HSDF THE PODCAST

Leveraging Artificial Intelligence for Non-Intrusive Inspection - Part 1

• 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. 

This episode discusses the evolution and future of the Non-Intrusive Inspection program, focusing on its integration with artificial intelligence to enhance border security. It highlights the importance of data analytics, workforce adaptation, and technology in streamlining customs inspections amid increasing volume and complexity.

•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 :

Good morning everyone. John Wagner, retired from Crescent and Water Inspection, joined here by some former colleagues and experts and we're going to talk about the Non-Intrusive In program also known as NII in CBP speak and you know this is a program these have been.

• John Wagner, former DEAC for Field Operations, CBP :

I'll just say in non-technical terms, these are the big x-ray systems that can x-ray cars and trucks. You know all different types of technology, all different form and functions and use cases for these. But you know, throughout my career in customs, you know, know we saw these when I first joined. We thought we were cool because we had an X-ray van. We could put packages through the van and X-ray them on the spot. You know that grew into building these big. You know systems where you could X-ray an entire truck at once.

• John Wagner, former DEAC for Field Operations, CBP :

It looked like a giant car wash and you know it's built from there. It's this program wash, and you know it's built from there, it's this program. And now you've got hundreds of different types of large scale, small scale, all different types of energy and technology to be able to do this and it's really become a cornerstone and a foundation for the inspection process and we want to talk about that today and like where it's going. And you know some of the real challenges you have and some of the exciting possibilities you have about what this will do for the land, border and the enforcement operation. So, diane, we'll start with you, like, can you kind of give us a sense of where the program's at, what it means, and some sense of, like, what's the current things you're most working on? No, thank you, and it's great to see you most working on.

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

No, thank you, and it's great to see you. It doesn't feel like you're retired. We see you all the time.

Sunil Madhugiri, Chief Technology Officer, CBP:

This is a good thing.

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

And happy to be here on the panel with everyone today my favorite USG partner, mr Sunil Madaguri for sure, who really is kind of the driver of all things that we're trying to do with the data integration and the AI with non-intrusive inspection technology. But well underway, about half of the deployments across the southwest border with the MEPs and the LEPs. That deployment is going to go through fiscal year 26. A lot of big construction projects wrapped around all of these kind of smaller construction projects at the ports of entry and at the same time really partnering with the Office of Information Technology to take all of the data, all of these kind of smaller construction projects at the ports of entry and at the same time really partnering with the Office of Information Technology to take all of the data, all of the sensors and tie that information together. And that has really been a lot of the workload for us.

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

In addition to these construction projects, but incorporating the artificial intelligence, machine learning, into it, we have a couple of business use cases for the algorithm deployed already. We're seeing some great results with those algorithms being deployed. But the reality is, you know, in terms of the scale and scope, when we think about, you know, we were scanning about 2% of passenger vehicles previously. We're going to increase that to 40%. That's going to equal about 29 million scans that we're looking at in terms of the volume that we have, so we could not staff our way out of it. That's where the artificial intelligence, machine learning really is going to be critical and building with that human-in-the-loop architecture I know the deputy commissioner mentioned kind of the weaving in that technology, that hardware and that software, into our most important resource, which is our frontline personnel. You know, in the fight against fentanyl, whether it's the agriculture threats that we're trying to tackle, I brought that one up first, right.

• John Wagner, former DEAC for Field Operations, CBP :

Got to always give a plug to agriculture.

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

Absolutely the biological threat issues that we deal with. But you know the priorities that we have. You know very matrix and leveraging the technology, even across the enterprise, and making sure that our partners in CBP and we're leveraging the tools and the emerging tech that they're identifying, incorporating it into ours as well. So a big landscape that we're looking at. I mean, you know it's over a million passengers a day. Just yesterday we had 270,000 vehicles cross our borders. Yesterday 48,000 trucks crossed our borders.

• John Wagner, former DEAC for Field Operations, CBP :

Just yesterday. Just yesterday, one day One day One day.

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

Four million de minimis packages I know the deputy mentioned that as well. The volume is substantial and it's not shrinking anywhere. So leveraging those tools and NII is one of the tools the artificial intelligence, but the intelligence infrastructure that we have to support it A lot of resource commitments, kind of across a number of different disciplines that tie together.

• John Wagner, former DEAC for Field Operations, CBP :

So the, the, the NIA systems today? You know they're used in the secondary environment. You mentioned some of the, the throughput numbers, but there's a CBP officer and they're sitting in the machine looking at the screen and interpreting what the results are and trying to use their training and their historical knowledge and their expertise to say wait, that's an anomaly, that shouldn't be there. To build these drive-through systems and to bring this up to. You mentioned 40% or up to 100% of the trucks and the cargo coming through, like that's a couple of hundred thousand scans a day. Now I don't think you have a couple hundred thousand officers. Since I left I don't think the hiring's been that great. But even with 30 something thousand officers, 25,000 something officers like Sunil, how do you best position now the workforce to take advantage of and interpret the enforcement productivity that's going to come out of a couple of hundred thousand scans a day? And how do you possibly interpret all that? Certainly can't have an officer there looking at the screen and trying to read all that sure, uh, thank you, john, it's a.

Sunil Madhugiri, Chief Technology Officer, CBP:

You know, whenever I come and talk with the panel, with ac sabatino, it's a tough act to follow. As you can can see, she's absolutely unbelievable. I've been working with her for a while. She's a great partner for us and always supporting us with what we do on the tech side. Without them, we don't exist. Right, that's the bottom line. She's absolutely. Thank you for everything you do, ma'am.

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

Thanks for the plug. I'll get you later.

Sunil Madhugiri, Chief Technology Officer, CBP:

No, no, she really is. It's unbelievable. So I think what the difference is with the EAC is that she's forward-looking and giving us opportunity to think for us at what. I'm a techie myself and I grew up in the industry myself, so been in the government for about five years. So to John's, your question right, the word AI technically has been there for a little while. Right, what has changed is some of the tooling around it. If you look at data in general and some of the things what CBP has been doing for a while especially our current commissioner who started the National Targeting Center and some of the things they were building working with the targeting systems for CBP, they were using anomaly detections, various things, for we used to call predictive analytics at that time. That was the former name, right for AI. Now they're kind of converted, because now each and every product people are developing has some intelligence built in there. So you're looking at things to do in a faster, better manner, Especially when you things to do in a faster, better manner, Especially when you mentioned about the images. That is exactly that, because you don't want humans, you don't have 100,000 officers to look at this stuff. So the AI component is playing a role. To summarize things, look at things Like the EAC said office is always in charge.

Sunil Madhugiri, Chief Technology Officer, CBP:

Ai is assist. That's what we say, right, AI is there just for assistance. Office always has the final say. But from our perspective, from the tech perspective, the cool thing what is happening now is that tech has dramatically changed, especially last seven to eight years, and especially now I wish I was 30 years younger. The Gen AI component is probably one of the biggest changes in the tech industry after dot-com. I don't know how many people remember dot-com days, but those days was exciting, right. Some were failures, some were good, right. Those companies have grown and become large companies and some of the startups have done an amazing job. So this is where we are right now, especially with a tech perspective. So how do we take the technology, John, and start applying that to NII component and things like that is what I'm trying to figure that piece out, right, Working with AAC and actually Jody Harden and team on the NII side, A and team on the NII side A lot of exciting stuff we are doing.

Sunil Madhugiri, Chief Technology Officer, CBP:

I'll talk a little bit more as the panel goes through, but we are. You know it's computer vision from looking out what the images look like for anomalies. As ESE said, we're deploying some models already, and then what we're going to start doing is how do we make life easier for the officer to interact with some of these things? Right, they all have a UX component to this. They also have some interactive data, getting information out very quickly for them. That's the part which is exciting now with Gen AI component that marriage between traditional AI, custom-built stuff. What we are doing currently with the Gen AI model is where the future is, and I have some examples I can give you as we go forward, but that, I believe, is going to help dramatically. What we do with NII. Another area, John, I'm going to talk a little bit more also is on the edge computing. That is changing the how we do business and that, I think, will help industry also.

• John Wagner, former DEAC for Field Operations, CBP :

Yeah, we'll come back to that in a minute. I got a few more questions for you too. But, peter, the industry perspective, hearing some of the use cases and some of what CBP is doing and your experience across even different sectors and commonalities, and some differences, some thoughts on where this work is headed.

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

Yeah, no, appreciate it, and I'm going to apologize in advance. I'm on the tail end of a cold, so you get the Barry White voice today.

• John Wagner, former DEAC for Field Operations, CBP :

So yeah, put me at the end. Should I sit?

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

over there. Yeah, sorry about that, but yeah. So I've been at Oracle now, I think, over a year and a half, so a short time, but in that time what I've seen, you know, not only are we supporting DHS super important customer of ours you know supporting DHS super important customer of ours where a lot of the data that is being processed for, you know, the border crossings and others are housed and secured and processed within Oracle. What we're seeing just broadly, though, is the application of using both AI infrastructure within the cloud and at the edge to be able to pull different disparate data sets together in order to have a more complete picture. So, from an NAI perspective, if you think of an analogy like in health, you know you have an x-ray for a patient that shows you know, through computer vision, there's a potential things within their lungs, but that's one piece of data. It has to be married up with many other pieces of data infused so you get a more complete picture of what the potential issue is and then how you potentially would fix that. And that is the same thing at the border. It's the same thing in defense, where we do a lot of work and intelligence. It's really the fusion of that data with the NII data together to create a more complete picture which allows agents to be more thorough, to do processing more quickly and also to discover patterns.

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

And we were at the border. It was several months ago now, cold medicine's kicking in, I can't remember when it was. And it was really interesting because when agents were looking at just the patterns of where they were finding contraband in just sedans, it was one particular make model color that was more indicative and likely to have contraband in it than any other. And it's like it's a consistent theme, because it's cat and mouse game, right, but like it's a consistent theme that as you're able to fuse that data together and see those patterns, you're better able to react in the moment from whatever scan or whatever it is that you're getting at that time.

• John Wagner, former DEAC for Field Operations, CBP :

Yeah, that's interesting. Cbp gets tons of information, yeah, and I think one of the end visions here is just like what you were talking about and like Pete was talking about earlier how do you package all that together to help that officer or agent make the best decision that they can make? So how do you take the cargo description, the manifest information right and fuse that with the imaging information and say, look, if this is 20 washing machines, is this what a scan of 20 washing machines should look like? Officers can see this now based on past practice and experience. Yeah, that's what washing machines pretty much look like.

• John Wagner, former DEAC for Field Operations, CBP :

What are these square bricks over here? Those don't belong there, right? So If that's the vision on how this could end up and the possibilities for are endless at this point, right, Would CBP can build operations around this. But, like, how close are we to seeing this? Like, what are you seeing with some of the testing? Dan, we'll start with you and then maybe Sunil. Like you know, what is some of the impact been? What are some of the things that are really working? Well?

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

You know what I think, and I'm sure you saw over the many years in your career as well that once the technology gets out into the field, you know we always a lot of training, a lot of messaging that has to go Like what is the role and responsibility, you know, of the frontline personnel? But you see them really take it and make it much greater than we could even imagine or envision. I think when we talk about the application of the AI, I mean these kids and I can call them kids now because I'm older than most of them they're pretty smart. They're pretty smart, they're very tech-oriented and you know seeing, you know these kinds of deployments, I think, the feedback and once we get you know we've had some of the algorithms deployed for the NAI the feedback that we're getting from the officers on the tweaks that we have to make. But again, some really good success with these first kind of business use cases that we've rolled out.

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

They're not the complex ones. We have homogenous loads. We have, you know, passenger vehicles. They're going to be challenging but you know focusing and getting the feedback from the officers like right on the specific areas of the vehicles and we know where you know the intelligence tells us. We know what prior seizures tell us. You know there's only so many places in a car that you can really conceal things. But also layering in the other data, right, how many times is that vehicle crossed? So it's layering in all that information algorithms that are also telling us about the travel right, the crossings, like looking at you know, like, does this make sense? So it's anomaly detection, not just in the NII scans. All of this has to be packaged for the officer. And when the deputy mentioned right, we're responsible for enforcing like 500 something laws. You know both CBP and other agencies.

• John Wagner, former DEAC for Field Operations, CBP :

Including agriculture.

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

In 90 seconds or less. Right, when you think about that, they really you know. So educating the officers is really important, but I think you know getting their feedback on the application of it is the most critical kind of stage that we're at.

• John Wagner, former DEAC for Field Operations, CBP :

Yeah, and I think that does a lot for the younger workforce too, as you're recruiting and retaining younger employees and they're working for an agency that's doing cool stuff. Right when we started, we had diskettes and we had dot matrix printers. Sunil, you mentioned edge plans. Talk a little bit more on that.

Sunil Madhugiri, Chief Technology Officer, CBP:

Sure, before we go to edge, can I talk a little bit about what EAC said about the packaging piece and some of the model deployed, because I want to pin the story right as to what exactly is going on. I tell you, it's what EAC mentioned with respect to some of the deployment models. Right now there's some of the smaller, simpler models. Even to do that, there's a lot of work has to go at the back end. And the number one you know the number one thing which we go through and I hope there's a quiz after this it's called data engineering. I mean, that is probably the toughest part of any AI work we do within CVP and anywhere else. And the same thing applied to us and ESC. She knows some of the challenges we went through. I tell you I never thought in a thousand years the hardest thing which we will deal with data labeling, data annotation. Data labeling was probably the most challenging part for us, not for anything else. It's because of supervised versus unsupervised models, and these models do behave differently. Yes, you can have an officer sit there and keep updating these models, but that's not the point. The point is to automate everything so that they can do their job and do everything else right. So that's what we are struggling with. But what other piece of what we have done is? I think it's from the industry perspective. Right from the beginning I had a vision. Working with EAC, sabatino and actually Jody Harden is saying that we need to have a common framework so all these models go through a similar pipeline. That means end result. That means beginning to the end. It's the same consistent feeling across the board, no different for software development, right? If you are software developers, if you guys are aware of that, how it works, have you heard of it? It's called a SecDev pipeline. This is called a data pipeline. Same concept Model comes in here, go through the process scanning, testing, everything else before deployment and then gets rehydrated again with new information. That's what we have done. So you'll see more and more. This is just only for NII. We are doing this across the board for all model deployment within CBP. And the reason is the EC mentioned about integrated package some of the laws which we have to follow to make sure we're not only helping CBP but across the board for other agencies. To do that, we need to have a consistent architecture across all of CBP. That includes Border Patrol, amo and, of course, office of Trade. That's what we do, that's my job, from my side. So, because of this now, because of the whole architecture discussion, what has happened is if you currently 60% to 70% of all data getting generated, john right now is at the edge. If you're a global entry customer thank you for your business, that one you're at the edge looking at a screen right now and it's doing something for you, generating some data and putting it out right. All those are edges. Now, if you're looking at a sensor, you're looking at NII scan we talked about. These are all edges for CVP right, border wall sensors, you just name it. Across the board, border Patrol has cameras, rofas cameras all those components are considered edges. That's where the data is going to be generated and this is where we need industry's help.

Sunil Madhugiri, Chief Technology Officer, CBP:

What is happening is the edge computing component for CVPs perspective. We're coming out of an architecture as to how do we deploy edge architecture, do processing of the edge, whatever needs to be done, including for NII, and then send the data back to a centralized in the cloud. We call it like a spoken hub model. That's what we're looking at. One more thing I want to tell you that when you talk to us, please, please, talk to us.

Sunil Madhugiri, Chief Technology Officer, CBP:

We're looking at open architecture. I believe in three words open, open, open. All right, it's all open architecture. No commodity stuff so we can swap out stuff. Put everybody's model in there, right, every cloud provider, every vendor who's providing models to us is what we're looking for. So that's exactly that's the vision across the board, right? Not only from both our perspective, and both EAC and actually Harden have been very supportive on this one, and you'll see some of the from our perspective, that architecture is. We just published our about to publish our edge strategy. That strategy talks about exactly this how do we take all these remote devices, bring them together, analyze for retention and everything else, plus also processing quickly, as EAC said, your 90 seconds to do certain things. How do you do that? You can't bring data back, process and send it back Too late. Physics doesn't go away, right? Law of physics still stays. So because of that, you have to process a lot of things on the edge, and that's the strategy we are following, john.

• John Wagner, former DEAC for Field Operations, CBP :

Right, and that was always the thing right the latency in the responses. You know when you get information back so you can decide what to do. Does this person or does this piece of cargo need more scrutiny? But you're sitting there waiting for the computer to the clock to go away and get the results back. So, yeah, when you can speed that up, commerce can speed up. To Peter.