AI Emergency Alerts Are Coming. Humans Might Not Be Ready
Generative AI promises faster disaster communication, but risks accuracy and public trust during crises.
1AI in Crisis Communication
Emergency managers are increasingly turning to generative AI to draft public alerts during extreme weather events. While agencies officially restrict these tools, experts note pervasive unofficial usage. The technology is marketed for flood assessment and resource coordination, yet its primary appeal lies in rapidly generating clear, actionable public safety messages.
2Human Limitations in Alerts
Human emergency managers often struggle with effective communication due to high stress, limited resources, and the use of technical jargon. These officials, who serve as the first line of defense, frequently omit critical information or include irrelevant details. Such communication failures have historically led to public confusion and tragic outcomes.
3The Cost of Miscommunication
Inadequate warnings have caused significant loss of life and public distrust. Following the 2017 Thomas Fire in Montecito, California, unclear evacuation alerts contributed to a mudslide that killed 23 people. Residents experienced warning fatigue and confusion, highlighting how poor messaging can be the difference between life and death during disasters.
4AI as a Force Multiplier
Large language models can process vast amounts of research and historical alert data in seconds. Experts like Rob Dale suggest that AI can outperform humans under pressure, especially for officials who rarely send alerts. By synthesizing best practices, AI helps draft precise, directive messages that human managers might overlook during emergencies.
5Improving Alert Quality
Research by Jeannette Sutton demonstrates how AI models can significantly improve alert clarity. In one instance, an AI model corrected a vague, error-prone flood warning into a structured message naming the specific location, evacuation instructions, and safety advice. These refinements ensure that critical information reaches the public without unnecessary, confusing jargon.
6Unofficial Adoption Trends
While official government adoption remains low, many emergency managers use personal accounts to access chatbots like ChatGPT or Gemini. Instructors at the National Disaster Preparedness Training Center report that trainees frequently present high-quality, AI-drafted alerts. This shadow usage indicates that the technology is already integrated into daily disaster planning workflows.
7Startup Solutions for Agencies
Companies like EM1 are developing specialized AI software trained on disaster research to outperform human managers. These models have been tested against wildfire and tsunami scenarios, matching or exceeding human benchmarks. Although these startups hold contracts with numerous agencies, current usage is primarily limited to internal planning and message workshops.
8The Authenticity Dilemma
Experts warn that relying on AI could backfire if the public perceives the content as inauthentic. Skepticism toward AI-generated graphics or social media posts may erode the trust that emergency managers rely on. In Syracuse, New York, public employee unions even sued over the use of AI for answering non-emergency calls.
9Risks of Warning Fatigue
Automating alerts could lead to an overabundance of messages, potentially increasing warning fatigue among residents. If managers send too many alerts, the public may stop paying attention to critical updates. Balancing the speed of AI-generated content with the need for meaningful, targeted communication remains a significant challenge for disaster departments.
10Human-in-the-Loop Necessity
Current technology is not yet capable of operating without human oversight. Seasoned managers can leverage AI to speed up communication, but inexperienced users may fail to verify inaccurate outputs. As climate disasters become more frequent, the industry must determine how to maintain public trust while utilizing AI for life-saving alerts.
Original article: Wired — AI Emergency Alerts Are Coming. Humans Might Not Be Ready