Recently Facebook has apologized after its AI spanked and awful label on a video of Black men. According to The New York Times report, recently, users watched a video posted by Daily Mail in which black men saw a quick inquiry them if they would like to “keep watching videos about Primates.” In this matter, a statement to the public came from social media giant Facebook in which they apologized for this unacceptable error. The technical team of Facebook also immobilized the recommendation feature that is liable for the message as it seems into the cause to avert serious mistakes like this from happening again in the future.
Statement of Dani Lever on Facebook AI mislabels
Soon after the disable of the recommendation feature, the company spokesperson Dani Lever said in a statement that “we are already in a phase to bring improvements to our AI and we know its not perfect till now, and we have to bring more improvements to make in this field. We apologize to everyone who may have seen this offensive recommendation”. Now many people are getting disappointed on Facebook for this huge mistake! No doubt, these kinds of mistakes are not acceptable from a reputed social media like Facebook. Still, we can’t do anything when it’s a mistake of Facebook AI technology which still needs huge improvement.
Artificial intelligence is not so far perfect.
When it comes to artificial intelligence, gender and racial bias are hardly a big issue! But it’s a unique case to the social media network. Facial recognition technologies are still very much perfect and usually incline to misidentify POCs and women. There are many examples before regarding the fail of AI in the internet world, and unfortunately, it’s still there. The previous year, the untrue recognition matches led to the wrongful arrests of two Black men in Detroit.
In 2015, Google photo tagged the photos of black people as “gorillas.” And reputed news portal Wired found that Google solved this issue by censoring the word “gorilla” from the searches and image tags.
Facebook hired professionals to fix the AI recognition issue.
The social media giant shared a dataset which is created by its AI community in a whole effort to combat the issue a few months ago. This dataset contains more than 40k videos featuring top-rated 3000 paid actors who shared their age and gender with the company. According to a report, social media giant Facebook even hired specialists to light their shoot and label their skin tones, so the Facebook AI systems can easily observe how people of different ethnicities look under several lighting conditions.
No doubt, this dataset was not sufficient to resolve the AI bias for Facebook, its clearly understandable that the AI community of Facebook still needs huge improvements ahead to restrict this kind of huge mistake in the future. No doubt Facebook is very serious regarding this issue, and hopefully, they will focus more on fixing these kinds of issues.
Also read: New iPhone 13 won’t be port-less