Facial recognition is proven to be inferior for black people. And, for the first time, a technology was sufficiently capable of showing 99,84% accuracy in these cases. She started from startup CyberLabs, located in the Rio de Janeiro. The other similar technologies that already operate in the market indicate a level of correctness that varies between 75,8% and 87,5%, which proves the frequent failure of the system in black peoplemainly in women.
The Brazilian company used the Racial Faces in the Wild (RFW) database, which has more than four million faces of different ethnicities, to help in studies that prove algorithmic racism, which has already been blamed for condemnation of innocents, or research that increases the accuracy of the technology.
For Marcelo Sales, president of CyberLabs, the technology was created to meet the demand for diversity present in the Brazilian population. The accuracy of the startup's easy recognition system also involves indigenous and Asian faces. Sales also informed that this diversity was also reflected in the information used to feed the system.
- Man sues police after being mistakenly arrested for facial recognition failure
- Doria vetoes facial recognition in the São Paulo Metro
- Microsoft fixes bug that corrupted NTFS drivers in Windows 10
Named as KeyApp, the new facial recognition system has been in development for three years and reached the market experimentally in March 2020 with an accuracy rate of 90,98%. However, the technology advanced after receiving support from the platform aimed at accelerating the training and conditioning of artificial intelligence called Nvidia Enterprise. CyberLabs pointed out, in an official statement, that the partnership brought surprising speed to the development of the system.
The KeyApp is being provided free of charge to public entities and has already been implemented at the Rio de Janeiro Operations Center and at Santa Casa de Itajubá, in Minas Gerais. The tool is also available in an app on the App Store (iOS) and Google Play Store (Android).
Have you watched our new videos on YouTube? Subscribe to our channel!