HSDF THE PODCAST

Leveraging Artificial Intelligence for Non-Intrusive Inspection - Part 2

• Homeland Security & Defense Forum

Share episode

Senior leaders from CBP and industry experts discuss advancements in non-intrusive inspection (NII) technology, focusing on how AI can enhance the accuracy and efficiency of detecting anomalies in cargo and passenger vehicles at the border. 

National security is at an intersection of technology and policy, with AI reshaping border control and data management strategies. The conversation highlights the critical need for trustworthy data, innovative infrastructure, transparency in AI deployment, and human expertise to effectively navigate the future landscape.

•Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP
•Matt Gilkeson, Chief Technology Officer and Chief Data Officer, TSA (invited)
•Sunil Madhugiri, Chief Technology Officer, CBP
•Peter Guerra, Global Vice President, Data & AI, Oracle
•John Wagner, former DEAC for Field Operations, CBP (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.

John Wagner, former DEAC for Field Operations, CBP (moderator):

Um, anything like the the government should be particularly focused on, or any advice you can give the government or even the industry to look at, like these are areas where things should be focused on or ideas that they should be considering to incorporate as things develop, like I think Sunil just mentioned. You know just some of the rapid changes in technology and the new things that are out there on the edge, the other edge, right what's on the horizon that they might want to be thinking about now?

• Peter Guerra, Global Vice President, Data & AI, Oracle:

Yeah, I appreciate that. I mean, specifically within the AI field, there's maybe two things top of mind and Sunil mentioned that. Actually maybe three, maybe four, all right, we'll see. So, of course, the data is the foundation of all AI and it's really important to understand that the AI's output is only going to be as good as the data and so forth. We all know that intuitively. However, doing it is really really hard, and so for government to invest in the way that Sunil was talking about, absolutely the right path. That's what we see across industries, across other government entities and so forth, moving in the way that he was describing Absolutely.

• Peter Guerra, Global Vice President, Data & AI, Oracle:

We see a lot of movement around data mesh in particular. We see a lot of movement around data mesh in particular, where you leave data where it is but you allow to have the ability to do analytical products that then can be shared broadly across an organization. We see that in the industry. You know we have the largest data mesh in the world processing all the financial transactions that run through our data mesh. It's like $3 trillion a day. That's high speed, like low latency. That type of thing.

• Peter Guerra, Global Vice President, Data & AI, Oracle:

Government focus in moving towards that, I think, is important to enable AI applications. Ai infrastructure is really important, especially as you're thinking about these models. They're getting large and they're complex and expensive, not only to build but also to run, and they're complex and expensive not only to build but also to run. We're deploying data centers and AI infrastructure GPUs like thousands per week right now for industry and government. I have seen, especially maybe more on the defense side, hasn't been as forward-leaning in being in a cohesive strategy in buying and procuring and securing that type of infrastructure, and so I think that's something government needs to be thinking about.

• Peter Guerra, Global Vice President, Data & AI, Oracle:

And then, lastly, what I would just say is, you know, generative AI is super exciting. I actually was at Microsoft before I was at Oracle working with OpenAI as all this was unfolding, and have been a researcher for a long time Super exciting. What I think government should be thinking about specifically, though, is really around transparency. Generative AI within a government context is, you know, all of these models are built by commercial companies, and they put filters. They have specific data sets that they use, not always known or published what those are, and so we had several POCs that we've been doing in the defense area where the generative AI had filters on it, which prevented us from getting what we expected was kind of the right answer, and it wasn't published, it wasn't known. So really, we're doing a lot of work, we're investing a lot of work in our labs and we're partnered with meta and cohere and some of these large language models that you know.

John Wagner, former DEAC for Field Operations, CBP (moderator):

They are focused on providing that transparency to government, but I think that's a really important, important, key thing that often isn't thought about when they're thinking about Gen AI deployments in particular Interesting so, Diane, thinking about all the investments CBP is making in the border drive-through NII systems, up to 100% of the scans coming in, getting good quality data from the industry that you can target and analyze and pick out what needs more scrutiny, what doesn't all the biometric work you're doing with pedestrians, with the vehicle systems, Fast forwarding 5, 10, 15, 20 years. What does that border of the future look like? You know, how will people and companies experience this? But also, what do you expect? How do you expect the risks to change? So the risks you're addressing now with the NII program and the scanning, some of those risks are going to change. Once that capability is there to address them, how do you monitor and react and shift to where those threats are going to go? The drugs are going to continue to come in. Maybe they go other places.

• Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP:

now, no, absolutely, and that's mentioning earlier about the investments have to be across a number of different disciplines and the intelligence infrastructure that we have to build out to support the frontline personnel. It's the data that we have, but it's also the intelligence that feeds that decision making and making sure that we have the right tools to get that data and that intel really integrated and building the systems, working with OIT, but also things like threat briefings, right, but we need the people, the intelligence research specialists, those type of investments that we need to continue to be focused on. I think, in terms of the newer technology that we're looking at, when those threats move or adversaries look, they're moving faster than us, you know, when it comes to seeing what we're putting up in terms of the enforcement posture, right, and we always have to have that redundancy there to make sure that we're kind of looking across the entire landscape at all times. But what else is out there? Right? You know that, those concealment methods, you know, and NII is again one of many tools that we have the AI, one of many tools in the toolbox, but some of the newer things that we're looking at, excited about the diffraction technology, and I do not have the technical expertise to really explain what it is.

• Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP:

But I know that it's really looking at like material structure and the composition of things with the technology. And it's exciting because we have resources like seized property warehouses where we have lots of materials to work with industry to help build libraries right, to identify things like synthetics, precursors, methamphetamine or cocaine, heroin. How do you find this right? And partnering with industry when that type of technology is starting to be developed? We're very much interested in doing that already working some of the details, but as these new technologies really start to develop I think biometrics you were right at the front line with trying to gain the advocacy for facial biometrics and why it was important. And I say and this is not tied to any particular administration looking back across the last 27 years of my career it's always great when you figure out a way to do something and there isn't a policy or oversight that exists yet to be able to roll it out.

• Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP:

You can get away with it for so long before they catch up when you can be in the position to write the policy and have the success and the advocacy before the policy and the oversight kicks in. But you have to do it right. You have to be judicious about it, because if you do it wrong and the oversight kicks in, it's a disaster. Right, but being at the forefront of that and really that's where government and industry have to be really focused on those key areas. I'm not advocating for running out and just moving away, but AI, very much the same. There's a lot of concern Facial biometrics, with the privacy, making sure that those are the cornerstones of the programs, that we have those protections and security, the data security, built in from the beginning. But there's a lot of great stuff that we can really dive into and write the policy about it.

John Wagner, former DEAC for Field Operations, CBP (moderator):

Yeah, that's a great point. I mean the government's in a unique position where they can develop a lot of these things and they can push the envelope on a lot of new ideas and implementation of it in a deliberative, structured fashion. And, like you mentioned, you have the things to build libraries. You have stockpiles of seized property, narcotics, different things that you can build these types of libraries against, to be able to test it.

• Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP:

Yeah, 127,000 pounds last year, Right Wow.

John Wagner, former DEAC for Field Operations, CBP (moderator):

Do you see, at some point with the land border, you know, have you put out any types of timelines as far as you know what percentage of cars and trucks will be scanned or how many pieces of equipment will be out there, and then like after the southern border, what's next?

• Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP:

Yeah, we have about 158 systems to deploy. We're about halfway through. Just under 80 deployed between MEPs and LEPs. But I mentioned earlier with the passenger vehicles right, it's you know, going from 2% to 40%, which is like 29 million scans, and I'm bad at math on the fly here, but with the commercial vehicles it's, you know, roughly like 15% to at least 70%. The goal is 100%, but we have to figure out some things with Mexico, maybe with non-intrusive technology in Mexico, to get to the 100% just because of the physical footprints.

John Wagner, former DEAC for Field Operations, CBP (moderator):

Right. Some of your ports are right on the border.

• Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP:

You're in the second car, you're in Mexico, right.

• Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP:

Like the lines look right here so figuring out those. But you know the vision you mentioned before that I kind of failed to answer. Right, it's that at speed. Right, that's the goal. It's to get the facial biometric. It's to get the data well in advance, you know. And we have some other things that we have to figure out for advanced passenger information across our land borders. Right, we have mobile applications.

• Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP:

When you submit mobile applications, we've learned a lot from our air industry partners on communicating, you know, with the traveling public. Why can't we do that in those spaces? Get biometrics in advance. We already have a lot of data for the crossing. So, at speed and really taking our officers away from do we have to talk to every single individual that crosses? We have to physically look at every single conveyance that crosses? No, we can automate a lot of that work and really focus on the threats and our frontline personnel, our officers, our ag specialists, you know, resourcing them to do what they do best, which is talk to people, the ones who we need to the threats and the unknowns.

John Wagner, former DEAC for Field Operations, CBP (moderator):

Right and, like you mentioned earlier, like they take the Snoop technology, they tear it apart, they find different ways to use it like we would never even think of, and it's really it's just the enthusiasm that they bring to that and, you see, you go out to the ports and you'd see how they've locally modified it to fit whatever you know needs that they need, based on logistics and other concerns. But with our closing few minutes let's open it up for any questions or comments, that we've got one right here.

• Sunil Madhugiri, Chief Technology Officer, CBP:

Here comes a microphone. Thank you, phil Hassel with LMI.

Audience:

You've talked a lot about the massive undertaking you need to deploy in AI around just the pure construction and physical infrastructure, but also the technology supported as well. Can you talk a little bit about the challenges you're looking at and foreseeing, as NII is being implemented, around the logistics tail of the sustainment?

• Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP:

Around, I'm sorry.

Audience:

The logistics of the sustainment and the O&M side of it.

• Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP:

I think you know, in terms of the technology and this is where kind of you know we talk about, like the open architecture, right, and not putting all our eggs in one basket but the long-term vision for us, I mean, we're looking at it with radiation portal monitors. Now Our plan can't just be to, you know, keep replacing panels in the radiation portal monitors right, there's better technology that's out there today. And I think I'm not going to say that we're not talking about the long term for the NII. Today we're just trying to get it in the ground and get all of the data integration done. But today we need to be looking at what that new technology is. You know, and I mentioned the diffraction technology, there's muon technology. That's out there. There's going to be continuing development and is it always going to be an MEP or an LEP? You know what is that future vision? But, again, that's where we're going to be overly reliant, frankly, on the industry to help us get to the next piece.

• Diane Sabatino, Acting Executive Assistant Commissioner, Office of Field Operations, CBP:

I think, in terms of the budget cycles, look, it is frankly, and I'll say it, too long for the technology to be deployed that we received the funding for back in 2019. But, frankly, with all of the different entities. Everyone was skin in the game. There's, you know, probably hundreds of different touch points when it comes to getting this technology in the ground, but what I will say is that a lot of lessons learned, that moving forward and looking at what that long-term vision is, you know we have some clear paths on how we're going to get there Right and mitigating. I think also, you know, in terms of funding, good stories and understanding who the advocacy groups are within USG and outside of USG, that we really need to touch early to get those you know deployments and find out what's next.

John Wagner, former DEAC for Field Operations, CBP (moderator):

Mr Ragland.

Audience:

Just a follow-on question. Thread back to Sunil Open, open, open. Here's where we tie it together. Things are moving in the AR world, rapidly bringing it home. Today we're listening to the plethora of comments coming out of New Jersey because of the drone observations and tracking. Yesterday in a CBP conversation it came up in the same way. But there's a parallel Open, open, open is where we need to go, but inside right now. So open, open, open is where we need to go, but inside right now. We have some proprietary system where the filters are being applied to the data and it's not getting back out to the operators analyzed, inferenced. As quickly as we can. We need to focus on that. One of the observations from yesterday's CBP conversation was look, look at where the leading bleeding edge information is coming from right now, and around the room it was the Ukraine, because they're dealing with the counter drone and the drone threat in real time. We didn't have anybody in the conversation yesterday who was linked to them because of the existing filters.

Audience:

So Sunil back to you Any insights on how we can accelerate the openness for the data, because the operators are starved.

• Sunil Madhugiri, Chief Technology Officer, CBP:

Right, it's an age-old question. Right, it's that technology keeps changing, right, changing, right. You know, I've been doing this for a long time. Tech has changed over the years. Right, and open, open standard it's a. It's a.

• Sunil Madhugiri, Chief Technology Officer, CBP:

It's a tough area because there's certain open standards which we know, right. Tcp ip, like you could think about your networking protocol, right, think if you don't have TCP IP, that's an open standard because there are certain things built on top of that, right. And there are other technology areas where, if you're moving data for example, linkedin came up with a concept called Kafka. Kafka is an open, you know, real-time architecture, sending data from point A to point B, for example, right, but legacy systems, as you mentioned, probably are not tuned for that kind of things. So what happens is that in a lot of cases, a lot of cases, the existing systems. You have to start working with that and start building some architecture on top of that which kind of takes advantage of these open systems. So ultimately, those systems are going to get replaced, right. So we are hoping that when you get replaced, replaced, the next system you're building and the industry you are coming to us, talking to us, are also working on these open standards, right and that that's. That's the age, age, old, chance that that doesn't go away right 50 years from now. There'll be something else. Something else you have to worry about open standards, but as of right now, we are very confident. From operating system onwards, upwards I know the whole stack for a long time Things are getting more and more open in general.

• Sunil Madhugiri, Chief Technology Officer, CBP:

I'll give you a good example. The good example I would say is the large language model, foundation model we are working on. These companies are working in such a way you know what CBP has done is we approved three models right now. We approved LAMA, which came from Meta, came from Meta right. Second one is Mistral and the third one is Gemma. These are all open source models. What does that mean? We are hoping not hoping, I know for a fact our developers will not be touching anything else, otherwise they are in violation of software development within CVP. That's one.

• Sunil Madhugiri, Chief Technology Officer, CBP:

Second thing is when we work with you guys and knowing our open standards are, when you come and talk to us, you need to understand that this is what open standards we are using as of right now, because this is a green left field, new stuff we are working on right now, existing systems. We are hoping that this is what other thing we have done. So we are putting what is called in between, like some kind of in between architecture which can emerge of those two things. Let's say, we haven't we talked about Oracle databases and things like that? How do we take those databases, which is like a rag model database? How do you take that marry with LLM, for example? Right? So we have a concept called agentic functions. We are creating functions. Those functions are the thing we just talked about. It can bridge that gap between the two. Does that make sense?

• Peter Guerra, Global Vice President, Data & AI, Oracle:

Can I add on to that real quick? You know, data sharing always an issue, right? Even if it's a corporation or government, et cetera. Where we're seeing government move to is more of a data mesh type architecture, like I talked about earlier, that Sunil is referring to as well. That helps enable at least when you know there's a filter. It allows you to share data in a different way, without necessarily having to move it, and also to share analytical products.

• Peter Guerra, Global Vice President, Data & AI, Oracle:

So we see that, like if you look at the Army, they publish UDRA, which is the way that they're thinking about all of the data they collect from mission to back office, to everything. How do they mesh that together and leave it in place? But also, from a technology perspective, allow the sharing more rapidly. And it's an event-driven architecture, meaning if there is a piece of data that somebody else needs to see, it's an event driven architecture, meaning if there is a piece of data that somebody else needs to see, it's driven by the event so that it gets published to the person that needs to see it. So it's things like that, those types of that's a technology problem. It's always a cultural problem, not a technology problem. But from a technology standpoint, that's what we're seeing more and more adoption of.

John Wagner, former DEAC for Field Operations, CBP (moderator):

Okay, we've got time for one quick question. Your hand went up first.

Audience:

Hey, I'm Thomas Anthony. I'm the founder of Analytical AI. We are in the NII program, and so the question was for EAC and for Sunil. So for us to build algorithms, we need that data. On the TSA side, we build a lot of algorithms for TSA. We go to their labs, have guns, knife shops and bags, collect that data. Is there a plan on the CBP side, Because we can't create?

Audience:

that data we need to get seizure data from CBP. Is there a plan to create that data? And so Nile said annotation is expensive, it's difficult, so if it can be collected in a controlled environment, it becomes really great for industry to build those algorithms for you guys. So is there a plan to go that route?

• Sunil Madhugiri, Chief Technology Officer, CBP:

I'm glad you asked. I forgot to mention that. So, yes, there is a plan and you're 100% correct, right? So what we are doing right now is that we are in the process of starting to create what is called creating synthetic data, and we are working. I can't name any companies, I'm not supposed to, I'm not, but we have what's called an iLab within CTO office and we are in the process of working with over four teams to create a synthetic data. 100,000 images are being created as we speak and you'll see some of the things that we just talked about. And again, as I mentioned, that is the hardest part, as you know. Since you are in the NIR space with us, you know how tough that is. So I'm hoping, the synthetic data, what we are creating Another thing I'm going to do here is we're going to do here, as CBP, is marry that with a computer vision concept, with LLM component, because what is happening is these LLMs can create synthetic data also.

• Sunil Madhugiri, Chief Technology Officer, CBP:

Now, you see, I'm saying I don't need a specialized company to do this anymore. We can actually do llamas of the world, three companies we have chosen for their models, for open source models they can start doing these things now. So that's why what will happen is the integration piece should be faster, because I can marry that piece with the NII. Your custom-based model, which you guys are building Make sense. So that's where we are. Just hang on If you talk to me offline.

John Wagner, former DEAC for Field Operations, CBP (moderator):

I can give you more information. Good Well, thanks everyone for the time. Thank Diane, sunil and Peter for your time as well too. I know you have very busy days, thank you.