In the fall of 2024, Hurricane Helene made landfall on Florida’s Gulf Coast before tearing north and dumping catastrophic rainfall on western North Carolina, creating one of the worst natural disasters the region had ever seen. More than 100 people died as flash floods swept away buildings, roads and critical infrastructure.
The widespread loss of power spread through the area, knocking out cellular towers, fiber optic cables and broadband systems — the same infrastructure that emergency responders used to communicate with each other.
“When things fail, residents can’t call for help, coordinating entities couldn’t speak to each other, and systems that we’ve created that need to be online were dependent on wires and signals,” said Sara Nichols, energy and economic development program manager for the Land of Sky Regional Council, which coordinates planning and development in four North Carolina counties.
Download the Straight Arrow app today to get the stories that matter free from manipulation, bias or agenda.™
Point phone camera here
Today, emergency management offices have a new tool to help them better prepare for and respond to disasters: artificial intelligence. But these tools share the same limitation as the cellphones and radios Helene knocked out — they need an internet connection to work.
That vulnerability is now backed by data. The AIDE Initiative, a research partnership between the Markle Foundation, Aspen Digital and RAND, recently released the first major study of AI in emergency management. It cataloged nearly 1,179 AI products from more than 710 companies. The findings suggest the tools built to help agencies prepare for and respond to disasters may be least reliable when they’re needed most.
What the research found
The study found that of the nearly 1,200 products researchers reviewed, 78% required a continuous internet connection to operate. But that dependency matters more at some points in a disaster than others.
Emergency management is generally broken into three major phases: preparedness, response and recovery. Preparedness is everything that happens before the storm, like planning, training and stockpiling supplies. Response is the immediate phase during and immediately after a disaster strikes. This can include search and rescue, emergency shelters and coordinating resources in real-time.
Recovery comes after the immediate crisis passes. This is when emergency response teams assess damage and help rebuild and restore infrastructure.
Jessica Jensen, senior policy researcher at RAND, told Straight Arrow that most of the AI tools she researched were marketed for the response phase of the disaster. She said that is problematic because most tools require a constant internet connection, and that’s not always available. For products more focused on preparedness and recovery, the internet is less of an issue.
“If you’re looking for a recovery product, recovery can last years after a disaster, then that internet connection is not as much of a problem,” Jensen said. “So it’s not as huge of a barrier, but it is a real barrier if you’re looking for a response product, potentially.”
Jensen said only about 11% of the tools in her study could tolerate intermittent connectivity, meaning the vast majority aren’t built to keep working when a disaster knocks out the internet.
But Jeremy Greenberg, senior adviser at the Aspen Institute, who helped research the real-world adoption of AI tools in emergency management, said many of those tools are intended for use inside emergency operations centers. Those facilities typically are outside the areas directly hit in a disaster and may have backup power and cell systems in place.
Greenberg cautioned against dismissing AI tools simply because they rely on an internet connection.
“Out of the tools that were assessed that do require connectivity, many of them … can work offline, and then we’ll plug back in once that connectivity is restored,” he said. “That in itself should not be a barrier to adoption for an AI tool.”
But using AI in emergency management raises another issue: privacy.
Jensen said the tools collect data like geolocation and personally identifiable information about people affected by a disaster.
“It’s not just that it’s gathering that data in their communities,” she said. “It’s that then the vendor is also holding that information, and it’s important to understand not just what data it’s collecting, but what the vendor’s rights are.”
‘Sweet spot’
Although emergency management offices play a crucial role in a community, more than half of local agencies operate with no more than one full-time employee, if any at all, according to the report.
“It’s not that the intelligence isn’t there,” Greenberg said. “It’s not that the knowledge isn’t there. It’s a time and capacity problem. Every emergency manager, whether you’re an office of one, one and a half, or 400, you are already stretched thin.”
But even smaller offices that don’t have the resources to navigate the market alone can find a practical starting point for choosing AI tools to help deal with disasters, Jensen said.
“If you are a low-resourced jurisdiction that doesn’t have copious information technology staff or funding, I would start with looking at large language models and general-purpose tools like translation products,” she said. “They tend to be low-cost and user-friendly products … that’s going to hit that sweet spot.”
Beyond individual product choices, the AIDE Initiative is also producing a practitioner playbook to help agencies evaluate their own needs before purchasing any AI tool. The group is hosting a summit in Austin, Texas, in September, bringing together technologists, academics and practitioners to keep the conversation going.
Greenberg said the research is a way to begin the discussion of AI tools in emergency management.
“We are not solving every single problem that’s out there,” he said. “That was not the intent, but the intent is to motivate the conversation and to make sure that our research is being used as a baseline to inform future action.”
Round out your reading
- All your questions about napping, answered.
- Trump’s $5,000 promise echoes past payouts that never materialized.
- Inside the effort to make data centers pay their share of electricity costs.
- Why did the Feds seize the ’largest Martian meteorite on Earth’?
- Photos and video show exactly where and when Trump visited Ground Zero after 9/11.
