AI tools are helping CBP, Commerce hasten IT modernization

Artificial intelligence tools are not new for the Customs and Border Protection directorate in the Department of Homeland Security.

CBP has been using these capabilities for things like facial recognition and entity resolution for years.

Thomas Mills, the executive director of the Cargo Systems Program Directorate at CBP, said the agency is now applying AI to modernize legacy code and seeing significant improvements.

“We have seen right about a 40% acceleration in in our ability to be able to move legacy components into modernized components, ranging from stored procedures and databases to parsing code on inbound messages. It’s a lot of stuff that’s been written in languages that I may know but others don’t, like C and C++. So that’s a big deal,” Mills said during a panel at the recent ATARC AI Summit, an excerpt of which played on Ask the CIO. “We’ve been using a lot for automated data extraction. I think one of the themes that you’ll get even with AI, the whole garbage in, garbage out principle is still relevant. So, if you’re using large language models (LLMs) against bad data, you will get bad results. So we are using AI to help in our data extraction, embed a lot in our data governance to clean up the data, so that when we use AI, it’s based off of a data set that’s a lot more true.”

]]>

The Automated Commercial Environment is CBP’s long-time major system modernization effort where AI is helping to accelerate development of new capabilities.

Mills said CBP developed several of the underlying infrastructure supporting ACE in the early 2000s. Though, he said, the agency recently moved off and retired its mainframe from the 1980s.

“Our original estimates around modernizing that stuff was a five to seven-year projection. The components that we thought would take a long time, we’ve seen a 40% compression in the timeline, so code generation is real,” he said. “Now I caveat it because you do have to have people that know what they’re doing taking a look at the results. We don’t just blindly trust what code gen puts out, but it does give you a lot of insight into what things like stored procedures in a database are doing, and gives you a lot more sense of where things are. It has resulted in us being able to accelerate some of our modernization timelines.”

Iterative development to predict costs

Mills added that for many of its software development projects, CBP is setting a two-week sprint to deliver capabilities to users.

“When we look at fielding capabilities incrementally, we look at the holistic cost of that, so we can actually build some estimation. Because the first time we did it, it’s like ‘holy cow, what are we actually paying for here?’ in the cloud. But mapping our capability development into small iterative chunks to build really does help us predict what we think that cost will be,” he said. “I think that’s been the key for us, sticking with small iterative development and then making the case to say, ‘hey you know here’s what we did, here’s what we found out through this use case,’ and then advocating for the funding. If we don’t get it, then obviously we can’t expand the way we want to. I think the key for us has been small chunks, fail fast and then find out ways to scale.”

Thomas Mills (left) is the executive director of the Cargo Systems Program Directorate, in the U.S. Customs and Border Protection in the Homeland Security Department. Matt Passos is the acting chief technology officer for the Department of Commerce and Ryan Palmer, the USAi architecture lead at the General Services Administration. (photo courtesy of ATARC).

At the Commerce Department, bureaus and offices are applying AI for modernization efforts as well as to improve mission and back office functions.

Matt Passos, the acting chief technology officer for Commerce, said the one common goal across all of the agency’s use cases is the data.

]]>

“For us, AI is really helping us make our data smart. We have a lot of smart people and a lot of dumb data, a lot of systems, older systems that we don’t necessarily have the money to modernize. We don’t have always the skill to modernize. Let’s not even start with system integrations. When you modernize a system, it’s not just that system, it’s every single thing that it touches. So, one thing we’re relying heavily on partners, and [the General Services Administration’s] USAi [platform],” Passos said. “They’re probably one of our greatest partners here, and we’re using application programming interfaces (APIs) to give every dumb data system access to AI. Now the folks that are experts in the mission, experts in the data, experts in the system, can have all this all this processing power to go and basically make their data a lot cleaner.”

Once Commerce can clean up and standardize its data, Passos said applying AI to mission and back office functions will be more successful.

He said bureaus, including Census, the National Institute of Standards and Technology and NOAA have systems that are using APIs to access data and run queries through the USAi platform.

“A great example of something we’ve been doing with data is for capital planning purposes, AI is absolutely outstanding. Figuring out the total cost of ownership for things is absolutely something AI is outstanding for,” Passos said. “If you work with the Office of Management and Budget or with GSA, you know you have to do a lot of the capital planning reporting. Some agencies do it just for compliance sake maybe, but we actually use our capital planning very seriously. AI is looking at the capital planning data and making investment decisions on the future, projecting success or failure, and gamifying our investments is something that AI is absolutely outstanding for.”

Reducing the risk of AI vendor lock in

Understanding the cost of AI tools is another key factor in how agencies are applying these tools to IT modernization efforts.

Passos said there still are a lot of questions about how industry will price AI today and whether it will carry over into the future.

“It’s going to impact our cloud costs, but how it’s going to be spread out, I’m not sure. I don’t think it’s a settled business. We are working with all the different vendors. We’re transparent about the fact that if we tell our stakeholders, ‘hey, your cost may spike. Here’s what it may spike with this one versus this one,’ and we’re putting the onus on the business owners to figure out what’s the value to you, which is I think the hardest thing,” he said. “If there’s anything that anyone can help Commerce, or I think the government to figure out, showing quantitative value of AI is a huge challenge, while you’re trying to implement and adopt it.”

Ryan Palmer, the USAi architecture lead at GSA, said addressing cost and vendor lock-in concerns are among the biggest challenges for the shared service.

He said during the panel that agencies need to take advantage of new models as they come out as well as integrate USAi with multiple cloud providers.

]]>

“A lot of times increasingly the models are available for multiple services as well or multiple locations. So being able to not marry an AI model, but rent the AI model that you need at the time, I think is something that is helpful to approach this as an agency or organization,” he said. “Some of the stuff we’re doing with USAi is we’re running evaluations. We’re pulling all the models from any of the leading American AI companies we can get, making those available from FedRAMP authorized platforms, and now agencies can confidently explore and test these solutions, and even integrate with them. They can connect their data sets. I think modernization in this AI era versus past efforts is about using AI to help people unlock data. That data is letting us do like modernization more intelligently. It’s letting us take a legacy system and expand its capabilities, and that doesn’t require a full rewrite of that system.”

Palmer added more than 20 agencies are using USAi today.

GSA is using the platform as well in several areas. Palmer said one example is using an API to expand the capabilities of an existing system that doesn’t require a full rewrite. Another, from a quality standpoint, is a use case around catalog products on the schedules program.

“We have a lot of really simple use cases, and by making USAi available within GSA and also to other partners, we’re seeing very smart people really look at how AI can actually help their job, and we see kind of a curve,” he said. “People start with exploring with the chatbot; they might start using the API on existing data or existing workflows, and then you start seeing that transition into ‘hey I’m going to change my workflow and start building around AI as like a capability.’ That’s where we really see like an evolutionary curve, but also some of that acceleration. I think people are talking about.”