MetroWest emergency managers hear AI briefing on use, limits

Madison Hoadley, AI Specialist for EM1, briefed the MetroWest Regional Emergency Planning Committee on April 2 on how artificial intelligence can be used by emergency managers, the risks involved and practical first steps for adoption.

Hoadley outlined an AI adoption maturity model and described real-world use cases for emergency management, emphasizing that tool selection matters: hallucination risk depends on whether a tool is open or closed and on the underlying large language model. During a follow-up discussion, she advised that adjusting a system’s outputs often depends on user feedback and iterative prompts rather than a single setting.

Committee members raised operational questions. Brian Simeone asked how often AI tools produce false or fabricated information; Hoadley said the frequency depends on model architecture and training data and recommended validation of AI-produced content before operational use. Chris Hampton asked how to make text ‘‘sound’’ less like it was generated by AI; Hoadley recommended guiding tone through explicit feedback to the system and refining prompts.

Michael Cassidy noted that the slides used for the REPC presentation were produced from an outline provided to the AI tool, and Hoadley confirmed the deck was an abbreviated version of a presentation used at a recent EM conference. She also shared contact information for follow-up assistance and next steps to pilot AI tools safely in local offices.

The briefing closed with committee members noting interest in practical demonstrations and in guidance on validation and documentation of AI outputs before they are used for public communications or operational decisions.