This archive report was first published on 20 December 2019.
December 20, 2019, marked the release of a US government study that shed light on the inaccuracies of facial recognition systems, particularly for non-white individuals.
The study, conducted by the National Institute of Standards and Technology (NIST), found that facial recognition algorithms produced false positive rates for Asian and African American individuals up to 100 times higher than for whites.
Researchers also discovered that two algorithms incorrectly assigned the gender of black females nearly 35% of the time.
These findings come as facial recognition technology is increasingly being used in various sectors, including law enforcement, airports, border security, banking, retailing, schools, and personal technology.
Activists and researchers have expressed concerns that the potential for errors is too great, and that mistakes could result in the jailing of innocent people, as well as the creation of databases that may be hacked or inappropriately used.
According to lead researcher Patrick Grother, false positives can lead to serious consequences, including lengthy interrogations, watchlist placements, and false arrests.
However, the study also found that some algorithms developed in Asian countries produced similar accuracy rates for matching between Asian and Caucasian faces, suggesting that disparities can be corrected with more diverse training data.
Despite this, Jay Stanley of the American Civil Liberties Union stated that the new study highlights the technology's flaws and biases, and that it is not ready for wide deployment.
