AI facial recognition Differences in recognition accuracy by race National research conducted by the National Institute of Research on December 21, 4:57
AI = Face recognition technology using artificial intelligence is progressing, and the national research institute in the United States examined the accuracy of 189 face recognition technologies that have been put to practical use in each country. Announced that the accuracy of was found to be low.
The National Institute of Standards and Technology of the United States announced the result on the 19th, analyzing the accuracy of 189 face recognition technologies that are widely spread and put into practical use in each country.
According to it, the accuracy of each recognition was analyzed using approximately 8.5 million face photos such as passports held by American government agencies, 18 million copies, and compared to white people in East Asia and Africa The accuracy of recognition was low.
In addition, women are more likely to be mistakenly recognized than men.
On the other hand, the face recognition technology developed in China has a relatively high ability to recognize East Asians.
Differences in the accuracy of face recognition technology due to race and gender are often pointed out in the academic world, and the main factor is the AI calculation processing method that forms the basis of the technology = bias in the data used when designing the algorithm It is.
The laboratory points out that “users who introduce face recognition technology should be used with the fact that there are differences depending on race and gender, and should continue to improve accuracy”.