Chris Kraft, Deputy CTO, AI/Emerging Technology at the U.S. Department of Homeland Security (DHS), speaks with Adita Karkera, Chief Data Officer, Government and Public Services at Deloitte, in a video interview about the three generative AI (GenAI) pilots spearheaded by the DHS, the challenges in the AI journey, and steps taken to address them.

At the onset, Kraft highlights three GenAI pilots that were chosen through the AI task force and included in the DHS AI roadmap released in March 2024. The first is with the U.S. Citizenship and Immigration Services (USCIS), wherein a GenAI chatbot is used to train officers who work with refugees and asylum seekers.

Kraft says that it relies on prompt engineering rather than a Retrieval-Augmented Generation (RAG) approach to provide details about the refugee or asylum seeker. He appreciates the team at USCIS for setting up a system that provides human-like answers. The tool offers supplemental and on-demand training, helping officers become more comfortable and efficient in their work.

The second pilot is with Homeland Security Investigations (HSI), says Kraft. He mentions that the agency has loaded millions of investigation records into a vector database, and users can search through that data by asking questions and generating summaries of the results.

During testing, HSI agents found connections like certain dark web monikers, says Kraft. This highlights the value of natural language processing and embedding information, then using AI to spot similarities and produce summaries.

The third pilot is with the Federal Emergency Management Agency (FEMA), and it is using GenAI to help develop hazard mitigation plans for underserved communities. These plans, which can be hundreds of pages long, are required to obtain funding, but smaller communities often lack the resources to put them together.

By using GenAI, FEMA aims to gather relevant information and automatically generate the plans, easing the burden and helping these communities secure the support they need, says Kraft.

When asked about challenges in the AI journey, he stresses ensuring having the right talent to implement and leverage the technology effectively. He notes that this talent could come in the form of contractors or federal employees.

The DHS AI Corps plays a crucial role in addressing this, says Kraft, and he believes it will become the largest dedicated AI team in the federal government. Then, he highlights the challenge of workforce readiness for AI, particularly GenAI, while ensuring a clear understanding of what the technology is or is not.

To address this, Kraft’s team has been holding educational sessions with different groups. In these sessions, the team not only introduces DHS-approved commercially available tools such as ChatGPT, Claude, Copilot, Bing Chat, Grammarly, and others but also focuses on practical use.

The real challenge, according to Kraft, lies in translating that the tools can be used in a limited way that is safe and responsible. Many in the organization, who are not as familiar or comfortable with the technology, require additional guidance, he adds.

Wrapping up, Kraft states that the next challenge revolves around how to use these tools. To address this, the teams are walked through specific use cases. When results don’t meet expectations, the team uses those moments to discuss why it happened and how to address similar situations.

These sessions are aimed at equipping the workforce with the understanding and confidence to leverage AI tools in their day-to-day work, says Kraft. The ultimate goal is to add efficiency to their tasks. In conclusion, Kraft says that while workforce orientation to these tools remains a challenge, the DHS is seeing significant progress and benefits from these efforts.

CDO Magazine appreciates Chris Kraft for sharing his insights with our global community.

I Chose the Public Sector to Be a Part of the Mission — US Department of Homeland Security Deputy CTO, AI/Emerging Technology