Decoding the Municipal AI Landscape: Generative AI-Driven Insights for AI Governance and Application in the City of Seattle
Human-AI Collboration Workflow to Identify Relevant Contents in Public Documents(In Progress)
This work is in development with Meixin Yuan, Jack Bernard, and Qifan Wu.
Abstract
We explore a novel human-AI workflow for identifying and categorizing grey literature related to AI governance and application, using the City of Seattle as a case study. Our goal is to move beyond simple keyword searches to accurately detect predefined AI topics, retrieve supporting evidence, and minimize hallucinations common in end-to-end LLM applications. The ultimate vision is to enable the public to find relevant AI topics in a large volume of public documents and provide a mechanism to trace back to the original supporting texts.
Our workflow has iterated on chunking strategies, retrieval methods, and opportunities for human validation. While the workflow remains in an early prototype stage, we share key challenges, design decisions, and preliminary findings that shed light on the complexity of using generative AI to understand municipal AI landscapes. We applied our methods to all public documents released by City of Seattle as a case study to test validity.
Keywords: AI governance; stakeholders; network
