The International Biometrics + Identity Association (IBIA) is trying to alleviate people’s fears about facial recognition, especially with regards to accuracy and racial bias. To that end, the organization has published a White Paper that looks at NIST test results to highlight the accuracy of some of the world’s leading facial recognition systems.
In the White Paper – titled Data Analysis of Facial Recognition Technology – the IBIA looks at the test results of Cognitec, IDEMIA, NEC, and Thales, all four of which are IBIA members. The analysis shows that the systems are consistently accurate for both black and white faces, which would seem to dispel any concerns that the public has about racial bias.
However, the IBIA did acknowledge that some algorithms perform better than others, as the NIST has reported in the past. That means that bias is present in many facial recognition systems, and can even extend to factors other than race. For example, the algorithms highlighted in the IBIA White Paper tended to perform better with men than women, regardless of the race of the people in question.
The IBIA is simply arguing that bias is not inherent to facial recognition, and is instead reflective of the quality of different algorithms. Facial recognition systems can be trained to recognize people of any demographic. The top providers have worked to deliver systems that can perform well for a broad population in a wide range of settings, and organizations that want to use facial recognition have a responsibility to choose the most accurate technology available.
The IBIA also suggested that facial recognition algorithms can identify people more accurately than real-life humans. The four systems in the White Paper boasted true identification rates in the 96 to 97 percent range, while humans average a 60 percent rate (with a range of 50 to 90 percent) with the naked eye. Having said that, the IBIA advocated for a combination of automation and human review, noting that the best results are achieved when people work with machines, and not when either is working alone. In that regard, the organization stressed that the responsibility for any errors ultimately lies with the people implementing a system, and that there should always be human oversight for any major decisions.
“Our mission is to educate and advocate for best practices, policies and laws that balance civil rights and liberties with the need to properly identify individuals,” said IBIA Chairperson John Mears. “We look forward to the opportunity to engage more broadly on these important topics and facilitate a thoughtful dialogue on how we can all do better.”
The IBIA is hoping that the White Paper will help make facial recognition technology more comprehensible (and less scary) to the general public, especially in the wake of a growing backlash and a rising number of bans. As it stands, 75 percent of Americans have reported that they have only a cursory understanding of facial recognition technology and best practices.
August 27, 2021 – by Eric Weiss