Facebook claims that it has removed more than 90% of hate speech from its platform in public. However, the company maintains that only 3 to 5% of the hate speech is removed. Facebook wants us believe that nearly all hate speech has been removed from its platform, but in reality it retains almost all of the content.
This obscene hypocrisy was revealed amid the numerous complaints, based on thousands of pages of leaked internal documents, which Facebook employee-turned-whistleblower Frances Haugen and her legal team filed to the SEC earlier this month. The leaks have drawn attention to Instagram's effect on teens health and News Feed's role in misinformation amplifying. However, it is alarming that Facebook has not taken steps to curb hate speech and used deceitful tricks to conceal this failure. This exposes how heavily Facebook relies upon AI for content moderation and how ineffective it is. It also highlights the need to force Facebook to be transparent.
In October 2020 testimony before the US Senate, Mark Zuckerberg referred to company transparency reports. He stated that these reports show that the company is proactively identifying hate speech. Zuckerberg also responded to questions on hate speech in testimony to the House by citing a transparency document: Facebook has almost every quarter's transparency report stating that hate speech moderation percentages were in the 80s and 1990s. A March 2021 document that was leaked shows that we may only take down 3-5% of hate speech on Facebook.
Facebook was it really caught in an outrageous lie? Both yes and no. Both numbers are technically correct, but they measure different things. The only measure that matters is the one Facebook has been concealing. It is irrelevant what measure Facebook has been publicly reporting. It's almost like if a police officer asked you how fast you were driving and you replied by saying that you didn't know and then bragging about your car's gas mileage.
Two ways can hate speech be flagged to be reviewed and possibly removed. It can be reported manually by users, or AI algorithms can detect it automatically. It is more efficient than manual detection, and it can also be detected proactively before users flag hate speech.
It's almost as if you answered every question about how fast you drive by ignoring them and instead boasting about how much gas mileage your car gets.
Facebook's publicized 94 percent figure is the proactive rate. This is the ratio of hate speech items that Facebook's AI detects proactively to the total number of hate speeches taken down. Facebook wants you to believe that this number indicates how much hate speech is removed before it can cause harm. However, it does not measure how important algorithms are in hate-speech detection.
Society is concerned about the amount of hate speech that remains on the platform. This can be captured by dividing the total hate speech instances by the number of hate-speech takesdowns. This rate is a measure of how much hate speech Facebook has taken down, and it's the number that Facebook tried not to disclose.
Haugen helped us to determine the takedown rate and it is quite dismal. Internal documents show that more than 95 percent hate speech posted on Facebook remains on Facebook. Zuckerberg claimed to Congress that Facebook had removed 12 million hate speech Groups posts. However, the leaked estimate now shows that there were probably 250 million hate speech pieces left. This is a staggering figure that shows how little has been done since the early days, when unregulated internet forums existed.
The complaint Haugens legal staff filed to the SEC made the matter more complicated by prominently stating in bold, Facebooks Records Consistent That Facebooks Statements Were False. Although Facebook technically didn't lie or misstate truths as the complaint alleges, it has repeatedly and unquestionably misled the public about how toxic its platform is and how bad it is at reining in hate speech.
