MIT’s Deb Roy urges human accountability for AI-written civic speech

AI can make fabricated public feedback sound like a crowd of distinct people, weakening trust in a channel meant to give citizens a voice. In an October 13 interview with MIT News, professor Deb Roy argued that the safeguard should be human accountability: someone must stand behind civic speech, even when AI writes it.

Speaking with MIT News writer Peter Dizikes, Roy discussed his essay, “Who stands behind the words?,” published in the Bulletin of the Atomic Scientists. Roy, a professor of media arts and sciences and director of MIT’s Center for Constructive Communication, draws a distinction between using AI to help someone express a view and allowing generated words to circulate without anyone responsible for them.

When fake comments lose their fingerprints

Roy traces his concern to conversations with government staff who review comments on proposed rules. After ChatGPT’s release, he began hearing suspicions that some submissions were not human-written. His worry is not simply that officials might count fake comments. He believes reviewers have quietly reduced the weight they give public feedback, making genuine participants less confident that anyone is listening.

As historical context, he points to the Federal Communications Commission’s 2017 proposal to repeal net neutrality rules. It received more than 22 million comments; Roy says the New York attorney general later found nearly 18 million were fake. Repeated phrasing and templates left signs of that fraud. His warning is that AI language models can produce fluent, varied comments, removing those clues without supplying a real person who holds the expressed view.

Standing and answerability have to travel together: The right to be heard comes with the obligation to answer for what you say.

Deb Roy, in his MIT News interview

Accountability without a public name

Roy uses “standing” informally to mean having a stake in a decision and a basis to be heard. At a town meeting, showing up creates friction, and participants generally expect speakers to have a local connection. He argues that language models and digital networks collapse those boundaries. “Answerability” adds an obligation: a speaker must be ready to explain a statement or support an accusation with facts.

That does not require every speaker to publish their name. Roy argues that an institution can verify a person and stand behind their words without exposing their identity publicly. Expressing a view requires showing that a distinct, eligible person is speaking. Making an accusation requires more: a trusted intermediary, such as a reporter, must be able to know the speaker’s identity. He proposes greater disclosure for powerful people and stronger identity protections for vulnerable ones.

Nor is his proposal a ban on AI assistance. Someone affected by a rule can use AI to draft a comment, adopt it as their own and be ready to answer for it. Roy considers that legitimate. His essay names four principles for designing accountable systems:

• Attributable authorship.
• A process broken into steps that can be inspected.
• Provenance—tracking where material comes from—as infrastructure.
• Widening who has standing rather than concentrating it.

Keeping participants able to overrule the technology

The Center for Constructive Communication and Cortico, the affiliated nonprofit Roy co-founded, already use technology, including AI, to identify themes in group discussions. Those discussions span civic meetings, schools and other organizations. The center’s human oversight process lets participants check and override technology-based outputs at key steps, with a person ultimately responsible for each public comment.

Roy also challenges the idea that trust necessarily means moving slowly. He recalls a community organizer describing growth as happening “at the speed of trust.” Roy initially heard that as a call for delay, then came to see that established trust can let people move quickly. His warning is about mismatched clocks: if AI deployment outruns democratic processes, decisions may be made without the public.