Directorate Aids AI Adoption for Homeland Security – Signal Magazine

  • Lauren
  • March 1, 2022
  • Comments Off on Directorate Aids AI Adoption for Homeland Security – Signal Magazine

The Department of Homeland Security Science and Technology Directorate could release its artificial intelligence and machine learning strategy implementation plan as early as this month and is growing a community of interest to foster the adoption of the technologies across the department.In August, the Science and Technology (S&T) Directorate released its first-ever artificial intelligence (AI) and machine learning (ML) strategy document. The strategy presents three goals: driving next-generation AI and ML technologies for cross-cutting homeland security capabilities, facilitating the use of proven capabilities for homeland security missions and building an interdisciplinary workforce trained in the technologies. It will likely be reassessed each year and updated when necessary.
The directorate’s strategy complements the department-wide strategy, which was published late in 2020. That document sets five departmental goals: assessing the potential impact of AI on the homeland security enterprise; investing in AI capabilities; mitigating AI risks to the department; developing an AI workforce and improving public trust and engagement.
Department of Homeland Security (DHS) officials describe AI as a game-changing technology. “Through this strategy, S&T will build and apply expertise to help the department fulfil the game-changing promise of AI/ML technologies while mitigating the inherent risks,” Kathryn Coulter Mitchell, the senior official performing the duties of the undersecretary for science and technology, writes in the directorate’s strategy document. “As we grow S&T’s proficiencies in artificial intelligence and machine learning, we ensure S&T continues supporting the department’s vision to enhance its capability to safeguard the American people, our homeland and our values through the responsible integration of artificial intelligence into the department’s activities.”
In the months since the strategy was published, the S&T directorate has developed an implementation plan that could be released this month or next, created multiple AI/ML-related working groups and is building a community of interest to help solve some of the challenges to AI adoption.
John Merrill, acting deputy director for the Technology Centers Division within the S&T Directorate, says that to create the pending implementation plan, the directorate worked with the department’s operational elements, including U.S. Customs and Border Protection, U.S. Coast Guard, the Transportation Security Administration and the Cybersecurity and Infrastructure Security Agency. “This is a collective effort. Because of the magnitude of AI and machine learning, ML, within the confines of the department, it has to be a collective effort. It has to be teamwork.”
Directorate officials were expected to hold their first implementation meeting in January with three working groups. The working groups likely will meet biweekly to determine the best path for AI/ML adoption. “Ultimately, our goal is to … have an actionable path or road map per se on how we can move this into action,” Merrill offers.
With agencies across the department already investing in or exploring potential investments in AI and ML, the plan may support future requests for additional research and development funding. “Our ultimate goal is that the implementation plan, when we push it out to the other components—and obviously, working within DHS S&T—is to use it as a guide to help us build our budget so that we have the appropriate justifications in place when we start moving forward and asking for additional funds,” Merrill explains, adding that officials hope to release the document this month or next.
The initial focus will be on AI/ML used for cyber infrastructure. S&T officials work closely with the experts at the Cybersecurity and Infrastructure Security Agency. “If you take a look at AI and ML … it is very tightly coupled with the cyber infrastructure as a whole,” Merrill explains. “Before we get to actual implementation, we need to look at the algorithms associated with AI and ML. When you do the algorithm assessment, you are actually going in and doing a cyber assessment as well.”
Merrill says he would like to see a lot of advances in AI and ML, including standards and benchmarks, but the advances S&T will push for will be largely driven by the programs and projects from across the department. “Reliability, accuracy, usability, interoperability, and security and privacy—those are some of the typical areas we look at for AI/ML implementation, but it’s going to be highly dependent on whatever the project or program is that we’re trying to address.”
In addition to the soon-to-be-released implementation plan, S&T Directorate officials are developing a community interest, or COI, for sharing information and overcoming challenges. If, for example, officials from one agency need a particular AI capability, they may discover another agency already is acquiring that capability and can share lessons learned. Additionally, by combining efforts, the two agencies could save time and money.
Merrill says he expected no more than 30 people for the first meeting held last fall, but more than 50 participated. Furthermore, the distribution list already has grown to about 75. Ultimately, personnel from the national laboratories, federally funded research and development centers, and industry also will be invited.
The group was expected to hold its second meeting in January. It is scheduled to meet monthly, but the frequency of meetings will be driven by the “operational tempo” and could be as often as weekly when necessary, Merrill suggests.
The COI will use an online platform for information sharing and collaboration but has not yet determined which one. “We’ve got some internal platforms we can use within DHS, but right now we’re still in discussions about what will be the best area for posting information and for collaboration,” Merrill says. “What we don’t want to do is reinvent the wheel and bring in some other capability nobody has access to.” Microsoft Teams, a part of Microsoft 365, is a possible solution, he adds.  
DHS S&T also is working with the National Science Foundation’s AI Research Institute program, allowing collaboration with researchers and experts in academia. If the COI personnel need an answer to a particular question, the institute researchers may be able to respond within a matter of days, Merrill says.
AI and ML systems can perform a number of tasks and support an array of homeland security missions. For instance, it might be applied to the Advanced Imaging Technology program, which uses backscatter X-rays and millimeter-wave devices to detect objects of concern carried into airports. The S&T Directorate works with the Transportation Security Agency on the program.
The goal is to use AI and ML technology to achieve a 99 percent accuracy level in detecting explosives or other concealed threats. “Granted, that’s very high, but you want to get to that point,” Merrill says. “[Advanced Imaging Technology] will be much faster than the human and transitioning through the checkpoint will be quick and easy.”
In some cases, the technology will perform rote or menial tasks, allowing agents to focus on more important duties. For example, watching multiple monitors for illegal border crossings can quickly become boring for humans, possibly resulting in distractions. “Let’s look at the southwest border, the number of cameras that are out there and having the visualization capabilities to decipher the difference between a human versus an animal. If a CBP agent has a number of screens on his desk, he doesn’t have to monitor those screens because the AI would provide that capability,” Merrill suggests.
Furthermore, AI and ML systems can be useful for assessing cargo containers at ports for potential threats. “If you have ever seen the Los Angeles Long Beach port of entry for those cargo ships and the number of containers that are there, having an AI/ML algorithm for screening capabilities would go a long way. Instead of having a human behind that screen identifying each cargo container that’s coming through, you could train that AI/ML capability to do that on their behalf.”