Tomcat
Professional
- Messages
- 2,695
- Reaction score
- 1,072
- Points
- 113
This conclusion was reached by the National Institute of Standards and Technology, according to Reuters. The audit found that facial recognition systems incorrectly identify blacks and Asians ten to one hundred times more often than whites. This means that when the technology is applied at airports, customs, and on the streets, black people and Asians can be unfairly detained by police officers.
The researchers also noticed a bias towards dark-skinned women, who are more at risk of unfairly detained than white women, simply because the system does not know how to accurately recognize their faces.
Who was checked?
The researchers tested 189 facial recognition systems from 99 different developers from around the world, however, Amazon did not participate in the study, which did not provide its algorithm. The Chinese startup SenseTime, valued at $ 7.5 billion, showed "a high level of false coincidences for all comparisons", the Chinese Yitu system showed itself slightly better, but scientists there, too, noticed a racial bias.
After analyzing Microsoft's facial recognition, the researchers found that the algorithm was ten times more likely to misbehave for black women than for black men. Microsoft said it is reviewing the report and commenting later.
Reaction
Rep. Benny Thompson, chairman of the US House of Representatives Security Committee, said the study was worse than he feared as customs officials continued to install facial recognition at checkpoints. “The administration should reconsider its plans for facial recognition technology in light of these shocking results,” he said.
The researchers also noticed a bias towards dark-skinned women, who are more at risk of unfairly detained than white women, simply because the system does not know how to accurately recognize their faces.
Who was checked?
The researchers tested 189 facial recognition systems from 99 different developers from around the world, however, Amazon did not participate in the study, which did not provide its algorithm. The Chinese startup SenseTime, valued at $ 7.5 billion, showed "a high level of false coincidences for all comparisons", the Chinese Yitu system showed itself slightly better, but scientists there, too, noticed a racial bias.
After analyzing Microsoft's facial recognition, the researchers found that the algorithm was ten times more likely to misbehave for black women than for black men. Microsoft said it is reviewing the report and commenting later.
Reaction
Rep. Benny Thompson, chairman of the US House of Representatives Security Committee, said the study was worse than he feared as customs officials continued to install facial recognition at checkpoints. “The administration should reconsider its plans for facial recognition technology in light of these shocking results,” he said.
