Breaking01 / AI
AI-Voyager aims to help citizens revise AI rules
The proposed system links answers to government documents, but study participants rarely followed citations to check the source.
Researchers introduced AI-Voyager in an arXiv paper on 16 Nov 2026, proposing a system that helps citizens work with AI to understand AI rules and draft revisions.
According to the paper, users can find local rules on AI in hiring or upload a national AI strategy. The system lets them read these through an affected worker's perspective. They can open passages supporting an answer and draft changes.
The researchers built the search collection from 3,653 government rules and policy documents. These are split into passages that retain page and edition details. The system is designed to search repeatedly within set limits and check each statement against evidence.
The researchers reported a study involving 24 participants unfamiliar with the two AI bills under examination. Most proposed revisions asked how provisions would apply. Participants treated citations as signals of supporting evidence but rarely checked the original text.
The paper reported search results on 360 new questions requiring the system to find relevant documents. It measured macro recall, an indicator of how much of the required evidence was retrieved, at 83.0%. The strongest comparison system scored 80.5%, while a version without repeated searching scored 73.7%; these are the researchers' own results.
The abstract did not disclose whether participants' proposed revisions were incorporated into policy.