Facebook apologized when its artificial intelligence hit a cheeky tag on a video featuring black men. According to the New York Times, users who recently watched a Daily Mail video featuring black men saw a message asking if they wanted to continue watching primate videos. In the statement sent to the publication, the social network lamented an “unacceptable mistake”. He has also disabled the recommendation function responsible for the message, as he is looking for a reason to avoid repeating such serious mistakes.
Company spokesman Dani Lever said in a statement, “As we said, while we’ve made improvements to artificial intelligence, we know it’s not perfect and we’re still making progress. We apologize to anyone who saw these offensive recommendations.”
Gender and racial distortions in artificial intelligence are not a unique problem for the social network: facial recognition techniques are not far from complete and tend to misidentify POCs and women in general. Last year, fake facial recognition competitions led to the false arrests of two black men in Detroit. In 2015, Google Photos called photos of blacks “gorillas,” and Wired discovered a few years later that the tech giant’s solution was to censor the word “gorilla” for image and tag searches.
The social network shared the dataset it created a few months ago with the artificial intelligence community to try and solve the problem. It featured over 40,000 videos featuring 3,000 paid actors with their age and gender at the company. Facebook has even hired professionals to illuminate its photos and tag skin tones so that artificial intelligence systems can learn what different ethnic groups look like under different lighting conditions. It was clear that the dataset was not enough to completely dissolve the illusion of artificial intelligence on Facebook, proving once again that the artificial intelligence community still has a lot of work to do.