Artificial intelligence (AI) and machine learning (ML) hold tremendous potential to enable first responders to better process information and drive faster and more precise response. However, these capabilities present certain risks. The Science and Technology Directorate (S&T) is working diligently to identify the best ways to help the responder community strike that balance, while incorporating AI in a meaningful way into technologies used in daily operations.
The appeal of applying AI to emergency operations is that these emerging technologies offer advanced capabilities to process information faster and enhance decision making. S&T is taking a multi-disciplined research and development (R&D) approach where AI is concerned; this means consulting directly with first responder stakeholders to gather their operational needs, as well as identifying ways to leverage cutting-edge sensors and communications channels to provide clear, quick and actionable data to establish accurate situational awareness.
For instance, S&T is partnering with the WIFIRE Edge program at the University of California in San Diego, providing technologies to support generation of high-resolution information on the fire environment, where weather and location information can quickly change. Development of AI/ML technologies alongside Homeland Security Investigations (HSI) contributed to Operation Renewed Hope in 2023, which resulted in the identification and rescue of 311 previously unknown victims of sexual exploitation from active abuse and the arrests of suspected perpetrators. S&T also developed AI-based tools for HSI to help counter the flow of fentanyl, which the agency says contributed to a 50% increase in seizures and 8% increase in arrests.
S&T aims to apply lessons learned from these successes by providing AI-based tools to aid law enforcement and first responders nationally.
“We’re looking at decision support tools with the human-in the-loop, because our responders are pretty confident that they do not want to turn it all over to AI yet,” said First Responder Portfolio Manager Paul McDonagh. “With the rapid increase in capabilities, we are adjusting to the needs of our customers and their demands.”
Assessing the Needs of Those on the Frontline
S&T is currently gathering requirements to inform a gap analysis that will soon be published, entitled: “Gap Decomposition: Automated Alert and Execution System for Responders,” that recommended two phases of AI/ML development activities in support of first responders. This research involved collecting feedback from first responders directly and it will drive S&T’s R&D strategy moving forward—first, by focusing on AI-supported decisioning for Emergency Operations Centers (EOC). The second phase, focused on field operations, will also enhance operations decision making support and situational awareness by leveraging AI to provide required information and active monitoring of the situation in support of the operations commanders. Progress has already been made since its release, and next steps will be to complete detailed requirements and uses cases while identifying potential commercial, off-the-shelf solutions, with the goal of delivering solutions quicky and cost-effectively.
Brady Robinette, lieutenant with Lubbock Fire Rescue in Texas and a member of S&T’s First Responder Resource Group (FRRG), said AI could go a long way in support of first responders who have to quickly document extensive incident detail while bouncing from one crisis to another. During a busy shift, time can get away from you and things can be too hectic to complete documentation in real time. “It could be some time before you get around to writing that report, and how many calls have you run since then? You know, you’re sleep deprived,” he said.
“People that are trained in emergency medicine did that because they want to put hands on people and help people, and documentation is just a necessary evil,” Robinette said. “And, so, if we could leverage AI to complete like 90% of the documentation and we would just have to go in there to put the finishing touches on it and make a few corrections, I think it would be a huge improvement. I think it would lend to better patient care.”
At the 2024 FFRG gathering in Las Vegas this spring, volunteer members from across the U.S., representing every response discipline, shared some of the challenges they face on the front lines and how emerging technologies could address them. They highlighted several ways AI could help responders process information and provide early warning, with potential advantages of less risk to human life, less bias in decisioning, less risk of emotional interference, fewer operator errors, constant availability and increases in speed and accuracy. Some of the possible applications included:
- Monitoring crowd size and assessing related traffic flow and choke points
- Triaging patients during mass casualty incidents
- Supporting report-writing and detailed documentation
- Providing real-time language detection and language translation in text or voice format
- Managing 911 incident call overload to parse information from similar incoming reports
- Filtering background noise to enable responders to hear mayday calls amid chaotic scenarios
- Digital mapping of scenes where there is low or no visibility, like in a smoke-filled home or building
Piggybacking on the feedback from the FRRG, S&T has been conducting a series of Emergency Management of Tomorrow Research (EMOTR) tabletop exercises with the greater response community where AI has continued to generate buzz, according to McDonagh. “The responders agreed they would like a decision support tool where AI is processing multiple feeds of information and alerting when rapid changes impact their operational plan. This way, they can make decisions faster with more accuracy during dynamic incidents,” he said.
This spring, a joint S&T and Pacific Northwest National Laboratory (PNNL) effort resulted in publication of an AI Landscape Assessment, and the State University of New York at Albany held a one-day hackathon focused on exploring innovative AI applications for emergency operations. PNNL facilitated a similar EMOTR event with Wisconsin Emergency Management at the State EOC in Madison, where participants focused on exploring the EOC of the future with any eye towards identifying opportunities to incorporate AI and automation.
”Emergency managers face hazards that continue to evolve and we, at S&T, have the opportunity to help them use technology to tackle that constantly changing and always demanding threat,” said Daniel Cotter, Executive Director of S&T’s Office of Science and Engineering. “These EMOTR tabletop exercises convened emergency managers and first responders to assess the impacts and benefits of emerging technologies on EM organizations via real-world scenarios. Together, this feedback and R&D coalesced into recommendations for the EOC of the Future.”
Exploring the Art of the Possible
S&T experts also test and evaluate commercially available technology for public safety missions. Additionally, S&T identifies capability gaps and funds development of groundbreaking new technologies to meet the operational demands of first responders nationwide.
S&T’s National Urban Security Technology Laboratory (NUSTL) is a go-to resource for testing and evaluating life-saving emergency responder technologies, including technologies that incorporate AI/ML capabilities. NUSTL’s System Assessment and Validation for Emergency Responders (SAVER) program informs emergency responder equipment selection and procurement decisions, publishing reports that evaluate technologies, such as smart stethoscopes, crowd analysis technologies and AI-facilitated EMS call center software, as well as more high-level analysis like that provided in their report: Artificial Intelligence/Machine Learning Technology Uses for First Responders.
S&T previously partnered with a variety of New York agencies to enable first responders to test an AI computer vision technology. First responder evaluators tested this tech, which automatically detects and classifies weapons through integration with existing security cameras and associated video management systems, in an operational scenario to provide feedback on its features and suitability for urban first responder organizations. S&T was able to collect feedback from questionnaire responses and debriefing remarks and provide to the commercial provider for the purpose of tailoring the tech to better fit the needs on the ground.
“The AI solution is being developed by AI companies, but before it becomes public-facing, particularly if it’s going to be rights- or safety-impacting, we need to understand, ‘How do we test that system to the best of our ability?’” said Brian Henz, S&T senior science advisor for AI.
S&T has a number of tech development projects in progress that use AI/ML to detect objects and explosives, with potential law enforcement applications in the future.
S&T is also advising on a pilot project involving commercially available AI-facilitated call center software that utilizes large language models and ML and is being tested out in a few districts to provide supplemental information to reaffirm a live dispatcher’s judgment and actions. The findings of this pilot will inform S&T research and recommendations on wider application and whether, in the future, AI could possibly be employed to predict response times based on gathered data, locations and environmental conditions.
"Our upcoming research into AI-assisted call handling is expected to represent a significant step forward in enhancing emergency response capabilities while addressing the increasing workload on call takers. Similar capabilities have shown promising results in identifying high acuity calls such as strokes and cardiac arrest, potentially improving response times for these critical situations,” said Norman Speicher, Office of Mission Capability and Support Program Manager for the pilot.
Weighing the Risks
While there is a lot of excitement in the first responder community about the potential benefits this tech could bring to bear, there is also a hesitance due to the possible risks.
Megan Bixler, senior technology strategist for the Association of Public Safety Communications Officials International and FFRG member, sees many potential benefits to implementing AI in Emergency Communications Centers (ECCs), but also warns that as a costly solution, the benefits will need to be weighed.
“Emergency communications can realize several benefits through the implementation of AI solutions into ECC operations. Some of these benefits are speed and efficiency of inputting data to minimize response times, help with providing reliable information to fellow responders, automation and optimization to augment ECC staff, predictive capabilities, cybersecurity, and enhanced situational awareness that will keep fellow responders and citizens safe,” Bixler said. “However, there must be a balance between implementing technology without governance, conducting a cost-benefit analysis, and using the technology to its fullest capability.”
According to McDonagh, in addition to privacy issues and AI limitations being used by law enforcement and other disciplines, law enforcement is concerned about the use of deepfakes, people being accused inappropriately and also the potential for abuse, like swatting or making a false call. So, there will need to be an implementation that fully accounts and controls for risks before it is broadly incorporated into emergency operations. “They all want it, but they know there are pros and cons, and they want it done correctly.” he said.
This article is the second in a new feature article series dedicated to S&T’s AI/ML R&D efforts. Additional information can also be found the Artificial Intelligence and Autonomous Systems webpage. For related media inquiries, contact STMedia@hq.dhs.gov.