A few years ago, a former Facebook employee, Frances Haugen, blew the whistle on Facebook, making a large collection of internal Facebook papers public. These documents, containing shocking information about the platform and company, became known as the Facebook Files. The Wall Street Journal, one of the three entities that received an original copy of the Files, published both a written and podcast series addressing some of the biggest and most shocking concerns present in the documents. These included mental health for teens, human trafficking, and algorithm changes that encouraged outrage and hate speech. The latter is most crucial for this discussion of what it means to engage democratically on social media platforms.
According to the Wall Street Journal’s podcast series on The Facebook Files, Mark Zuckerberg claims that Facebook identifies and removes 89% of hate speech content before anyone ever sees it. This seems like a very impressive amount, though it is a little concerning considering the sheer amount of hate speech that ends up actually being seen by users. Unfortunately these numbers are not as clear cut as Zuckerberg makes them sound. One internal document from the Files, titled “Demoting on Integrity Signals Is Not Enough,” says that ” we only take action against approximately 2% of hate speech on the platform” (Wall Street Journal). Ok. Wait a second. 2%? What happened to 89% of hate speech being taken down before anyone sees it?
Here’s part of the problem: Facebook uses AI to moderate the platform. It makes sense when you think about just how much content a human moderator would need to get through each day, and how expensive that would be for the company. In 2019, human moderators for hate speech cost Facebook $100 million. Using AI was a cheaper, faster option. The only problem is that the AI moderator struggled to identify differences that would be obvious to a human, like fighting animals and a car crash, or a video game and a real shooting. Suttle nuance makes things even more complicated, like distinguishing between satire and real hate speech. The AI was just not able to identify these essential differences (Wall Street Journal).
Let’s get back to those percentages that seem so contradictory. Both can be true. It is most likely that, of the 2% of hate speech content Facebook removes, 89% is taken down before anyone sees or reports it (Wall Street Journal). That means that about 98% of the hate speech content on Facebook’s platform is not taken down. Now that’s a scary thought.
How does this effect democracy and political dialogue? In a previous work (Social Media Algorithms: What Pushes All Your Buttons), I discussed how social media algorithms are designed to push content that drives up engagement levels. This content tends to be the kind that evokes emotions, like cute puppies and heavily controversial political statements. Hate speech is the kind of content that gets pushed by the algorithm, the kind of content that users reshare, the kind of content that goes viral. This hateful content clogs the pipes of democratic engagement with posts that inspire anger, fear, and cruelty. Such content is not beneficial to dialogue by any means.
So what can be done? Whistleblower Frances Haugen suggested that Facebook could change the way users reshare content on the platform. Rather than clicking reshare, Facebook could require users to copy and paste the post to a new post of their own. These few seconds of extra time could get people to think just a tiny bit more about the content they are resharing. Facebook could even limit the number or reshares per post to two times for each user (Haugen, Yale ISP). Unfortunately, Facebook has continually prioritized profit over the safety of users. Haugen clarifies that “no one at Facebook is malevolent,” but their priorities are misaligned (Haugen, 60 Minutes). It will take conscious effort and choices by both Facebook and its users to promote democratic dialogue rather than hate speech on the platform.
Works Cited
Deepa Seetharaman. “Part 7: The AI Challenge.” The Facebook Files. From The Wall Street Journal, 18 October 2021, https://open.spotify.com/episode/7M7BM2h5tRuOAaqw86MAGM?si=7uOgnj4dSBSAoeIE1XOAKA.
“The Facebook Files: What Next? Panel 1: The Activists.” YouTube, uploaded by Yale ISP, 7 October 2021, https://youtu.be/YUubanGIZc0?si=qxH2YRYqeiPBmtRt.
“Facebook Whistleblower Frances Haugen: The 60 Minutes Interview.” YouTube, uploaded by 60 Minutes, 3 October 2021, https://youtu.be/_Lx5VmAdZSI?si=c-Y9laXs9GUO7RO.
