In the wake of the American Civil Liberties Union’s test where 28 members of Congress were wrongly identified as arrestees via its Rekognition facial recognition service, some of those lawmakers have now raised serious concerns to the tech giant.
For now, Amazon has neither answered their questions, nor Ars’.
“We request an immediate meeting with you to discuss how to address the defects of this technology in order to prevent inaccurate outcomes,” wrote Rep. Jimmy Gomez (D-California) and Rep. John Lewis (D-Georgia), in a letter to CEO Jeff Bezos.
Those two members of the House of Representatives were among the 28 that were falsely identified as being amidst a group of 25,000 mugshots.
The 28 also included Sen. Edward Markey (D-Massachusetts), Rep. Luis V. Gutiérrez (D-Illinois), and Rep. Mark DeSaulnier (D-California), who separately wrote to the company on Thursday, asked the company to answer specific questions and requests, with a deadline of August 20.
Please describe in detail how Amazon tests for facial recognition accuracy, how often Amazon tests, and whether these results have been independently verified.
Please also describe in detail show Amazon tests for bias in its facial recognition results, especially racial bias.
Please provide a list of all law enforcement or intelligence agencies that (1) Amazon has contacted or otherwise communicated with regarding acquisition of Rekognition and (2) currently use the Rekognition service. In addition to your response to this letter, we encourage Amazon to include a list of government Rekognition customers in its next transparency report.
Amazon did not respond to Ars’ question as to whether Bezos or any other company executive would be meeting with these concerned members of Congress.
Another legislator, Rep. Emanuel Cleaver (D-Missouri), who was not among the 28 identified, also brought up other issues:
This is exactly what @keithellison and I warned about. Also why I remain extremely concerned that @amazon‘s Rekognition will further exacerbate issues of racial bias in society, and particularly where deployed by law enforcement. https://t.co/8CgUxeIujn
— Rep. Emanuel Cleaver (@repcleaver) July 26, 2018
In a statement sent to Ars late Wednesday and other media outlets on Thursday, Amazon said that its 80 percent confidence threshold was not precise enough for human faces, a point that the ACLU pushed back on.
Jacob Snow, an attorney with the ACLU of Northern California, said in his own statement on Thursday:
In addition to remaining silent on these very real concerns that members of Congress, community groups, and Amazon’s own employees, shareholders, and consumers have raised repeatedly, Amazon is acknowledging that Rekognition – a product that it aggressively markets to law enforcement – can and will misidentify people by default. That’s downright dangerous, and there’s more:
We know from our test that Amazon makes no effort to ask users what they are using Rekognition for. Instead, Rekognition sets one default: the same 80 percent we used in running our test.
We also know that Amazon’s website, right now, shows the use of an 80 percent confidence for recognizing human faces. It shows that Amazon is recommending an 80 percent confidence score in ‘Face-Based User Verification.’ If an 80 percent threshold is not ‘appropriate for identifying individuals with a reasonable level of certainty,’ why is Amazon highlighting that confidence level for recognizing human faces?
On Thursday afternoon, Jesse Freund, an Amazon spokesman, contacted Ars to say that our earlier reporting on the ACLU report was “inaccurate.” However, he declined to explain further without going “on background,” which we declined. He also did not explain why he would not speak on the record.