Both Reddit and Twitter failed to remove over 70 percent of the antisemitic content ADL uncovered with first-ever independent artificial intelligence tool
New York, NY, February 28, 2022 … The ADL (Anti-Defamation League) Center for Technology and Society today announced the launch of the first version of its Online Hate Index, a machine-learning system that detects hate targeting marginalized groups on online platforms. The Online Hate Index represents the first independent, cross-platform measurement of the prevalence of antisemitic content on social media and, critically, is centered on the lived experience of the community targeted by hate. This is the first tool using artificial intelligence that has been trained by experts in antisemitism and by Jewish volunteers.
In its inaugural use, the antisemitism classifier found that Twitter and Reddit failed to remove roughly three out of four of the antisemitic posts ADL had initially detected on their platforms, even after the posts had been up for nearly a month.
The deep learning that powers the Online Hate Index is trained to identify antisemitic language using connections between observed social media content and language that is labeled antisemitic by human annotators. Through repeated interaction with labeled data during the training process, the model becomes more accurate. Once fully trained, the model can analyze English-language text and predict whether that text is antisemitic far faster and for a far larger volume of content than any human moderator could.
“For the first time, we are using the combined powers of artificial intelligence and ADL’s expertise to uncover antisemitic content online at scale,” said Jonathan Greenblatt, ADL CEO. “Technology companies are not sufficiently transparent about the effectiveness of their anti-hate policies and product interventions. So we created a tool that offers a clear picture of the state of hate on social media in real time, and we will use this tool to hold those social media platforms accountable for how well they proactively take down hate and how well their content moderators respond to reports.”
In August 2021, ADL used the Online Hate Index to detect antisemitic content on Reddit and Twitter, and then returned several weeks later to see if the platforms had removed the antisemitic content on their own. At that time, ADL found that both Reddit and Twitter failed to remove at least 70 percent of the content ADL uncovered as part of this investigation. Specifically, 79 percent of antisemitic tweets of the representative sample of content still remained on Twitter, and 74 percent of antisemitic comments of the representative sample remained on Reddit. But even after ADL reported this hateful content and more than two months after ADL first found it, 56 percent of antisemitic Reddit comments and 57 percent of antisemitic tweets remained on the platforms.
“While Reddit and Twitter have far more to do, they have both made substantial recent strides in addressing antisemitism and hate online. In this light, we offer our recommendations to help them better address these broader societal problems of online—and offline—hate and antisemitism,” said Greenblatt.
Typically, hate prevalence on social media platforms is only reported by the tech companies themselves, which use opaque internal metrics that provide results that are neither verifiable nor useful in making comparisons between platforms. Reddit and Twitter are more transparent than most, but tech companies generally do not publicly report the prevalence of hate targeted against specific groups on their social media platforms, and they do not provide data on who is targeted and which identity groups they belong to -- even when many targets self-report that information. Some platforms, such as Facebook, have even gone so far as to hide what they do know about who is targeted by hate on their platform from their own auditing teams and other efforts at accountability.
To better address antisemitic content online, ADL developed recommendations for both platforms and legislators. For Twitter and Reddit, ADL recommends that the companies develop content moderation tools and practices specifically designed to protect marginalized communities, submit to regular and comprehensive third-party audits informed by outside experts on online hate, enforce policies on antisemitism and hate consistently and at scale, publicly report key metrics on hate content targeting specific groups, and provide researchers with greater access to data. For the government, as part of its REPAIR plan to decrease hate online, ADL encourages policymakers to prioritize regulation and reform focused on transparency and accountability, and to support research and innovation in curbing online hate.
Building on ADL’s century of experience building a world without hate, the Center for Technology and Society (CTS) serves as a resource to tech platforms and develops proactive solutions to fight hate both online and offline. CTS works at the intersection of technology and civil rights through education, research and advocacy.