The Department of Homeland Security (DHS) has successfully completed the first phase of three generative AI (GenAI) pilot programs, gaining the knowledge necessary to keep rolling out the developing technology for a variety of missions, a top DHS IT official said today at IBM’s Think Leadership Exchange event in Washington, D.C., produced by MeriTalk.
David Larrimore, executive director of the Chief Technology Officer Directorate in DHS’s Office of the Chief Information Officer, said the pilots – which involved improving investigative summaries and training immigration officers to interview asylum seekers – “created technical infrastructure and knowledge” that will enable DHS agencies to “test out these technologies.”
But Larrimore also emphasized the pilots showed that data quality needs to be substantially improved to enhance the performance of GenAI models before DHS can fully integrate the technology.
“The hardest part we had was recognizing that our data was – not to use the term garbage but it was garbage,” he said. “We had to work really, really hard to make sure that the LLMs (large language models) could use it in a performance way.”
“Once we, as a department, solve that data problem,” he added, “we’re immediately going to see a huge cliff that we’re going to be able to jump over in leveraging generative AI to support these cases.”
DHS has been moving rapidly on AI and GenAI, and in March became the first Federal agency to roll out a comprehensive AI roadmap to integrate the technology into a variety of uses.
Among the DHS initiatives was the launch in February of an “AI Corps” designed to hire 50 new AI experts to work on projects such as helping to combat child sexual exploitation and abuse, delivering immigration services, securing travel, and fortifying critical infrastructure.
The hiring initiative stems from last year’s White House executive order that focuses on promoting and adopting safe AI technology to help advance Federal agency mission goals.
Larrimore announced in March that DHS had already received more than 2,000 applications for the AI Corps, and DHS recently said it has onboarded 31 technology experts as part of the initiative.
Larrimore has been something of an AI evangelist for DHS, calling publicly for increased use of AI and speaking regularly at technology events. He was featured earlier this year as one of WashingtonExec’s “Top AI Execs to Watch in 2024.”
DHS announced last month the completion of the first phase of the GenAI pilots that Larrimore discussed today. They took place in three DHS components: U.S. Citizenship and Immigration Services (USCIS), Homeland Security Investigations (HSI), and the Federal Emergency Management Agency (FEMA).
The USCIS pilot involved a training application that allowed immigration officers to interact with a GenAI tool to practice interviewing potential refugees seeking asylum.
The HSI pilot used LLMs to produce summaries of law enforcement reports to bolster efficiency and precision, while FEMA’s pilot used an LLM to help state and local governments generate draft hazard mitigation plans to better respond to emergencies.
Mohamed Ali, a senior vice president who heads IBM Consulting, discussed the three use cases with Larrimore at today’s event, saying he found them “very fascinating” and describing the USCIS pilot as “probably by far the most successful.”
Larrimore agreed, saying USCIS is confident that the GenAI tool that helped agents interview asylum seekers will be “an effective long-term mechanism for training.”