Some support can scale. Civic responsibility cannot.
People keep asking whether AI can help rescue democracy. While that is a fair question, it is also too broad for where we stand right now. A better question is whether we are trying to use AI to support people, or whether we are slowly using it to replace the human parts of civic life that actually matter.
That distinction matters because moderation and counterspeech are not the same thing.
Moderation is about rules, removal, classification, and enforcement. Counterspeech is different. It is people choosing to respond to hate, lies, and dehumanization in public. It is not only about what gets said, but about someone saying something, standing by it, and making a norm visible to others. That is one reason I found Can AI Rescue Democracy? Nope, It’s Not Funny Enough so useful.
Anyway, the point is, without people willing to participate (and just letting AI take over), civic life does not work, and that is what today’s post is about.
Moderation and counterspeech are not the same thing.
I do not believe content moderation or counterspeech should be automated by AI outright. For example, AI could support people in chat moderation rather than replace them entirely. However, automating any part of what people do in community chats or forums introduces additional considerations. Tone, timing, context, relationships, community norms, and escalation all matter.
That is generally true in community spaces. It is even more true when the issue is hate speech, because counterspeech stands apart, even if it shares some characteristics with moderation.
Research points in that direction, too. One paper found that contextualized counterspeech generated by large language models can be ineffective and may even backfire. You can read more about that here.
Overall, that is not simply a classification problem. It is a human act and part of why ethics is most useful when it helps people act. It is also one reason that what gets tracked cannot and should not replace the human judgment that democratic life still depends on.
P.S. for those of you who have never automated anything: it is not magic. You have to tell the computer, tool, or AI exactly what to do, step by step. It is highly tedious. Technology only works with the instructions it gets. When people skip details or context, things break, and then they have to redo the work because they took shortcuts without really understanding what they built. We are starting to see this more and more, with “innovative” companies hiring back for positions they tried to automate with AI (Here’s a recent article: Companies Fired Workers For AI. Now They Want Them Back).
Let’s not take that route with counterspeech.
People seeing people speak up still matters.
One of the strongest ideas in Susan Benesch and Cathy Buerger’s essay is that online debate should not be outsourced to AI. They argue that counterspeech matters in part because people see others doing it. That can encourage bystanders to join in, make hateful norms look less normal, and show targeted people that someone is willing to stand beside them. Over time, that is also how people could perhaps interrupt what gets normalized before silence starts looking like consent. Anyway, we know that AI can generate text, but it cannot create that same kind of solidarity.
That point stuck with me because it gets at something far deeper than efficiency. The issue is not whether a machine can produce a passable reply, but whether democracy still means what it used to.
That is part of what has helped keep democratic participation alive for generations.
It was not the Industrial Revolution. Not being silent. Not capitulating because you think someone else can take care of it when you are the one helping steer the ship.
*takes a deep breath*
That is also why Louis Brandeis’s line still lands: “public discussion is a political duty.” Whitney v. California may be old, but the point still holds.
AI may help with support, but not substitution.
That does not mean AI has no role at all.
In Counterspeakers’ Perspectives: Unveiling Barriers and AI Needs in the Fight against Online Hate, Jimin Mun, Cathy Buerger, Jenny Liang, Joshua Garland, and Maarten Sap found that people face real barriers to counterspeech. Some have no time because it is usually not paid. Some do not know what to say. Others worry their response will not matter.
On the other hand, some dread retaliation, harassment, or mental exhaustion.
Ultimately, it seems, people need help with research, safety, translation, drafting, and timing. Perhaps a solution that addresses all of those (at once) could help lower the barrier to speaking up, especially in a society that often protects corporate interests more aggressively than its own citizens. In that sense, AI could be useful. But the same research also raised strong concerns about authenticity, agency, and whether AI involvement would make counterspeech feel less sincere or less human. That matters because what gets trusted starts to weaken when public participation feels synthetic.
It’s about authenticity and agency, not AI.
I think that part gets missed when people talk about AI in public life. They assume the problem is whether the output sounds good enough, relevant, impactful, or engaging. Whatever helps them to slurp up more vanity metrics. However, the deeper issue is who is actually choosing to respond.
For example, if a person uses a tool to gather facts, translate something, or think through a response, that is still their judgment and choice to help. That is part of who gets a say before a system starts speaking in their place. However, if a system starts speaking for them, that is something else.
If people stop speaking for themselves, the norm shifts.
This is the part that feels biggest to me.
The danger is not only that AI might get something wrong. The danger is that people may begin to treat democratic participation as one more annoying task to automate. Reply to the troll. Draft the message. Manage the conflict. Push the button. Move on.
But democracy cannot survive on outsourced gestures, because if fewer people speak up, fewer people see that speaking up is still possible.
That is not some abstract future problem. It is a present-tense question about what we are normalizing now. It is quite literally creeping authoritarianism in real time, and what it looks like in the U.S.
At the end of the day
I am not against using AI to reduce burdens. I am opposed to pretending that reducing the burden is the same as preserving democracy.
Some support can scale. Fine. Some tools may help people research more quickly, draft more clearly, or feel better prepared. Fine.
But democracy still needs people willing to show up, speak up, and make their values visible in public, because freedom is not passive. That is part of the infrastructure people need to act, not just a feature that can be outsourced to a tool.
That is also why I do not think AI can speak for us.
It can support people, assist with some tasks, and help lower some barriers. But if we want healthier norms, stronger communities, and a public culture that is less willing to tolerate hate, then people still have to do the human part themselves.