When policymakers draft regulations, a public comment period is usually opened. Legislators have long learned to spot mass‑produced fake feedback, but artificial intelligence now enables submissions that look authentic, making it harder to tell whether a real person is behind them. In an essay titled “Who stands behind the words?” published in the Bulletin of the Atomic Scientists, MIT professor Deb Roy, director of the Center for Constructive Communication (CCC), argues that even when AI crafts text, accountability and answerability must remain. CCC and its affiliated nonprofit Cortico have facilitated group discussions in civic meetings, schools, deliberative assemblies and private organizations, using tech tools—including AI—to surface themes in comments while allowing participants to inspect and override algorithmic outputs at every key step, ensuring a human is ultimately responsible for each public remark.
Roy says the issue came into focus with the launch of ChatGPT. Although his PhD work dealt with what he now calls “Puny Language Models,” the depth and interactive capability of ChatGPT revealed a missing element: standing. In legal terms, standing means the party bringing a claim has a concrete interest; Roy suggests a similar informal notion should apply to everyday discourse—speakers should be real individuals with a stake in the issue. Large language models can generate fluent prose but lack any entity that can be held to account.
A striking example is the 2017 Federal Communications Commission net‑neutrality proposal, which received over 22 million comments, nearly 18 million of which were later identified as fabricated. Those fake submissions left “fingerprints” such as templated phrasing; modern LLMs erase those traces, leaving reviewers unable to distinguish genuine voices and eroding the legitimacy of public input.
Roy pairs the concept of standing with “answerability”: anyone granted a right to be heard must be prepared for cross‑examination and to back up claims with facts. Privacy does not have to conflict with answerability—institutions can verify a speaker’s eligibility without exposing their identity. AI can assist in drafting, summarizing, and pattern detection, but a person must ultimately stand behind the words. To embed this, he outlines four design principles: attributable authorship, a decomposed and inspectable process, provenance as infrastructure, and widening rather than concentrating standing. The guiding rule is that autonomy is permissible only when it is not unaccountable.
The phrase “at the speed of trust, not at the speed of tech” emerged from a conversation with community organizers in Appleton, Wisconsin. Trust takes time to earn; once established, technology can be deployed rapidly. If AI moves faster than democratic processes, decisions may be made without public oversight—a breaking‑point scenario already unfolding.
CCC’s work includes the Real Talk initiative, the Youth Voices program, and the New Social Forms project. Real Talk creates small, consent‑driven spaces where participants share authentically, while technology surfaces larger patterns. Youth Voices brings the same approach to high schools, enabling students, teachers and administrators to co‑create policy recommendations. A prototype called VoiceTrace, built by PhD student Cassandra Overney, links every summary or recommendation back to the specific voices that produced it, ensuring traceability. New Social Forms examines workplace collaboration, where AI empowers individuals but can fragment team ideas. By extending early collaborations with Google, the project aims to build collective communication tools that preserve accountability in civic‑like work environments.
Review: The interview highlights the hidden dangers of AI‑generated civic discourse and offers concrete design principles to keep humans at the end of the accountability chain, reinforcing the health of democratic dialogue.