Facebook tags & # 39; handmade, asking privacy issues

HYDERABAD, India / SAN FRANCISCO (Reuters) – In the past year, a team of 260 contractors in Hyderabad, India, has seen millions of photos, status updates and other content posted on Facebook Inc. from 2014.

Workers classify items according to five "dimensions," as Facebook calls them.

These include the subject of publication – whether it is a food, for example, a selfie or an animal? What is the occasion – a daily activity or a major event of life? And what is the intention of the author – plan an event, inspire, make a joke?

The work aims to understand how the types of articles published by users on its services evolve, said Facebook. This can help the company to develop new features, potentially increasing the use and advertising revenue.

The details of the effort were provided by several employees of the outsourcing company Wipro Ltd for several months. The workers spoke on condition of anonymity for fear of reprisals on the part of the Indian company. Facebook then confirmed many details of the project. Wipro declined to comment and referred all the questions to Facebook.

Wipro's work is one of more than 200 content tagging projects proposed by Facebook at any given time and employs thousands of people around the world, company officials told Reuters. Many projects aim to "train" the software that determines what appears in the news feeds of users and that feeds the artificial intelligence behind many other features.

Labeling efforts have not been reported before.

"It's an essential part of what you need," said Nipun Mathur, director of product management for AI at Facebook. "I do not see the need to go."

The content labeling program could pose new privacy concerns for Facebook, according to legal experts consulted by Reuters. The company faces global regulatory investigations of an unrelated set of alleged privacy breaches involving user data sharing with business partners.

Wipro workers reported having a window to life when they were viewing a holiday photo or a post-commemorative image of a deceased family member. Facebook has acknowledged that some messages, including screenshots and those with comments, may include user names.

The company said its legal and privacy teams needed to endorse all labeling efforts, adding that it had recently put in place an audit system "aimed at ensuring that privacy expectations are met and that the settings in place work as intended. "

A former Facebook Privacy Officer, speaking on condition of anonymity, however, expressed his discomfort with the fact that users' messages are examined without their explicit permission. The European Union's General Data Protection Regulation (GDPR), established a year ago, contains strict rules on how companies collect and use personal data and in many cases require specific consent. .

"One of the key elements of the RGPD is the limitation of purpose," said John Kennedy, partner of the law firm Wiggin and Dana, who worked on outsourcing, privacy and privacy. # 39; IA.

If the goal is to look for positions to improve the accuracy of services, this should be stated explicitly, said Kennedy. The use of an outside supplier for the job might also require consent, he said.

It is still unclear how GDPR will be interpreted and whether regulators and consumers will view Facebook's internal labeling practices as problematic. The main European data protection officer declined to comment on any concerns.

A Facebook spokesperson said: "Our data policy clearly states that we use the information that people provide to Facebook to improve their experience and that we can work with service providers to help them in this process.

US Sen. Mark Warner, a Democrat and leading social media critic, told Reuters in a statement that major platforms "are capturing more and more user data, for wider and wider uses, without any corresponding compensation for the user. "

Warner said he was developing legislation that would require Facebook to "disclose the value of user data and tell users how their money is monetized."

THE PROJECT

Managed content labeling, also called "data annotation," is a growing industry, with companies seeking to leverage data for AI training and other purposes .

Autonomous car manufacturers such as Waymo of Alphabet Inc. have asked their labellers to identify traffic lights and pedestrians in videos to reinforce their artificial intelligence. Voice assistant developers, including Amazon.com Inc., ask users to annotate the sound of clients to improve the ability of the AI ​​to crack speech.

Facebook launched the Wipro project last April. The Indian firm received a $ 4 million contract and trained a team of about 260 labellers, according to workers. Last year, the job consisted of analyzing the positions of the previous five years.

After that, the team in December was reduced to about 30 people and moved to the job labeling each month compared to the previous month. The work should last at least until the end of 2019, they said.

Facebook has confirmed the staff changes, but declined to comment on the financial details.

The company said its analysis was underway, so it could not provide any results from labeling or product decisions. He did not explain to taggers the purpose or results of the project and the workers indicated that, based on their limited opinion, they concluded that selfies are becoming more popular.

The Wipro and Facebook taggers reported that these publications were a random sample of text-based status updates, shared links, events, mailings, videos, and photos, including snapshots. screens posted by users on different Facebook messaging applications. The messages come from Facebook and Instagram users around the world, in languages ​​such as English, Hindi and Arabic.

Each article goes to two labellers to check for accuracy, and a third one they disagree, said Facebook. Workers reported seeing an average of 700 articles per day. Facebook said the target average is lower.

Confirmed Facebook labels in Timisoara in Romania and Manila in the Philippines are involved in the same project.

Among Facebook's other tagging projects, a collaborator in Hyderabad of outsourcing provider Cognizant Technology Solutions Corp said he and at least 500 colleagues were looking for sensitive topics or profane language in Facebook videos.

The goal is to form an automated Facebook tool allowing advertisers to avoid sponsoring videos, for example, for adults or politicians, said Facebook. Cognizant did not respond to a request for comment.

Another labeling application has been the social marketplace buying feature, which automates category recommendations for new ads by starting to let taggers and product experts categorize some existing ads, said Mathur. , from Facebook.

PRIVATE POSTS

Facebook users are not offered the opportunity to refuse to have their data tagged.

At Wipro, the messages reviewed include not only public messages, but also those that are privately shared with a limited number of user friends. This ensures that the sample reflects the range of activities on Facebook and Instagram, said Karen Courington, director of product support operations at Facebook.

Facebook's data policy does not explicitly mention manual analysis.

"We provide information and content to suppliers and service providers who support our business, for example by providing technical infrastructure services, analyzing the use of our products, providing customer service, facilitating payments or by conducting surveys, "states the policy.

The European GDPR also requires companies to delete user data on demand. Facebook reported having the technology to regularly synchronize tagged posts with deletion requests and changes to content privacy settings.

Facebook and other companies are testing techniques to reduce the need for outsourced labeling, in part to analyze more data more quickly and cheaply. For example, artificial intelligence training data for wire feeds and photo descriptions for the blind come from hashtags on Instagram, said Mathur of Facebook.

"We try to minimize the number of messages we send," he said.