Who Actually Said That? AI Is Breaking Civic Trust

A recent essay by MIT professor Deb Roy in the Bulletin of the Atomic Scientists raised a deceptively simple question: when someone — or something — speaks in our public square, who is actually accountable for those words? It’s a question that feels urgent now that AI can generate thousands of realistic public comments that no longer look obviously fake.

As someone who writes for the internet for a living, I find this idea both fascinating and unsettling. We’ve spent the last decade worrying about whether we can tell if content is made by AI. Roy’s point pushes us further: even if we know it’s AI, the deeper problem is that no human is willing to stand behind it. And without that, civic conversation starts to collapse. This is my take on why that matters, and why fixing it is much harder than it sounds.

The Fake Comment Problem Isn't New, But AI Supercharged It

Government agencies asking for public feedback is a cornerstone of democracy in many countries. Before a new regulation is passed, there is usually a comment period. Anyone can write in. In theory, it’s a direct line between the people and the policymakers.

In practice, that line has always been noisy. Lobbying groups, advocacy organizations, and even pranksters have long used form letters and copy-paste campaigns to flood inboxes. Staffers know this. They have developed a kind of institutional immune system for spotting mass-produced feedback — identical phrasing, suspicious timing, or thousands of emails from the same template. It was manipulative, but it was legible.

Generative AI breaks that immune system. Now, instead of 10,000 identical letters, you can generate 10,000 unique, personal-sounding stories. A single operator can create comments from a "concerned mother in Ohio," a "small business owner in Arizona," and a "retired teacher in Oregon," each with different wording, different anecdotes, and different emotional tones. None of them are obviously fake, but none of them correspond to a real person who will answer for that opinion if questioned.

This is not just about spam. It’s about a shift in scale and believability. We moved from a world where faking public consensus was labor-intensive and easy to spot, to one where it is cheap, fast, and plausible. The cost of manufacturing the appearance of public support has dropped to near zero.

Accountability Is More Than Knowing Who Typed It

When we talk about accountability, most people immediately think about identity: we need to know the real name behind the account. But Roy’s framing is more nuanced and, in my opinion, more useful. Accountability isn’t just about a name tag. It’s about answerability.

The Difference Between Speaking and Standing Behind Speech

Think about how accountability works offline. If you stand up at a town hall meeting and say you oppose a new housing project because it will increase traffic, people can challenge you. Your neighbor can ask, "Which intersection are you worried about?" A city planner can show you a traffic study. You have to respond, clarify, or concede. Your words are tethered to you.

That tether is what makes conversation constructive. You are not just broadcasting an opinion into the void; you are entering a relationship where you can be questioned, where your reputation is on the line, and where you might even change your mind.

AI-generated civic comments sever that tether completely. There is no person to ask a follow-up question to. There is no one who feels the social cost of being wrong or misleading. The text exists, it may even influence a policymaker who thinks it represents genuine public sentiment, but it floats free of any human responsibility. It’s speech without a speaker.

This distinction is crucial because it explains why simple AI-detection tools won’t save us. Even if we could perfectly label every comment as "AI-generated," we still wouldn’t have solved the accountability gap. We would just know that a ghost wrote it.

Why Civic Spaces Are Uniquely Vulnerable

You might ask: so what if some comments are fake? The internet is full of bots and fake reviews. We learn to live with it.

But civic conversation is different from product reviews or social media drama. Its entire legitimacy depends on the idea that it reflects real people with real stakes. If policymakers can’t trust that the feedback they receive represents actual constituents, the whole exercise becomes theater. Worse, it becomes vulnerable to manipulation by whoever has the best AI tools.

There is also a chilling effect to consider. Imagine you are a real citizen who took the time to write a thoughtful, personal comment about how a policy affects your family. If you suspect the docket is already flooded with 50,000 synthetic stories, why would you bother? The sheer volume of unaccountable speech can drown out and discourage accountable speech. The people with the most to lose — those without organized lobbying power — are the first to disengage.

In that sense, the danger isn’t just that AI will trick lawmakers into making bad decisions. It’s that it will make ordinary people feel like their participation is pointless. And a democracy where people stop participating because they feel outnumbered by machines is in deep trouble.

The Trade-Offs We Don't Want to Admit

If the diagnosis is a lack of answerability, the obvious prescription is to demand more verification. But here is where my opinion gets complicated, because every solution comes with a real cost.

Verification vs. Anonymity

One idea is to require strong identity verification for anyone submitting a public comment — a government ID, a verified address, a digital signature. That would certainly make it harder to fake thousands of personas.

However, anonymity has a vital role in civic life. Whistleblowers, undocumented residents, LGBTQ+ youth in unsupportive households, or employees criticizing their industry — these are people who need to be able to speak without putting their name on the line for fear of retaliation. If we build a system where only fully verified, fully public citizens can speak, we will silence some of the most vulnerable and important voices. Do we really want to trade one form of exclusion for another?

Free Speech vs. Flooded Speech

Another tension is between protecting free expression and protecting the conditions for expression to matter. In the US and many other democracies, you have the right to say almost anything, even anonymously. But you don’t have the right to have your speech amplified by flooding the zone with synthetic copies of itself.

We have no good legal or cultural framework for "flooded speech." We know how to handle one person lying, but we don’t know how to handle one person using AI to sound like ten thousand people. Is that still free speech, or is it a form of fraud against the public process? My view is that it’s closer to the latter, but drawing that line in law without overreach will be incredibly difficult.

What Real Accountability Could Look Like Without Building a Dystopia

So what do we do? I don’t think the answer is a single national ID for the internet, and I don’t think it’s hopeless either. The most promising approaches, including those explored by researchers like Roy at the MIT Center for Constructive Communication, focus on designing friction and responsibility back into the system.

First, we need to separate the ideas of identification and accountability. You don’t necessarily need to reveal your legal name to be accountable. You could participate through a system that verifies you are a unique human and a relevant stakeholder (like a resident of a district) without publishing your identity. Think of it like a cryptographic proof: "I am a real, distinct person affected by this policy, and I am willing to be contacted for follow-up by moderators," without saying "My name is Jane Doe at 123 Main St."

Second, we need to design platforms that reward answerability. Instead of just a comment box, what if civic platforms asked for a commitment to dialogue? A comment that includes a willingness to answer one follow-up question could be weighted more heavily than a drive-by statement. That small piece of friction — the expectation that you might be asked to elaborate — would be trivial for a real human but expensive to fake at scale with AI.

Third, institutions need to be transparent about how they weigh input. If agencies openly state, "We read all comments but give more weight to verifiable, deliberative input," it reduces the incentive to flood the system in the first place. The goal isn’t to censor synthetic text, but to make it less powerful.

My Take: We Need to Choose Friction on Purpose

For years, tech culture has worshipped frictionlessness. Make it easier to post, easier to comment, easier to share. That logic made sense when the bottleneck was human effort. Now that AI has removed the effort bottleneck, frictionlessness is a bug, not a feature.

This is the opinion part, and I’ll own it: I think we need to intentionally reintroduce friction into civic spaces. Not to make participation harder for real people, but to make impersonation harder for machines. A little more effort to prove you are a real, answerable human is a price worth paying to keep the public square from becoming a hall of mirrors.

That will require uncomfortable compromises. It will mean some people will have to do a bit more to participate. It will mean platform designers have to prioritize trust over engagement metrics. And it will mean we have to stop pretending that more speech, no matter its source, is always better for democracy.

The core insight from Roy’s work that stuck with me is this: words matter because people stand behind them. When we break that link — when we allow words to circulate with no one responsible for them — we don’t just get more noise. We erode the very idea that talking to each other can help us solve problems together.

AI didn’t create that problem, but it made it impossible to ignore. The question now is whether we will redesign our civic infrastructure to demand that someone, somewhere, is willing to say: "Yes, those are my words, and you can ask me about them."

This article was inspired by reporting on a Q&A with Prof. Deb Roy of MIT, as published by MIT News and discussed in his essay in the Bulletin of the Atomic Scientists (source). The analysis and opinions above are the author's own.

____________

This article was created by an AI writer using the topic ideas from this post => https://news.mit.edu/2026/qa-building-accountability-behind-our-words-and-ai-civic-conversations-1013