Coronavirus puts new stresses on Facebook's content review systems
Facebook's latest Standards Enforcement Report, released Tuesday, details the scope of the company's efforts to police its platforms during the onset of the coronavirus pandemic, including massive efforts at labeling misinformation and relying more on artificial intelligence to weed out content that violates its standards.
The big picture: The pandemic is pushing the limits of Facebook's systems and policies in new ways, as it struggles to meet a commitment to take strong action against harmful or misleading public health information.
Driving the news: Facebook says it worked with more than 70 fact-checking partners to put 50 million fact-checking labels on posts based on 7,500 articles shared on its platform in April.
- The company also says it's leaning more on AI-based content moderation than ever before, in part because it had to send more people home during the coronavirus pandemic.
- AI now accounts for nearly 89% of Facebook's content removal, up from 80% the previous quarter. The company will start allowing some of its human reviewers back in the office on a gradual basis.
Details: Facebook CEO Mark Zuckerberg also said on a press call that since March 1, the company removed more than 2.5 million pieces of content related to the sale of masks, hand sanitizer, wipes and test kits that were being sold by people trying to scam users for money.
- "The challenge is that for everyone trying to exploit the situation, there are also a lot of people who are trying to help," Zuckerberg said. "Sometimes it's hard to detect fraudulent sales that we should stop versus well-intentioned sales we should help."
- Zuckerberg says that the company is using machine learning tools similar to those it uses to stop drug sales for this effort.
Between the lines: The company also said that it's creating more open-source efforts to take down bad content.
- The company says it has created a database of 10,000 "hateful memes" to help researchers learn how to spot hate speech online.
- It's launching the "Hateful Memes Challenge," a first-of-its-kind online competition hosted by DrivenData with a $100,000 total prize pool to help better detect hate memes.
- It has also created a "Deepfake Detection Challenge" to crowdsource better ways to identify deepfakes on its platform globally.
In addition, Facebook says its report now provides data from Instagram in newly broken out categories for hate speech, adult nudity and sexual activity, violent and graphic content, and bullying and harassment.
- The company also announced new Instagram controls to combat bullying and hate speech.
- Zuckerberg said that Instagram has taken down a significantly greater amount of suicide-related content this quarter than the quarter prior.
By the numbers on Facebook:
- Hate speech on Facebook: The tech giant took action on 9.6 million pieces of hate speech content on Facebook last quarter, up from 5.7 million the previous quarter.
- Fake accounts on Facebook: Facebook says its proactive rate in removing fake accounts remained above 99% through Q1 2020. It estimates that the prevalence of fake accounts at approximately 5% of its worldwide monthly active users (MAU) on Facebook.
- Hate speech on Instagram: The company says it took action on 805.9 thousand pieces of hate speech on Instagram in Q1 2020, up down from 843.6 thousand from the previous quarter.
What's next: Zuckerberg says Facebook's Standards Enforcement Report for next quarter should show more impact from the coronavirus, since the pandemic didn't really take hold across the world until March, affecting only the end of the quarter covered by the current report.