Should We Clean Up Bad Speech With Artificial Intelligence Before It Happens?

Photo by cottonbro on

Internet trolls are pervasive. Their comments can be found on websites, Apps (like Instagram), and online groups. “Once a message enters cyberspace, millions of people worldwide can gain access to it. Even if the message is posted in a discussion forum frequented by only a handful of people, any one of them can republish the message … And if the message is sufficiently provocative, it may be republished again and again. The extraordinary capacity of the Internet to replicate almost endlessly any defamatory message lends credence to the notion that ‘the truth rarely catches up with a lie’. The problem for libel law, then, is how to protect reputation without squelching, the potential of the Internet as a medium of public discourse…” – “Silencing John Doe: Defamation and Discourse in Cyberspace”, (2000) 49 Duke L.J. 855 at pp. 862-865.

The publication Wired has reported that dozens of newsrooms are now using commenting platforms like Coral and OpenWeb to address libel and to monitor online comments. These platforms use a combination of humans and algorithmic tools to flag and then remove harmful statements.

Another approach that has recently gained traction is using machine learning to prevent comments. The model identifies negative language before the comment is posted. The technology then encourages users to rethink the toxic comment before they hit publish. Wired reports that “Rather than automatically rejecting a comment that violated community standards, the algorithm would first prompt commenters with a warning message: ‘Let’s keep the conversation civil. Please remove any inappropriate language from your comment’.” By gently nudging the writer, many writers revise their comments.

Similarly, Instagram has taken to using a machine learning model to detect bullying messages on their platform. Wired notes that “Before someone posts a mean comment, the platform can prompt them to write it more nicely; it can also proactively hide these types of toxic comments from users who have turned on its offensive comment filter.”

This new technology inspires questions. Does using AI in this manner interfere with free speech? Should it be available to anyone? Would it be reasonable for governments to use it? Could this technology reduce litigation around defamation?

In Grant v Torstar Corp., 2009 SCC 61, the Supreme Court of Canada held that both goals of (i) protecting freedom of expression and (ii) protecting people’s reputations are worthy of legal protection:

1. Freedom of expression is guaranteed by s. 2(b) of the Canadian Charter of Rights and Freedoms. It is essential to the functioning of our democracy, to seeking the truth in diverse fields of inquiry, and to our capacity for self-expression and individual realization.

2. But freedom of expression is not absolute. One limitation on free expression is the law of defamation, which protects a person’s reputation from unjustified assault. The law of defamation does not forbid people from expressing themselves. It merely provides that if a person defames another, that person may be required to pay damages to the other for the harm caused to the other’s reputation. However, if the defences available to a publisher are too narrowly defined, the result may be “libel chill”, undermining freedom of expression and of the press.

3. Two conflicting values are at stake – on the one hand freedom of expression and on the other the protection of reputation. While freedom of expression is a fundamental freedom protected by s. 2(b) of the Charter, courts have long recognized that protection of reputation is also worthy of legal recognition…

Given the difficulty in balancing protecting people’s reputation and their freedom of expression, technology should be used carefully. I see a particular benefit for platforms that rate professionals in deploying this technology. Many lawsuits about defamation have arisen from comments on platforms like Rate MDs and Yelp. See the following cases as examples:

By using technologies that balance people’s right to freedom of expression with the protection of reputation, there is a chance to reduce defamation lawsuits. Online forums that rate professionals should use these programs. A slanderous comment or even a comment poorly worded can disproportionately or unjustly hurt a professional’s reputation. Unduly harmful comments should be flagged before they are posted.

***If you are curious to read more about this area of law, see the Supreme Court of Canada’s decision Bent v. Platnick, 2020 SCC 23.

(Views are my own and do not represent the views of any organization. This article was originally posted on Please see Heather Douglas Law at for more information.)

%d bloggers like this: