Facebook has apologised after its AI tagged as primates, a video featuring black men. Facebook said the AI error was “unacceptable”, and disabled the system, reported BBC News.
The social media website launched an investigation into the matter and issued an apology: “We apologise to anyone who may have seen these offensive recommendations.”
This is not be the first time racial bias and discrimination has been detected in AI systems. Google Photos app, in 2015, labelled images of black people as “gorillas”, forcing them to formally apologise and explain how “appalled and genuinely sorry” they were. The Wired had however, reported in 2018, how simple it was to remove the tags from the picture and censor the word “gorilla” in relation to the image. Google spokesperson had claimed that “gorilla” was censored from the searches, and that similar words like “chimp”, “chimpanzee” and “monkey” were also blocked after the embarrassing incident.
In May of this year, twitter admitted racial bias in their previews of cropped image, using a “saliency algorithm”. The image cropping tool preferred to identify white males over black men, compelling the company to immediately apologise and promise its users to fix the error.
Last year, in 2020, Facebook introduced its new “inclusive product council” and equity team for its sister application, Instagram which would examine algorithms that showcase racial biasness. The company has insisted that “We disabled the entire topic-recommendation feature as soon as we realised this was happening so we could investigate the cause and prevent this from happening again”. They claim that they have improved their AI algorithm which is not perfect, but they will continue making progress with it.