Drawing the Line

Online platforms commonly use AI and machine learning to moderate speech at enormous scale. How should society protect people from harmful speech without giving private companies the power to decide what speech is acceptable?

It’s no surprise that we live in a world that can’t keep anything consistent except one of two things. The advance of AI and social media. Two things that a large percentage of the world lives by because they tell them everything they feel they need to know. But what happens when we as humans can’t even distinguish what is real and what isn’t. Because as the world pours more and more money into the advancement of AI, that also means online platforms that we get our information from can be controlled by AI. For example, in the most recent politics, political groups have been using AI-generated videos to spread false information about opposing candidates and the same thing goes for state governors as well. By using AI, we, as humans, have an entity that can say something “perfect” for society to hear. Yet, what happens when the harsh realities of the world are being covered up by the little good things we “think” are happening. Those harsh realities pile up on top of one another and eventually come to the surface, and suddenly we don’t know how to handle it all at once.

This is where the question of AI-driven speech moderation becomes one of the most important debates, especially in our generation. Online platforms like Facebook, YouTube, TikTok, and X, formerly known as Twitter, already rely heavily on AI and machine learning to moderate speech at an enormous scale. Millions of posts, videos, and comments are flagged, filtered, or removed every single day by algorithms that most users never see and never fully understand. The speed and scale at which content spreads online makes human-only moderation practically impossible. There are simply too many posts, in too many languages, happening too fast for any team of human reviewers to catch everything in time. So, as a result of this, many companies turned to AI to fill that gap. But in doing so, they hand over an insane amount of power, the power to decide what ideas get amplified and what ideas get silenced, to a system built and controlled by private corporations.

That raises a fundamental question: who gave them that authority? These are not elected governments. They are not public institutions bound by constitutional protections of free speech. They are businesses, and businesses have interests. Their interest is in keeping users engaged, keeping advertisers happy, and avoiding public scandal. When those interests align with protecting people from genuine harm, moderation works well. But when those interests conflict, when removing harmful content means losing traffic, or when flagging misinformation means upsetting powerful political allies, there is no guarantee that the AI will be guided by the public good rather than the bottom line.

The problem runs even deeper than corporate bias. AI systems are only as fair as the data they are trained on, and that data reflects the biases, blind spots, and cultural assumptions of the people who built them. An AI trained largely on English-language content from Western sources will not moderate Arabic, Swahili, or Hindi content with the same accuracy or sensitivity. An AI built to detect hate speech may disproportionately flag content from marginalized communities who use reclaimed language in ways the algorithm doesn’t understand. Studies have repeatedly shown that AI content moderation systems have higher error rates when dealing with dialects, slang, satire, and cultural context, exactly the kinds of nuance that matter most when deciding whether speech causes real harm or simply makes someone uncomfortable.

At the same time, we cannot pretend that doing nothing is a neutral option. The spread of AI-generated digital forgery, coordinated disinformation campaigns, and incitement to real-world violence is not an abstract concern. They are happening right now, and the consequences are measurable, in radicalized individuals, in election interference, in violence against communities targeted by coordinated hate campaigns. Leaving platforms completely unmoderated does not protect free speech in any meaningful sense. It simply hands the microphone to whoever is willing to be the most extreme, the most dishonest, or the most manipulative.

So what is the answer? The honest truth is that there is no perfect solution, but there are better and worse frameworks. Any system that gives a single private company unchecked power to define acceptable speech is dangerous, regardless of how good its intentions are. What society needs instead is transparency, clear public disclosure of how moderation systems work and what rules they follow. It needs accountability, independent oversight that can audit AI moderation for bias and error. And it needs democratic input, public conversations, not just corporate policy documents, that determine where the line between harmful and protected speech should actually be drawn.

AI is a tool. Like every tool in human history, what it builds or destroys depends entirely on who is holding it and what they have been asked to do. The question is not really whether AI should moderate speech. The question is whether the people directing that AI are answerable to the public, or only to themselves.