It’s not a simple task to remove thousands of hate-filled postings each day on the world’s largest social network.
Facebook said Tuesday that over the past two months, it has removed roughly 66,000 posts that were identified as hate speech. The social network disclosed the statistic as part of a series of public postings intended to show how it deals with controversial topics like fake news.
Facebook (FB) currently defines hates speech as “anything that directly attacks people based on what are known as their ‘protected characteristics’— race, ethnicity, national origin, religious affiliation, sexual orientation, sex, gender, gender identity, or serious disability or disease,” wrote Richard Allen, Facebook’s vice president of public policy for Europe, the Middle East, and Africa.
Get Data Sheet, Fortune’s technology newsletter.
However, it’s often that there “isn’t a clear consensus” on certain posts in which the “the words themselves are ambiguous, the intent behind them is unknown, or the context around them is unclear,” Allen wrote.
In these cases, Facebook internally reviews the posts in question and decides whether to delete them or not, Allen explained. These are often difficult decisions that can end up unintentionally upsetting multiple groups of people. For example, when African-American activist Shaun King reposted hate mail he received on Facebook to denounce it, Facebook deleted his posting, then chose to restore it after taking in account the context of his post, Allen explained.
“Our mistakes have caused a great deal of concern in a number of communities, including among groups who feel we act — or fail to act — out of bias,” Allen wrote.
Although Facebook is exploring artificial intelligence technologies to automatically filter hate speech, Allen explained that Facebook still heavily relies on its users “to identify and report potential hate speech.”
“But while we’re continuing to invest in these promising advances, we’re a long way from being able to rely on machine learning and AI to handle the complexity involved in assessing hate speech,” Allen wrote.
