Outsourcing employees on Facebook are checking private postings for private visits for 'AI training'
by
Like many high-tech companies, Facebook has developed an AI that automatically categorizes content on its platform. In order to develop such content classification AI, it became clear that many outsourced employees around the world are seeing 'people's personal postings on Facebook.' Content that people see is said to include post content that has been published only to a few friends.
Facebook 'labels' posts by hand, posing privacy questions-Reuters
https://www.reuters.com/article/us-facebook-ai/facebook-labels-by-hand-posing-privacy-questions-idUSKCN1SC01T
Facebook's contract workers are looking at your private posts to train AI-The Verge
https://www.theverge.com/2019/5/6/18530887/facebook-instagram-ai-data-labeling-annotation-private-posts-outsourced
In general, training AIs that classify multiple content based on content requires a lot of learning data. The training data is human-labeled 'labeling' of sample content, and Facebook has outsourced the labeling process to a company called Wipro in Hyderabad , India.
In Wipro, the 260 workers hired to label the content, whether the photos in the content are self-portrait or what was taken of the food, what kind of scene was taken, how? We classify whether it was submitted by what purpose (whether it's a joke, trying to influence someone, or trying to gather friends). Wipro employees label photos, videos, stories, etc. posted on Facebook and Instagram, each content is judged by two workers, and the second of two judges is the third person The final decision is made by the
Facebook is responding to Reuters inquiries, Wipro workers are also labeling private content that users share with a few friends, data includes personal information including the user's name I admit that there are cases. Besides Wipro, there are nearly 200 labeling projects around the world, and there are thousands of people employed for labeling related to Facebook's AI development.
by
'AI is the heart of what people need,' said Nipun Mathur, director of AI product management at Facebook, highlighting that AI is what Facebook needs. According to Facebook, AI developed using data labeled by workers is used to display advertisements, post descriptions for visually impaired people, and to identify politically biased content and adult content. about.
Outsourcing of labeling work to low-paying countries is commonplace in other high-tech companies, in order to develop AIs and object identification AIs for use in self-driving cars. However, if the data used for labeling is private or if the outsourcing company's privacy awareness and security are low, this labeling operation may lead to the leakage of the user's privacy.
Facebook has told Reuters that the legal and privacy team has approved the data labeling process. He also said that he recently introduced a new audit system to ensure that privacy is properly protected.
Related Posts:
in Web Service, Posted by log1h_ik