WASHINGTON—In response to new claims from the American Civil Liberties Union of California (ACLU), which has run pictures of California state legislators through a facial recognition program that matches images to a database of criminal mugshots, the Information Technology and Innovation Foundation (ITIF), the world’s leading think tank for science and technology policy, released the following statement from ITIF Vice President Daniel Castro:
The ACLU is once again trying to make facial recognition appear dangerous and inaccurate. But independent testing from the federal government has consistently shown that facial recognition technology is highly accurate. It now exceeds the accuracy of humans at identifying faces.
This is the second time the ACLU has released misleading findings. Last year, it used dubious methods to claim that facial recognition had high levels of inaccuracy, but it generated false matches by setting an artificially low confidence threshold of 80 percent instead of 99 percent. The ACLU claimed at the time that companies like Amazon were not clear about what the threshold should be. That wasn’t true then, and it isn’t true now. In the past year, Amazon has repeatedly stated that any sensitive application of facial recognition, such as for law enforcement purposes, should only be using high confidence thresholds. So, for the ACLU to repeat this kind of test a year later, while apparently not changing its methods—and still refusing to share its data—is disingenuous and misleading. Claims that are not observable, testable, repeatable, and falsifiable are not science. It’s agenda-driven public relations, and policymakers should ignore it.
For additional background on the ACLU’s facial recognition testing, see:
- “Banning Police Use of Facial Recognition Would Undercut Public Safety,” by Daniel Castro and Michael McLaughlin, ITIF, July 30, 2018