Hate Speech – How far should social media censorship go, and can there be unintended consequences?
By Anonymous, 10/14/2018
Today, there is a lot of harmful or false information being spread around the internet, among the most egregious of which is hate speech. It may seem a difficult task to stop hate speech from spreading over the vast expanse of the internet, however, some countries are taking matters into their own hands. In 2017, Germany banned any hate speech, as well as “fake news” that was defamatory, on social media sites. Some social media sites have already begun censorship of hate speech; for example, both Twitter and Facebook disallow it.
This raises the question, should there be social media censorship of hate speech, and could there be disparate impact on certain groups, particularly on protected groups? Some may argue that there shouldn’t be censorship at all; in the US, there is very little regulation of hate speech, since courts have repeatedly ruled that hate speech regulation violates the First Amendment. However, others may take the view that we should do all we can to stop language which could potentially incite violence against others, particularly minorities. Furthermore, since social media companies are private platforms, ultimately they have control over the content allowed on their site, and it makes sense for them to enhance their reputation by removing bad actors. Thus some social media companies have decided that while it not appease everyone, social media censoring is positive for their platform and will be enforced.
Could social media censoring of hate speech lead to unintended consequences, that unintentionally harm some groups or individuals over others? Firstly, since hate speech is not always well defined, could the list of phrases included as hate speech disproportionally affect certain groups of people? Certain phrases may not be offensive in some cultures or countries, but may be in others. How should social media sites determine what constitute hate speech, and is there a risk that certain groups have their speech censored more than other groups? In addition, could the way hate speech is monitored could be subject to bias from the reviewer or algorithmic bias?
Some social media sites do seem to recognize the complexity of censorship of hate speech. Facebook has a detailed blog which discusses its approach to hate speech; it details that the decision to determine whether a comment is hate speech or not includes a consideration of both context and intent. The blog even provides complex examples of words that may appear offensive but are not because the phrase was made in sarcasm, or the use of the word was to reclaim the word in a non-offensive manner. This is an important recognition in the censorship process; among many issues in ethics, there is often no absolute right or wrong answer, with context being the major determinant, and this is no exception.
Facebook also notes that they “are a long way from being able to rely on machine learning and AI to handle the complexity involved in assessing hate speech.” Algorithmic bias is thus not an issue as of yet, but more importantly, it is good that there is not a rush to use algorithms in this situation; something as context based as hate speech would be extremely difficult to flag correctly, and doing so is certain to lead to many false positives. It does however mean the main identification of hate speech comes from user reports and employees who review the content. This could lead to two additional forms of bias. Firstly, there could be bias in the types of posts that are reported or the people whose posts get reported. There could be a particular issue if there are certain posts that are not being reported because the people they target are too afraid to report it, or if they do not see the post. This is a difficult problem to address, although anonymous reports should somewhat resolve the issue of the fear of reporting. The second type of bias is that the reviewers themselves are biased in a certain way. While it is difficult to remove all types of bias, it is important to increase the understanding of potential sources of bias, and then address the bias. Facebook has pledged that their teams will continue learning about local context and the changing language, in an effort to combat this. It is a difficult battle, and we must hope that the social media companies are able to get it right; in the meantime, we should continue to monitor that hate speech censorship censors just hate speech, and no more.
BBC News, “Germany starts enforcing hate speech law”, retrieved October 13, 2018 from
Twitter, “Hateful Conduct Policy”, retrieved October 13, 2018 from https://help.twitter.com/en/rules-and-policies/hateful-conduct-policy
Facebook, Community Standards – Hate Speech, retrieved October 13 from https://www.facebook.com/communitystandards/hate_speech
Facebook Newsroom, “Hard Questions: Who Should Decide What Is Hate Speech in an Online Global Community?”, retrieved October 13 from https://newsroom.fb.com/news/2017/06/hard-questions-hate-speech/