IBIA Emphasizes Range of Algorithm Quality in NIST’s Report on Face Recognition

The International Biometrics + Identity Association (IBIA) is trying to put a more charitable spin on a recent NIST report that examines the performance of various facial recognition algorithms. The report was based on the NIST’s evaluation of 189 different algorithms.

In its response, the IBIA takes issue with those who raised concerns about accuracy and possible bias, both of which were observed in the NIST report. The IBIA essentially argues that any low-performing algorithm should be ignored, placing the focus back on algorithms that have performed well.

To that end, the IBIA notes that the best are more accurate and more reliable than the human eye. The sheer number of algorithms make it difficult to draw generalizations about facial recognition, especially when considering the stark differences in performance.

The IBIA’s ultimate goal is to promote facial recognition as a law enforcement tool, and to prevent the blanket bans that have become increasingly common in the United States. The organization has already called for San Francisco’s initial ban to be repealed, but that is yet to deter municipal and federal lawmakers from trying to regulate the technology.

The IBIA goes on to claim that “no prudent organization” would use one of the lower-rated algorithms, though that does not necessarily correspond with observed reality. Multiple government agencies – both in the US and elsewhere – have deployed facial recognition algorithms that have performed poorly or demonstrated clear racial biases. For example, New York’s Metropolitan Transit Authority was unable to recognize a single face during its trial, while London police are rolling out an algorithm that may be only 19 percent accurate.

The IBIA is right to point out that facial recognition technology has improved dramatically in the last few years, and that the top algorithms have made significant strides in their efforts to reduce racial and gender bias. Even so, the organization’s attempt to dismiss the concerns of privacy advocates may be slightly overzealous in light of such recent examples. The debate around facial recognition is more contentious than it’s ever been, and advising people to disregard the potential flaws seems unlikely to alleviate the current backlash.

–

February 20, 2020 – by Eric Weiss