According to British newspaper Daily Mail, talk about the tool called Face Depixelizer on Twitter has spread to be tested by users, who have found that the technology is unable to properly treat dark-skinned faces, as the tool has turned a picture of Barack Obama into a white man.
Face Depixelizer relies on an AI tool developed by a team at Duke University, which uses a method called PULSE that searches through AI-generated images for high-resolution faces to match those that look similar to the input image when compressed at the same size.
One of Duke University’s researchers’ team used a machine learning tool with two neuronal networks, one of which evolved the human faces that create artificial intelligence, which simulated the one that was trained on them, and the other takes this output and decides whether it is persuasive enough to be confused with the real thing.
Duke University bragged that its system was able to convert a 16 x 16 pixel image to 1024 x 1024 pixels in a few seconds, which is 64 times the resolution.
Robert Osazoa Ness, a machine learning blogger, tested two images and the results produced faces that looked completely white to people who were not.
While Business Insider indicated that the failure may be due to the data set used for training artificial intelligence, if there is a lack of diversity in the images that are entered into the machine learning algorithm, you will not be able to perform correctly.
The MIT researchers released a report in 2018 revealing the way the AI system collects data, which is often racist and sexual.
“Computer scientists often rush to say that the way to make these systems less biased is simply to design better algorithms,” said lead researcher Irene Chen, the doctoral student who wrote the paper with MIT professor David Sontag and postdoctoral colleague Frederick Johansson.
But algorithms are only good when using the right data, and our research shows that you can often make a bigger difference with better data.