Meta Platforms is restructuring its internal research department, drawing on employees from individual divisions like Facebook, WhatsApp and Instagram to staff a centralized unit that will provide services to the entire company. The research will span everything from societal topics of politics, equity, health and climate to credibility topics like misinformation and account safety. The new division will be managed by Meta head of research Pratiti Raychoudhury. Additionally, Meta is deploying the new Few-Shot Learner artificial intelligence system to help moderate content, identify trends, monitor data and implement rules.

Raychoudhury said in a statement that centralizing the research team is optimal for collaboration and team learning, and will allow the company to consistently apply its findings across all platforms. The reorganization “will affect about 50 researchers” and no jobs are expected to be lost as a result, according to Bloomberg.

Bloomberg positioned the move as at least partially due to allegations the Meta companies minimized negative research results when it was profitable to do so, noting the company is also “refocusing its business on new innovations in virtual and augmented reality.”

The changes follow a damaging series of internal research document leaks by whistleblower Frances Haugen, who testified before legislators in the U.S. and Europe throughout the fall. The leaked documents revealed the company’s challenges gathering research from around the globe and appropriately applying its learnings in far-flung places.

“Internal discussions revealed worries that moderation algorithms for the languages spoken in Pakistan and Ethiopia were insufficient, and that the company lacked adequate training data to tune systems to different dialects of Arabic,” Wired reports, noting Meta is deploying the faster-learning Few-Shot Learner artificial intelligence moderation system to monitor content.

The new Few-Shot AI reportedly understands more than 100 languages and can parse images as well as text. “Few-Shot Learner makes it possible to automate enforcement of a new moderation rule in about six weeks, down from around six months,” Wired writes.

Meta says “Few-Shot AI contributed to a decline it recorded in the worldwide prevalence of hate speech from mid-2020 through October this year,” although specifics as to the performance of the new system, deployed earlier this year, were not released.

Few-Shot Learner gets its name because after being “pretrained on a firehose of billions of Facebook posts” it is then able to apply its machine learning across widely varying categories with relatively little information, Facebook AI product manager Cornelia Carapcea explained to Wired. Whereas conventional moderation AI “might need hundreds of thousands or millions of example posts before they can be deployed,” Few-Shot Learner can “work using just dozens,” thus, the “few shots.”

Related:
Senate Commerce Chair Urges Probe of Meta’s Facebook Over Whether It Misled Businesses, The Wall Street Journal, 12/9/21
Senators Want Social-Media Apps to Share Research, The Wall Street Journal, 12/9/21
As Facebook Plans the Metaverse, It Struggles to Combat Harassment in VR, CNET, 12/9/21