Dark Reading is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them.Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

Threat Intelligence

5/14/2020
12:18 PM
Connect Directly
Facebook
Twitter
LinkedIn
RSS
E-Mail
100%
0%

Facebook Fails to Staunch Coronavirus Misinformation

The social media giant in April affixed warning labels on 50 million pieces of content

As Mark Zuckerberg this week detailed the results of the company's latest Community Standards Enforcement Report, he also revealed that Facebook is being inundated with coronavirus misinformation and disinformation—and that the company has been struggling to stop it.

Zuckerberg, who said he was proud of the work Facebook's content moderation teams have done, also acknowledged that it's not stopping enough COVID-19 misinformation. 

"Our effectiveness has certainly been impacted by having less human review during COVID-19, and we do unfortunately expect to make more mistakes until we're able ramp everything back up," he said.

The report, released Tuesday, says that improvements in the company's machine learning systems helped Facebook remove 68% more hate posts in the first three months of 2020 than it did from October to December 2019. These same systems are responsible for detecting 90% of hate speech posts before they're posted by Facebook users, the company claims. Between the company's human moderators and automated systems, Facebook says it acted on 9.6 million pieces of hateful content in the first quarter of 2020, up from 5.7 million pieces of hateful content in the fourth quarter of 2019.

On Instagram, the detection rate increased from 57.6% in the fourth quarter of 2019 to 68.9% in the first quarter of 2020, with 175,000 pieces of content removed—35,200 more than the previous quarter.

The company also published a separate blog post detailing how it has handled COVID-19 misinformation thus far: In April, it placed warning labels on 50 million pieces of content, based on 7,500 articles by 60 independent fact-checking organizations; and since March 1, Facebook has removed more than 2.5 million pieces of content for attempting to fraudulently sell face masks, hand sanitizer, disinfecting wipes, and COVID-19 test kits.

But because Facebook does not allow its human content moderators (who just won a $52 million lawsuit over mental health problems developed from reviewing that content) to access potentially sensitive Facebook data from their home computers, the company has been relying on artificial intelligence systems more than before. 

Facebook's numbers are impossible to verify since the company does not allow independent audits of its systems. And in the age of COVID-19, that takes on added importance, say experts, since medical misinformation can encourage people to take health risks that put their lives at stake.

Look no further than the viral spread of the "Plandemic" video, which asserts a fake COVID-19 narrative, and has been reposted to YouTube nearly as fast as the company can take it down for violating its COVID-19 misinformation standards. Facebook explained in a blog post from its AI team what makes automatic detection of misinformation challenging, an assertion that Bhaskar Chakravorti, senior associate dean of international business and finance at Tufts University’s Fletcher School, agreed with in an emailed statement.

"Facebook has done more than the other social media companies in controlling misinformation by turning to fact checking organizations," he wrote, but cautioned that the company is still missing "about 40 percent of misinformation" on its platform.

One way to reduce the impact of misinformation and organized disinformation from political sources would be to label advertisements and posts from political organizations, says Pablo Breuer, co-founder and vice president of the Cognitive Security Collaborative. Misinformation stands out because it's often promoted nearly simultaneously by accounts with no connection, and because of the attention it receives: Legitimate news has a baseline signal that "peaks" because it's newsworthy, then "degrades" because people have seen it, he says. 

"What happens in a lot of misinformation is that you get multiple peaks, and you get those peaks because there are bots that are putting it out there to different audiences," Breuer says. "We've known for a long time that the propagation of misinformation is different from regular information."

Ultimately, the challenge Facebook faces is existential: Its services depend on users sharing information, and the more they share, the more ads the social network can show you, he notes.

"Anything that makes you react in a visceral, emotional way, instead of a cognitive way, is good for traffic and good for the bottom line of these companies. It's detrimental to them to limit this."

 

 
Learn from industry experts in a setting that is conducive to interaction and conversation about how to prepare for that "really bad day" in cybersecurity. Click for more information and to register
 
Related Content:
 
 
Seth is editor-in-chief and founder of The Parallax, an online cybersecurity and privacy news magazine. He has worked in online journalism since 1999, including eight years at CNET News, where he led coverage of security, privacy, and Google. Based in San Francisco, he also ... View Full Bio
 

Recommended Reading:

Comment  | 
Print  | 
More Insights
Comments
Threaded  |  Newest First  |  Oldest First
brodiegrave
50%
50%
brodiegrave,
User Rank: Apprentice
5/17/2020 | 10:18:48 AM
Re: How to censor a dumbed down society with dark, self-centered agendas?
I guess too
COVID-19: Latest Security News & Commentary
Dark Reading Staff 8/14/2020
Lock-Pickers Face an Uncertain Future Online
Seth Rosenblatt, Contributing Writer,  8/10/2020
Hacking It as a CISO: Advice for Security Leadership
Kelly Sheridan, Staff Editor, Dark Reading,  8/10/2020
Register for Dark Reading Newsletters
White Papers
Video
Cartoon
Current Issue
7 New Cybersecurity Vulnerabilities That Could Put Your Enterprise at Risk
In this Dark Reading Tech Digest, we look at the ways security researchers and ethical hackers find critical vulnerabilities and offer insights into how you can fix them before attackers can exploit them.
Flash Poll
The Changing Face of Threat Intelligence
The Changing Face of Threat Intelligence
This special report takes a look at how enterprises are using threat intelligence, as well as emerging best practices for integrating threat intel into security operations and incident response. Download it today!
Twitter Feed
Dark Reading - Bug Report
Bug Report
Enterprise Vulnerabilities
From DHS/US-CERT's National Vulnerability Database
CVE-2020-17475
PUBLISHED: 2020-08-14
Lack of authentication in the network relays used in MEGVII Koala 2.9.1-c3s allows attackers to grant physical access to anyone by sending packet data to UDP port 5000.
CVE-2020-0255
PUBLISHED: 2020-08-14
** REJECT ** DO NOT USE THIS CANDIDATE NUMBER. ConsultIDs: CVE-2020-10751. Reason: This candidate is a duplicate of CVE-2020-10751. Notes: All CVE users should reference CVE-2020-10751 instead of this candidate. All references and descriptions in this candidate have been removed to prevent accidenta...
CVE-2020-14353
PUBLISHED: 2020-08-14
** REJECT ** DO NOT USE THIS CANDIDATE NUMBER. ConsultIDs: CVE-2017-18270. Reason: This candidate is a duplicate of CVE-2017-18270. Notes: All CVE users should reference CVE-2017-18270 instead of this candidate. All references and descriptions in this candidate have been removed to prevent accidenta...
CVE-2020-17464
PUBLISHED: 2020-08-14
** REJECT ** DO NOT USE THIS CANDIDATE NUMBER. ConsultIDs: none. Reason: This candidate was withdrawn by its CNA. Further investigation showed that it was not a security issue. Notes: none.
CVE-2020-17473
PUBLISHED: 2020-08-14
Lack of mutual authentication in ZKTeco FaceDepot 7B 1.0.213 and ZKBiosecurity Server 1.0.0_20190723 allows an attacker to obtain a long-lasting token by impersonating the server.