HSDF THE PODCAST

Future Ready Insights into CBP's Emerging Requirements - Part 2

• Homeland Security & Defense Forum

Share episode

Preparing for the future, CBP is prioritizing emerging requirements such as advanced data analytics, AI-driven decision-making, and resilient infrastructure to address dynamic security challenges. 

 This episode delves into the intricate relationship between AI, data management, and national security, highlighting challenges and potential solutions for improving operations at CBP. The discussion emphasizes the importance of data cleanliness, accuracy, and innovative partnerships for effective AI integration. 

  • Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP
  • Jeremy Ocheltree, Director, Innovation Team, Office of the Commissioner, CBP 
  • Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP 
  • Ed Hicks, Business Development Manager for Federal and Artificial Intelligence, Dell Technologies
  • Justin Doubleday, Federal News Network (moderator)

 This discussion took place at the HSDF’s Border Security Symposium on December 11th, 2024 

Follow HSDF THE PODCAST and never miss latest insider talk on government technology, innovation, and security. Visit the HSDF YouTube channel to view hours of insightful policy discussion. For more information about the Homeland Security & Defense Forum (HSDF), visit hsdf.org.

• Justin Doubleday, Federal News Network (moderator):

Any other thoughts on, just on, data management interoperability?

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

I mean one. It's less of a thought, more of a challenge. I think you can all help. I think everybody in this room understands that they have data problems. Everybody understands that there's a lack of data cleanliness. It's not curated and optimized for the use case you're trying to apply it to. There's a host of these other problems, but you're all building AI solutions and.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

I fully suspect that there's going to be a solution out there that can help me clean my data, that can help me curate it into a fashion and format in which I need to use it to apply it to, let's just say, non-interested inspections. I'm trying to talk about anomaly detection and non-interested inspections, and I need to cull through a massive amount of data that I have to identify the types of data I want to use to apply, to refine the model AI can help me do those things too.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

I think we need to also look at ways in which we can optimize the challenges that we have, and then, we have people like Ryan, on the other hand, on the side that can say this is the storyline I want to see the story told, say this is the storyline, I want to see the story told. Are there solutions out there that can help me get to the editor faster? Because I know that doing it manually is a challenge we've all seen and we've all sat there and sat through for years and we're still talking about it. So are there solutions out there that can help us speed these processes along? And I'm fully supportive of piloting both solutions to help get me to the river and then solutions that are the boat to get me down.

• Ed Hicks, Business Development Manager for Federal and Artificial Intelligence, Dell Technologies:

One more, if I can add on that as well. I mean, what we are starting to especially leverage in some of the AI work that we're doing is to basically leverage the AI to basically self-label. So we'll use technologies such as federated transfer learning, where I'm actually learning at one location where I have a lot of information and I can self-label or have that information self-label at a different location, and we're seeing better accuracy with that than if you're actually having a program that goes out and actually does that or tries to do that curation. So what you're looking at is by leveraging, you know, natural language processing, by leveraging the ability to have that system that can be looking into something where there's a lot more data and provide that insight to a much smaller footprint. We can then utilize that to label and we're starting to do those kinds of things. So we're seeing that.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

And then I only need to leverage humans at the end of that to comb through and determine accuracy of a smaller subset of data. So we've reduced the amount of time we have to have agents and officers away from their primary responsibilities looking at these data sets when they're helping me define accuracy of a smaller percentage. I think there's just an incalculable number of gains we can make throughout our roles and administrative problems and documentation problems and tasking responses that we provide to the various people. We have those types of things where we're only having humans review and verify and validate accuracy after we've gone through kind of a cleaning round. There's just a lot that can be done.

• Justin Doubleday, Federal News Network (moderator):

And in terms of the accuracy of these models. I mean there's a lot of baggage that AI comes with in terms of concerns around accuracy, bias, privacy. Does CBP have a framework for approaching that? What should folks know about how they need to kind of gate their solutions right?

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

Absolutely so. Just as long as Ryan's kind of talked about the history of technology, we've been using artificial intelligence at CBP for all over a decade. Machine learning models started in some of our earliest use cases. We've been using it for facial recognition for our travel verification service for some time.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

There's a significant amount of testing that goes on in the background before any of those models approach operation. But I think, when we start talking about foundation models, engineering, ai models and models of that scale, the thing we're going to rely on is an industry partnership here that says I'm going to need to see behind the curtain and know what Oz is doing here, because there's some reporting that we're going to have to be able to provide that we have to work with industry on so that we can ensure, one, the American public, but two, the officers and agents who are going to rely on this information to do their job that they can trust the information presented to them as exactly as possible. We are concerned about. We're very concerned about bias.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

We're very concerned about introducing bias to the data itself and introducing inaccuracy. But again, a lot of this is a partnership that's going to have to be built together.

• Justin Doubleday, Federal News Network (moderator):

NIST has a lot of new data standards out. There's a lot of new ISO standards around these models.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

I think they're all going to be very valuable in the long run. But I think right now we're at the point of saying I have a data standard, now we need to get together to say, okay, we've found a way to make sure everything's meeting these, and that's a road we're probably going to have to walk together. I think, when it comes to artificial intelligence, there are going to be very few, if any, like true COTS solutions. Let's say this is a commercial off the shelf, because nothing is off the shelf when we're talking about integrating it into the government network, into the government enterprise. There's going to have to be cleaning done. There's going to have to be integrations done.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

There's going to be work on both our side and the vendor side.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

So I think a lot of those things introduce new challenges that need to be both tested around and iterated around. So, again, I think the point where I'm going with that is it's a partnership. Now, Like before, and amongst all other things, we're moving towards an avenue of direct like it's going to be, you know business-government partnerships that lead us to this road of success, I tell you on the privacy front, incredibly important to us at CBP, right so, as it relates to AI and, well, any type of technology or system really.

• Jeremy Ocheltree, Director, Innovation Team, Office of the Commissioner, CBP :

So, just so folks understand the process behind that, I mean, clearly there's federal privacy laws, and we're obviously lockstep following those laws and then supplement that with internal policies and NSOPs, et cetera. But in advance of deploying any new technology, we do what's called a privacy threshold assessment. We document exactly what are the essential elements of information that we're going to be collecting here, how is it going to be used, who sees it, so on and so forth. Have that engagement with CBP privacy. If it's needed, that'll get elevated to the department, but there's this kind of workflow that happens where we make sure we're on the righteous side of the privacy laws in the United States. So that's a core element of every single project that my team does as bleeding edge innovation with industry. We have those conversations frequently and if we need to bring the lawyers, we bring the lawyers. But, yeah, privacy is very, very important. It takes forever, though, to do Like tech transition.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

ATO.

Audience:

I mean.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

What we haven't talked about is using AI to improve our internal business processes, and we heard things today like speed of the mission and privacy is important. We're not going to take shortcuts on PTAs, but what eats my lunch is we have these prototypes using non-proprietary, open standard, open architecture, unlimited government rights technology to prove that you can automate machine reasoning, which is structuring data in a graph format to create good data products for machine learning. But when then we want to actually test it? On data, even internally, we have to go through many steps to verify that we have a law enforcement purpose, a need to know, and that all our data is unclassified. But we still have to be deliberate in its usage and handling instructions, and that could be rapidly accelerated, so we don't have to spend months and months on making sure that we're validating. You can have the tools to validate and test that the machine is going to be more accurate and make less errors than people, as long as the people coded it right. And when Josh mentioned standard, the one standard ISO 23818-1 and-2, is a standard for top-level ontologies and basic formal ontology that has been endorsed in different agencies of the government, and I hate standards. No one wants to hear another standard. But this standard allows you to be loosely coupled and make interoperable your existing data standards and formats. That's the key I would throw out to industry. And I know I said ontology. I said I promised I wouldn't, so I owe someone dinner now, but probably a lot of people.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

Applied ontology classic ontology is from philosophy how do humans reason about the nature of the universe? Applied ontology is how do you make that computable? And that really took off in the early 2000s in bioinformatics because they had massive amounts of data and they wanted to code the human genome. Pharmacy companies wanted to be first to develop new drugs. So now, 20 years later, there's billions invested in open source ontologies based on that standard that you can reuse and extend from. And so I mean I'm not, you know, silicon Valley venture company. We're a ragtag bunch of bottom up folks that use budget dust and most of the funding comes from charity to prove these projects. But that's the piece that fits, I think, is going to that standard.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

When you want your data to be interoperable, look at using basic formal ontology as the baseline for your data models. You don't have to conform and change your words. You have to map what your words mean, to that international standard and that's what we found has made progress. And the last point you know applied ontology. Who cares? Well, we believe in it so much. There's nine people in PhD programs studying this stuff now. Six of them are Border Patrol agents and there's 23 in the program overall getting PhDs in applied ontology for artificial intelligence. Maybe the demand signal there is we need to look at training more people and offering that. Maybe CBP needs PhD-level support. And I'll tell you that information is actually free. I'm not going to endorse it here, but you can take these courses for free and you only have to pay if you want the degree. You can get the same training in your own industry and companies If you find that right person that has the aptitude and willingness to learn. You just need like one. You don't need a lot of this.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

No one ever has to know or hear about ontology again there you have it All right.

• Justin Doubleday, Federal News Network (moderator):

All right, folks, we got about 10 minutes left. We're going to open it up to questions from the audience anyone. I've got more questions, but yeah, people need more coffee. Um, all right, I think. One thing we also just wanted to explore and we've touched on this a little bit um you with anomaly detection. You just mentioned business process engineering, but are there other use cases for this really broad umbrella? We're talking about AI, machine learning, autonomous systems that you think folks here should know about that are coming out, that you're working on.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

Oh yeah, I could talk about this part all day. So I think the reality is CBP is 60,000 people spread out across four operational components and two business components.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

Right, we have challenges in the Office of Trade that are numerous, in both data-relevant systems and then systems in the trade side, that we need to understand what the supply chain looks like. How do we better manage the supply chain resiliency coming into the United States and how our products are looking when they leave or when they're exported from the United States? I think we have some significant challenges in tracking things that are being tied directly to forced labor. But then I think you pivot that and now we start talking about what does a supply chain bill of materials start to look like Now? Do we have to wonder about what the code, the code base, starts to look like when you're outsourcing other country coders in development? Do I need to ensure that isn't impacting the models and the software deliverables that are being provided back to the United States government?

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

If we look over the office field operations, we see significant challenges in AI required to help us in non-intrusive inspection. Then we also see challenges that are hardware related that everything is going to have an IT Nexus hardware with an AI component, but we have smaller footprints in ports of entry and limited time to make decisions, so in some ports of entry they were developed. Later they were built with larger scale, we can actually install technologies to support the mission. But in some of our older, most thoroughly traveled ports of entry in the United States, there's no gap between the country of Mexico and when you enter, primary inspection. So we need technologies that can help us better inform decisions in that window of time that is almost non-existent and in that space right there and lack of footprint, you tell me.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

What information can I derive, as somebody's pulling up to enter the United States for me to determine whether or not they're bringing in dangerous narcotics, whether they're bringing in something that you know $1 million of fentanyl could take the lives of hundreds of United States citizens how do I start making those decisions faster? And not, you know, not even talking about AI here. What hardware solution? What can I look at? At the undercarriage of a vehicle that might point me in the direction that we need to have a further discussion. The border patrol always the same problems there's a lot of land, there's not a lot of connectivity, there's almost no power, and so we have to get solutions out there that can kind of impact that challenge across the board.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

And then air and marine operations the same thing but in the sky and on the water, and then air and marine operations the same thing, but in the sky and on the water. I think we're here to talk about any of those, but the individual components themselves will be able to walk you through day-to-day problems. And what I'm trying to look at right now is are there areas within artificial intelligence that we can help them, kind of triage, some of the problems and things that we can automate to get them back to doing that primary mission and that's going to be a challenge of this whole year is to say what are the biggest things that we can solve through ai. Because we don't know yet. Because the technology's candidly so good that it's reading my daughter, helping me read my daughter's new bedtime stories every night, because I no longer have to make them up, I can say, hey, ai, give me a good idea for a bedtime story.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

And I don't have to answer.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

why is the sky blue anymore? We can have a conversation and learn and talk about it. So if I can do that at home, think about what people are just now realizing at work.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

So people threaten me with hard questions and I just want to say I don't see anybody standing up.

• Justin Doubleday, Federal News Network (moderator):

So come on, oh, right here asking you guys.

Audience:

I got a mic coming to you from the back right. I have to say I think the crowd will agree with me when I say you've been one of the more entertaining panelists today.

• Ed Hicks, Business Development Manager for Federal and Artificial Intelligence, Dell Technologies:

You really brought sort of a Dennis Miller on Monday Night.

Audience:

Football aspect to this, If I can use your words, when you look at the opportunity around generative AI and application with CBP where is that opportunity for? The budget dust that pays for the wheels in the suitcase.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

What's that low-hanging fruit? Semantic technology and knowledge graphs. So you can hear the LLMs that's common right? Well, the Gartner impact radar for the last two years had. Impact radar means it's in the middle of the five ring, as we call it. If you're shooting at a target, it's here now and it's impactful. Llms right near the middle.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

The other thing that was in the middle was knowledge graphs, and many of you probably have never heard of knowledge graphs and what that means. So that is the semantic technology how you code, using ISO standards, your data unambiguously and precisely and your data models when you need it and underpinning your large language models with knowledge graphs. So RAG, or achievable augmented generation, goes get you there, but GraphRag having a knowledge graph tied to an international standard, underpinned by first-order logic which is computable, to then test empirically if what you ask, you get the answers they're correct and you can verify that. Explainable, traceable that's the opportunity and the cost of that is thinking. These things are hard to talk about and hard to explain. The cost of doing it is professional services. A couple people, a couple modelers can start working through the data sets I mean. Now, if anybody's tracking.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

Office of Director of National Intelligence released two weeks ago, their intelligence community data reference architecture and right in the middle it says you need your enterprise knowledge graph, you need your enterprise ontologies and then, when you have your data domains and your data orchestration or application layer, you reference that semantic layer, that information layer, to make sure that when you need to be accurate and precise and you're reasoning over complex problems and not all AI needs to do that, because we've been using AI for a decade in traditional machine learning models which can maybe inference or infer from smaller data samples into a greater population this is inferencing over a graph, where you're making connections that may not be explicit to analysts because the problem is too complex. There's too many different data sets across too many different organizations, with a complex adversary exploiting things like the minimus, where we know the data is there, but it's hard to make those connections. Implementing it it's hard to talk about and think you only need a few people to do it. That's the cost, but that's not popular.

Audience:

It's easier to say.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

LLMs, Even though we're running some LLMs in the FIBRING boom knowledge graphs, but you haven't heard probably much about that.

• Justin Doubleday, Federal News Network (moderator):

A few more minutes for questions here, if anyone has them. All right, and Jeremy, I wanted to actually turn to you and just ask about I know that Invent has a number of pilot projects that they're working on at any one time Any recent successes in terms of transitions, Any new ones that you would want to highlight for folks that are kind of just getting off the ground?

• Jeremy Ocheltree, Director, Innovation Team, Office of the Commissioner, CBP :

We do. I do have one or two comments. Oh, sure, sure, if I may, a couple, maybe just kind of forward-looking comments.

• Jeremy Ocheltree, Director, Innovation Team, Office of the Commissioner, CBP :

Josh did a really great job, I think, summarizing some of the needs of the three uniform components in trade, which is sometimes often overlooked but incredibly important as it relates to the economic prosperity of our country. The trade mission is really really important to CDP. I think of technology as being an extension of the senses for our agents and officers and people in the field doing work right. I think we've made significant progress, specifically in the Border Patrol over the last few years with the Team Awareness Kit, issuing all of our agents cell phones so you have some computational power with you and the ability to get blue and red and a greater understanding of what's happening around you. So again, it can inform the decision making and makes our agents safer and more effective Great first step. I know OFO wants to do things that are similar to that Our folks that are working in the POEs are currently. You can't imagine the volume, the incessant flow of traffic and pedestrians consistently coming up and I'm consistently having to bang on keyboards, logging in systems to retrieve information, to identify whether or not I need to continue to, you know, engage with this person or send them off the road. I know the vision there is, I think, somewhat similar to what we've been able to accomplish with the Border Patrol is to get that person away from that computer so that I can now use my actual senses to take a look at what's happening around me. Right, there are people in those cars that have micro-expressions. There's things happening that can be perceived. There's disclaimers in those cars, people are trying to, you know, throw you off. You miss it all if you're staring at a keyboard, right? So that's I think.

• Jeremy Ocheltree, Director, Innovation Team, Office of the Commissioner, CBP :

One other challenge maybe we could put out to industry here is that we need more capabilities that are going to get our agents and officers moving around and being human being, law enforcement officers and executing their mission. And that's just one example. I'll stay with OFO. I know the vision from some of their leadership in the airport environment. Airport modernization is important to Miss Sabatino right now. You know how can we continue to use biometrics? There's some really interesting pilots that they're doing. I'm not claiming credit for this right now, but with the global entry population facial biometrics you just walk through and it knows, based on your initial submission for your global entry package. It knows and confirms who you are and recognizes that we need to talk to you or we don't. So this more seamless travel experience is really important to OFO.

• Ed Hicks, Business Development Manager for Federal and Artificial Intelligence, Dell Technologies:

Yeah, thank you, pilots, I was going to add one last piece I was going to say. A lot of what we're seeing in the airports, especially from the airport perspective, for TSA to your point is leveraging different technologies such as like LIDAR, where we've got a, instead of a two-dimensional object of a video, we're actually looking at a four-dimensional object of a three-dimensional object moving through time, space and across multiple cameras, etc. And so what really does happen in that case is, as you pointed out, as you're walking in through the airport, as you're walking through customs, we're able to leverage technology that's basically privacy preserving, because it's really you're on a unique identifier. We're able to go ahead and bring you into that environment in a secure fashion and be able to maintain your privacy, but also be able to create a unique identifier based on behavioral analytics as well, based on that individual. So there's a lot of technologies that are coming down the pipe, that are actually some things that we would be able to leverage in border patrol, that are being used today in airports.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

I would say that would be a new system of record and a new biometric system of record. So there would be a ton of testing that would need to go involved in anything about that.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

I'm just going to put that out there to be in with, because creating a new biometric tag to an individual, such as facial recognition we're talking about that through LIDAR is not something we're currently doing and I would say, prior to us doing anything like that, I just went out in the open like a lot of testing for veracity and accuracy for anything related to that.

• Ed Hicks, Business Development Manager for Federal and Artificial Intelligence, Dell Technologies:

And a lot of that is really being driven by the industry and that's why you know, like I said, it's not, you're absolutely correct. The biometric aspect of it could potentially be.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

I have one short alibi. Alibi of the bore patrol is when you're on the firing range you have a misfire, you get to shoot again. Earlier I said something like that. We're using non-proprietary, open source, open standard. I don't want industry to think that doesn't mean not profitable, because it's in the technical approach of your application of those types of technologies and your expertise that I think is the opportunity here. I just wanted to make that point, thank you.

• Jeremy Ocheltree, Director, Innovation Team, Office of the Commissioner, CBP :

We've discussed this in the past. It's okay to have a black box when the information going into it and the information coming out of it needs to be something that we can meaningfully integrate.

• Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP :

I think that's what you're saying, that's right, Don't need to know the algorithm, I just need to know the data schema so we can have an ontology to map your schema to other schemas and make them interoperable, machine readable in a language agnostic way that's precise and ambiguous, computable and improvable logically.

• Justin Doubleday, Federal News Network (moderator):

I think we could go for a lot longer, but we're getting the plug so thank you again.

• Joshua Powell, Deputy Chief AI Officer, Office of the Commissioner, CBP:

It's like a seal at the zoo like what did that guy say?

Audience:

Thank you Thanks.