MIT Professor Deb Roy Wants Every Public Text to Have Someone Answerable for It

MIT professor Deb Roy has articulated a principle that sounds like a rule from accounting but hits generative AI where it is most vulnerable: someone specific should answer for every piece of text in public life. The word he uses is "answerable": the person who is prepared to answer for what was written. Where the text came from does not matter. A human may have written it, a language model may have written it, or the two may have written it together. The interview appeared in MIT News on 13 October 2026 and focuses on civic conversation, the space where the line between a human voice and model output blurs fastest.

The announcement of the piece contains no figures and no names of specific products. It is a condensed statement of a position, so what follows covers that position and its implications, not research results.

Who Roy is and why he is talking about this

Roy has spent many years studying how public conversations work and how technology affects them. He heads the Center for Constructive Communication at the MIT Media Lab and previously served as Twitter's chief media scientist. For him the topic is practical: the question is how to build spaces where people hear each other instead of drowning in a stream of automated noise.

The logic of the principle

The idea ties responsibility to the act of publication. The author can be anyone or anything, and the signature under the text belongs to whoever answers for it to readers. Several things follow from this.

• An "AI-written" detector stops being the central tool: what matters more is knowing who released the text into the public sphere.
• An anonymous text with no accountable party falls outside the category of full participation in a discussion.
• Using AI as a tool is acceptable as long as a person or organization is ready to vouch for the result and defend it.
• Autonomous agents that publish text without an owner fall outside such a rule.

Where it gets hard

The principle is simple on paper and hard to implement. A single remark in a public discussion can be produced by a chain of parties: the model developer, the agent operator, the platform, the client behind a campaign. Which of them is "answerable" is determined not by technology but by rules and agreements. Some regulators are already moving in this direction: the European AI Act, for example, contains transparency requirements for synthetic content. But labeling answers the question "who generated this," while Roy asks a different one: "who answers for this."

Scale is a separate problem. When text costs next to nothing and the number of comments, letters, and appeals to government bodies keeps growing, checking each author's accountability by hand is impossible. What is needed are verification mechanisms that do not turn the public sphere into a system of total identification.

What you can do right now

Editorial teams and companies do not need to wait for regulators to apply this logic.

• Assign a named owner to every public channel where AI operates: a blog, a chatbot, a newsletter, an agent's account.
• Record in the publishing workflow who approved the text, even if a model wrote the draft.
• For AI agents that write on behalf of an organization, decide in advance who is responsible for a mistake and to whom complaints should be addressed.
• Do not substitute a "created by AI" label for accountability: it says how the text was produced, but not whom to approach with a question.

For now this is a statement of a position, not a standard. But the question "who is answerable" will still have to be settled by everyone who publishes text produced with the help of neural networks.