by the ACLU. Among the 28 lawmakers wrongly matched with mugshots were civil rights icon Rep. John Lewis, major immigrant-rights figure Rep. Luis V. Gutierrez and three senators.
The American Civil Liberties Union did the test, using Amazon’s Rekognition software, to try to show the faults in the tool, which civil libertarians have been complaining about. The ACLU said black and Latino lawmakers were more likely than the average member of Congress to be matched with a mugshot. “Congress must take these threats seriously, hit the brakes, and enact a moratorium on law enforcement use of face recognition,” the ACLU said. “This technology shouldn’t be used until the harms are fully considered and all necessary steps are taken to prevent them from harming vulnerable communities. The ACLU built a database of 25,000 publicly available mugshots then ran photos of members of Congress against its database using Rekognition. The test used the default setting of an 80 percent “similarity” score between two photos to produce a hit.
Amazon Web Services questioned the ACLU’s methodology, saying the 80 percent similarity score “is an acceptable threshold for photos of hot dogs, chairs, animals, or other social media use cases, it wouldn’t be appropriate for identifying individuals with a reasonable level of certainty.” The company said it recommends using at least a 95 percent threshold for important tests such as law enforcement activities. The company said it recommends using at least a 95 percent threshold for important tests such as law enforcement activities. The company also said Rekognition is generally used as a narrowing tool, not a final determinator. They expect people to use the results to inform decisions.
The company did not address the ACLU’s findings for minorities: Nearly 40 percent of the lawmakers with false matches were racial or ethnic minorities, though just 20 percent of Congress falls into those categories. Women fared much better. Only one of the 28 false matches was a woman — Rep. Norma Torres. Rekognition also seemed to have a slight bias against Democrats: Though Republicans are a majority in both the House and Senate, 15 of the 28 lawmakers falsely flagged were Democrats. Given the ACLU’s findings on black and Latino false hits, that could be explained by the higher ratio of minorities within the ranks of Democratic lawmakers.
https:// www.washingtontimes.com/news/2018/jul/26/amazons-facial-rekognition-software-matches-congre/