AI-Voyager links AI civic reasoning to the exact text of government documents
A research team describes a public-interest system where residents and AI jointly read, question and redraft AI governance rules, built on thousands of government documents.
In plain words
AI-Voyager is a research prototype that lets people explore government rules about AI with AI assistance. A user can find local rules, read them as an affected worker would, and draft changes. The system grounds answers in specific passages from official documents.
What was announced
The system, introduced in an arXiv paper, indexes 3,653 governance documents from city, state, federal and national governments as passages that retain their page and edition. Retrieval uses bounded agentic search that verifies each statement, so a resident can open the passage behind any answer or draft a revision to a rule on, for example, AI in hiring.
On 360 new questions requiring document retrieval, the authors report 83.0% macro recall versus 80.5% for the strongest baseline and 73.7% without iterative search, rising to 99.1% when documents were fixed in advance. In a study of 24 participants new to two AI bills, most proposed changes asked how provisions would apply; participants selected explanation lenses, and while citations signaled grounding, they rarely prompted participants to check sources. The authors present this as a design space for public participation in AI governance.
What it does not show
- Evidence is limited to the paper's abstract, so methods, deployment status and peer review are not established.
- Reported recall figures are the authors' own evaluations and have not been independently confirmed.
- The participant study involved 24 people new to two specific AI bills, so broader behavior is unknown.
Sources
Source-support review completed 17 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 17 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.