Report: Agencies Should Turn to AI Before Disaster Strikes
But government agency-led applications have much room for improvement.
NASA-funded researchers applied artificial intelligence to Facebook user location data captured as two fires wrecked northern California in 2018 and gained new insight into people’s evacuation movements and behaviors when disaster strikes, which could strengthen future response. The Defense Innovation Unit and Carnegie Mellon University’s Software Engineering Institute are collectively crafting datasets to teach AI tools to assess buildings and structures after natural crises occur, and ultimately augment and increase the accuracy of damage estimates.
These are two of many examples detailed in a new report from the Partnership for Public Service and Microsoft that explores how the maturing technology can improve disaster resilience and response, and considerations and actions governments should pursue when adopting AI to boost preparedness, recovery and relief. The report suggests agencies improve data collection and access, make proactive instead of reactive moves, collaborate with other organizations—and more.
“While some governments, companies and universities have already used AI in this field, most are still in the early stages of use,” officials wrote in the report. “However, AI technologies could contribute to disaster preparation and response better than any other technology or innovation in operation now.”
The organizations interviewed 23 employees with relevant experience from either federal, state and local government, academic institutions, or private sector insights and also examined scientific studies and government preparedness plans to inform the report. “It’s notable how well-suited AI is to improving disaster resilience, and AI’s impact on disaster resilience could be substantial and positive,” Partnership for Public Service Vice President Katie Malague told Nextgov Friday.
“Disaster resilience is heavily dependent on information collection, analysis and dissemination, and those are areas in which AI tools made great advancements over the past two or so decades,” Malague said.
Defining AI is an ever-evolving feat, but in this case, researchers refer to the technology as “computers and software performing tasks we typically associate with people, such as recognizing speech or images, predicting events based on past information, or making decisions.” Malague said while every organization works with its own definition, there are elements of those definitions that are common.
“For example, most definitions compare AI capabilities to what was dubbed human intelligence, or tasks that humans typically do,” she said. “You see that as the basis of our definition.”
Researchers highlight a range of instances throughout the report that demonstrate the technology’s capabilities to help forecast future and impending disasters in real time, assess damage in the aftermath, and predict and evaluate impacts. In terms of the latter, the report notes that AI tools “could assess the impact of natural disasters, using data on people and details about communities, such as the age or height of structures, the number of hospital beds or the location of fire stations,” counterpose the information captured with past data, and ultimately identify areas that are vulnerable to certain events, “or where and how quickly a neighborhood could be evacuated from the path of a tsunami, based on population, evacuation routes and land elevation.”
The research mentioned above, which tapped location data from Facebook users who opted-in to be tracked during two destructive northern California wildfires in 2018, revealed evacuation patterns that demonstrated that shelters are not the only options people pursue when forced to leave their homes. Looking at movements of people returning home in the aftermath also shed light on areas where houses were destroyed versus others where they were habitable.
“Analyzing information like this could help emergency responders understand how people behave during disasters, which could inform future emergency response efforts, according to researchers,” the report notes. “It could also inform officials, who must decide how many shelters to set up or how many evacuation routes need to be created. And it could reveal factors that speed or slow evacuations, which could save lives during fires.”
Other examples presented highlight how data supplemented by artificial intelligence could improve the detection of disasters. Compared to employees using traditional earthquake detection tools within the Oklahoma Geological Survey, a state agency, for example, researchers from Harvard University and the Massachusetts Institute of Technology found more information and earthquake occurrences by leveraging AI analysis of seismological datasets, in a 2018 study.
“Government employees, such as those working for the Oklahoma state agency, could accomplish their mission more effectively with AI tools that help them analyze data and information,” the report notes.
A director within the National Oceanic and Atmospheric Administration also detailed how the agency’s tsunami-detection capabilities could be supplemented by AI.
A large portion of the report focuses on how the technology can boost California’s resilience and response, in particular, because, according to the researchers, “few places are likely to feel AI’s impact more than California, which is prone to more natural hazards than almost any other state in the nation.” A case study included in the analysis offers an in-depth look into the impacts that the AI-powered online software WIFIRE has had on the abilities of several California fire departments to predict wildfire’s spread. Data from remote weather stations, still images from cameras around the region, and satellites underpin the technology’s prediction, and in late 2019, the Orange County Fire Authority launched a pilot program with commercial companies to create an additional AI tool that further confirms WIFIRE’s predictions. In the report, the Los Angeles Fire Department’s chief said the technology has enabled the team “to more accurately send sufficient firefighters, fire engines and helicopters to the fires within minutes as compared to hours.”
Though AI is introducing promising results, there’s still a long way to go before federal and state agencies fully realize its potential. According to the report, agencies and organizations should strive to enhance data access, quality and usability. Malague noted that during the research it was “also interesting to see how the quantity of data available for AI is so different for different types of disasters,” noting that NOAA has abundant data on hurricanes, but predicting tornadoes requires years of collecting data before there’s enough captured to sufficiently train an AI tool.
The researchers reviewed government emergency and disaster-focused plans to prepare for the report, which Malague said “often rely on using information from disparate sources to, for example, respond to a disaster, whether it’s from different government agencies, social media, 311 calls or 911 calls.” AI holds the potential, however, to “pull together information from all those different sources and help ensure that first-responders do not miss important pieces of data that could make their disaster response more effective,” she said.
Increasing the use of AI in disaster response also demands that government officials shift to a more proactive approach, which could also prove to be financially beneficial, the report highlighted. The Federal Emergency Management Agency has shown that $1 spent on mitigation saves $4 in recovery costs, on average. Increased collaboration, data integration and platform-sharing must also occur across governments, and tools that are created to enhance response must be incredibly user-friendly for the emergency responders tapping into them. Human expertise to check and improve AI’s support is necessary as well, and subject matter experts with experience across disasters and weather events—and data science—will better ensure success.
“Adopting new methods or technologies often requires considerable change management across disciplines, and it can be an uphill climb for government at all levels to successfully adopt new practices,” Malague noted. “That said, profiling case studies of where these efforts worked well can help outline a roadmap for other organizations.”