HSDF THE PODCAST

From Innovation to Action Integrating Technology in Law Enforcement - Part 1

• Homeland Security & Defense Forum

Share episode

Integrating technology in law enforcement transforms innovation into actionable tools, enhancing capabilities in crime prevention, investigation, and community engagement.

This episode explores how law enforcement agencies like HSI and CBP are integrating AI and technology to enhance national security operations. Key topics include the challenges of achieving interoperability, the importance of accurate data, and the necessity for seamless integration of new tools into agents' workflows.

  • Michael Prado, Deputy Assistant Director, Homeland Security Investigations
  • Ryan Riccucci, Division Chief, Enforcement Technology & Operational Programs, CBP
  • Martina Melliand, Director of the Enterprise Analytics Division, CBP
  • Bogdan Frusina, Founder, Dejero
  • Luke McCormack, former DHS CIO (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.

• Luke McCormack, former DHS CIO (moderator):

I appreciate it, megan, and appreciate all of you sticking around. I always like to start these panels with my now famous pop quizzes. I happen to read the White House executive order. Reread it. This is the one on artificial intelligence that was issued October of 2023, october 30th. So I said, well, let me find out how many times integration is cited in the executive order. Anyone want to guess Less than 10. One time I was really disappointed. Law enforcement Take a guess? No panel members guessed because someone's going to get some coffee. Anyone want to take a guess? Law enforcement how many times? Come on. How many Close enough? 11. Not bad. Dhs 37 times it was cited.

• Luke McCormack, former DHS CIO (moderator):

All right, we did have a last minute substitution. I love the Border Patrol's nimbleness. Thank you very much for doing that. All right, mike, let's start with you at HSI. Let's start with you at HSI. I want to talk to you about, like you know, you all have a monster responsibility with national security, protecting on child exportation, et cetera, illegal import of goods and items. It's a big portfolio. Give me a and, if you could, you know, enrich us in speaking on behalf of all of HSI versus perhaps your own unit. How's the technology? What's the state of the technology out there these days. You know you've got a case management system out there. Does it need to be refreshed? How's Raven doing, or is it just? Hey, give me some new PCs and radios and mobile phones and we're good to go.

• Michael Prado, Deputy Assistant Director, Homeland Security Investigations:

That was the old days. That was definitely the old days, just a mobile radio Softball to set them up?

• Michael Prado, Deputy Assistant Director, Homeland Security Investigations:

Yes, mobile radio and a cell phone and a badge and a gun and go out and make your cases, but obviously times have changed.

• Michael Prado, Deputy Assistant Director, Homeland Security Investigations:

Just very briefly speaking on behalf of HSI, I know a couple of my colleagues were on a panel earlier talking about the fentanyl crisis.

• Michael Prado, Deputy Assistant Director, Homeland Security Investigations:

I know a couple of my colleagues were on a panel earlier talking about the fentanyl crisis.

• Michael Prado, Deputy Assistant Director, Homeland Security Investigations:

Everything within HSI is and you used that word integrated earlier in your opening remarks your pop quiz.

• Michael Prado, Deputy Assistant Director, Homeland Security Investigations:

That's really the underlying premise of everything we're doing at HSI and certainly within the DHS Cybercrime Center is trying to integrate our technology to keep up and keep a pace with the adversaries that we're tasked and required to go after. You know they've gotten more sophisticated, so in turn we're trying to keep up. You talk about the AI governance document there. The executive order that certainly guides us, has guided us, especially when it comes to some of the evolving technologies that we're involved in has guided us, especially when it comes to some of the evolving technologies that we're involved in and you mentioned, raven, trying to get beyond that cell phone and mobile radio now having essentially a handheld, personal AI assistant for each special agent, each analyst in the field to be able to tap into all of the disparate networks, disparate systems that we have not just within HSI but within DHS and within law enforcement and US government as a whole, is absolutely critical to be able to move at that speed that's necessary to combat today's 21st century cyber adversary.

• Luke McCormack, former DHS CIO (moderator):

Appreciate that and interesting outlook on some of the sort of technology landscape at HSI, sure to accelerate here in the not too distant future. Let's talk about PMOD. I want to talk about the capability roadmap and when you look at that and I think about the interoperability and some of the coordination across the various offices, you know how does one sort of look at that and make sure that these things are being instilled properly thought through as far as looking at that different technology and then making sure it's getting incorporated smoothly into those various entities?

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

That's an excellent question and I would point to open source. You can find PMODs I don't know the title of the design manual for interoperability and there is something very important in there that's called the levels of conceptual interoperability model and that model has six levels, starting with the basic levels of integration, which is systems and syntactic. Can you connect things, can they exchange information? But as you move up that levels of conceptual interoperability model, you get to level six, which is the highest level of interoperability, which is what they call the reference layer. What that means is you have to have the ability to have a shared conceptual reference model that has the same embedded semantic context or meaning. What I mean by that is, in our data models you have embedded computable meaning to an interoperable standard like the one I mentioned at an earlier talk basic formal ontology.

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

So before we even started, I should have asked for a pop quiz myself, like how did we even define integration? Are we integrated here in this room because we're together? Are we integrated because we're on a panel? I don't know the answer to that, but how I think of it is, you know I can call China. We're integrated, at least through communications. I may even be able to speak Chinese or have a translator.

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

Are we integrated? Yes, I think. Are we interoperable? Yet I don't know. I would say I would need the capability to be able to exchange information and understand meaning in a way that creates value and solves a problem that we couldn't individually achieve. I think that metaphor works generally for agents when we talk about systems. But to answer your question and sum up, the capability roadmap is a map for maturation and evolution, and then the design manual for interoperability is us wanting to walk up that ladder to get to interoperability at the reference layer. So if I want to be interoperable with ICE, cbp, fbi, nato, then we'll have a reference model that is language agnostic, where we can do machine-to-machine communication using those standards to solve problems.

• Luke McCormack, former DHS CIO (moderator):

Appreciate that, All right. We're going to go over to CBP and talk about data analysis and ground zero, right? And if I think about some of these enhanced technology, obviously we've talked throughout a good part of the day about some of these capabilities which, undoubtedly, you all are starting to layer into your environment there. What does that look like? How do you use that? Your environment there. What does that look like? How do you use that? How do you make sure that's going to be integrated smoothly, properly and then allow these decisions to be made with the operators?

• Martina Melliand, Director of the Enterprise Analytics Division, CBP:

Sure. So you know, generative AI and machine learning are really impressive and have been making improvements significantly over time, but in my experience, I've often found that they aren't the right tools for the questions that we have. I've heard it quoted that 80% accuracy is really great for some tasks and horrible for others. Imagine if our airlines had an 80% accuracy in landing airplanes. I feel like in our CBP mission space, we are very similar. We don't have the luxury of 80% accuracy. We have lies on the line national security missions.

• Martina Melliand, Director of the Enterprise Analytics Division, CBP:

So when we talk about machine learning and AI, we often talk about it as if it's a thing.

• Martina Melliand, Director of the Enterprise Analytics Division, CBP:

It's not a thing, it's an outcome.

• Martina Melliand, Director of the Enterprise Analytics Division, CBP:

It's an outcome, and what we're doing is we're buying the capability for our machines to learn to reason, and the outcome of that is AI to learn to reason, and the outcome of that is AI. Those are the two things that we've really been focusing in on, and when we go back and reflect on this need that we have to be accurate, we are really focusing in on the foundational piece I think that you've heard a lot about today already, which is how do we get good data? It doesn't matter what AI you throw on top of your data if you don't have good data. So we have to apply meaning into our data, and the goal of which is to be able to have our operators trust the analysis, trust the tools that we are putting into their hands at the end of the day, so that we have the foundation for human machine teaming. Michael mentioned it earlier he's putting devices in the hands of HSI agents so that they're able to rely on that as a tool. We want to take that even further and have that human machine teaming capability available for our agents, officers and pilots.

• Luke McCormack, former DHS CIO (moderator):

On the line. So do you feel so far it's been successful? Do they trust it or is it like whoa? This is Certain things.

• Michael Prado, Deputy Assistant Director, Homeland Security Investigations:

Okay, fair enough.

• Martina Melliand, Director of the Enterprise Analytics Division, CBP:

I think that we have a long way to go to be able to say that we trust our data. I mean, I think that's the question that it would go back to Do we trust the data that we have in our systems?

• Luke McCormack, former DHS CIO (moderator):

Fantastic. All right, let's talk from an industry perspective and I want to talk about we're going to get right into it. Let's talk about the discussions around requirements with the operators right, and the folks that are generating these requirements, having these conversations and trying to bring these conversations left of a procurement so that you're not in sort of this quarantine type of situation. And you've had a lot of experience in this area. Perhaps you'll have some recommendations on what you've seen, what's been working, perhaps what's not working.

• Bogdan Frusina, Founder, Dejero:

So I think the first thing that we deal with a lot is silos, and what I mean by that is conflicting requirements from different departments within the same agency for the same thing, which is kind of ironic. So, for example, you have Border Patrol requiring Border Patrol, requiring devices for communications, for example, that are small, diverse, some body worn, but then at the same time you want vehicle you want, like the diversity of the requirement is significant. And then at the same time you have the tower program, which requires a fixed location for the same type of communication device, for example. The requirements itself are actually the same. It's like I need uptime to a level of a certain amount versus the money I want to pay. I want five, nines uptime, or I want three, nines uptime and I'm willing to pay $100 or $300, whatever. That is right. It's a scale that you decide on that.

• Bogdan Frusina, Founder, Dejero:

But today what happens is you're taking the tower program or you're taking a particular program itself and you're creating the requirements only around that program, but yet 90% of that standard that you've built can be utilized across 30 programs or 40 programs. What that does? It gives you scale, it gives you efficiency, it gives you a lot of cost reduction and the last 10%. You do customization and I find that a lot of times these programs individually are tasks with the requirements completely separate from the others and they don't communicate. Yet they have the same requirement.

• Bogdan Frusina, Founder, Dejero:

I need an uptime, or five nines, and I need the speed of whatever 10 megabits or 100 megabits or whatever that may be so I can move the data from the field into a processing center, because, at the end of the day, anything that you do at the edge, anything that happens in the field, it has to communicate back and forth, whether that's information to the agents or that's information from the agents back to a central environment where you may use AI to process some of that, or you may use documents for storage, for identification, or you may use it for an operation at that moment in time, because it's real time and it's important, whether that's video, whether that's text, whether that's an image, whatever it may be.

• Bogdan Frusina, Founder, Dejero:

And I find it's interesting that the diversity so what I've found so far that has worked very well is actually talking to the field and then bringing that conversation from the field back into a PMOD conversation and going, hey, this is actually what I've heard, that the field happens and then they don't always have that full-fledged conversation and that actually creates a little bit of a.

• Luke McCormack, former DHS CIO (moderator):

I've lived it. Yes, Not on the same page, right.

• Bogdan Frusina, Founder, Dejero:

And then you've got the standardization world, which is the OIT side of it, which tries to drive security infrastructure, which is extremely important, right. So standardizing across all these things I think would be very helpful if there would be a lot more collaboration on things like communications, on things like standardization of data, as you've just described it right what is data and how does it communicate? What is a standard that you require for accuracy from an AI processing perspective, and how quickly and how much data do you need to move, and what data is pertinent for your HSI operators in the field at that moment in time?

• Luke McCormack, former DHS CIO (moderator):

Let me do this. I should have mentioned that we will take a little bit of time at the end here for some questions, so please make sure you have some of those teed up. I'm thinking that some of you may want to weigh in on this conversation, but I also think some of the panel members may want to weigh in on this. Thoughts about what is being described here. Any of you?

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

All right, fire away.

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

I think that it's human nature to jump right to the problem and I think that the first of all, I agree with everything you said, but we skip over the first requirement, which is are we capturing the subject matter, expertise to understand and frame the problem, and then are we representing that in an unambiguous and precise way for both people?

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

And sometimes what we're talking about here are machines to use, and I don't think there's any real new requirements. I think there's, except for one, a more strict requirement in our knowledge representation for how we actually capture that to use information. I mean, a famous physicist recently said a world-class physicist that information has, has, uh, physical properties and velocity it makes up, it's a building block of the nature of the universe that we should consider it the fifth state of matter and, in order to actually harness what he's saying there is, treat it not as something that is just this ephemeral, like, yeah, data interoperability, information, knowledge, but treat it as a commodity and and drive requirements about how you capture, represent and take these artifacts or data products and make them consumable, but from an interoperable way in terms of that levels of conceptual interoperability model Dr Andreas Tolk actually he was the author of that published paper.

• Bogdan Frusina, Founder, Dejero:

So shout out to him. That's well said and I also think that, frankly, it's one of the most complex things to do something extremely complex and make it simple, it is the hardest thing to do and the reality is you've got age gaps between the agents and awareness and knowledge and willingness to use technology. I think that's also part of that's influencing requirements and needs for the various systems. You know, from our perspective, from a communications perspective, what we've tried to do is we've tried to simplify the world. We've tried to create a standard communication system that allows you to plug in your uptime and your capacity requirements at the edge and in various formats, whether it's vehicles, bodies or buildings.

• Bogdan Frusina, Founder, Dejero:

So people, vehicles, buildings that's how we look at it and then you plug in whatever links you ultimately want and those will drive an uptime and a capacity requirement that you can now define that across everything and then just move the data back into the same environment, the same system on the other side and you prioritize the information as you need to.

• Luke McCormack, former DHS CIO (moderator):

Speaking of technology integration. I was going to go.

• Michael Prado, Deputy Assistant Director, Homeland Security Investigations:

I got a good points here and, like Martina was mentioning a moment ago, the data is important, right, and the tools that we need to govern that, to collate it, to be able to access and analyze that data Each of our agencies and obviously the private sector. We're fantastic at collecting data. We're fantastic at vacuuming up for lack of a better term all the data. I've got 235 domestic field offices and 90 foreign offices full of HSI special agents doing the investigative work. That doesn't include the massive amount of personnel that CBP and Border Patrol have all dumping information into disparate systems, siloed systems, being able to use that data. That data exists.

• Michael Prado, Deputy Assistant Director, Homeland Security Investigations:

So the AI tools that we need, the AI tools that we're trying to develop in-house and that we're trying to procure from all of you, are those tools that will connect those dots amongst and kind of go back to that one DHS model that we've all kind of grown up with, at least professionally, going back to the creation of DHS back in 2003. So that's the necessity because, again, it's not a matter of being able to collect the data, it's being able to take the data, interface it with everything else that's available to an investigator or to an officer and be able to action, that to be able to determine and predict with a high degree of probability where the next incursion on the border is going to be, where the next container is going to try to be smuggled into, where the next transnational criminal network is going to try to scam a financial institution, for example. So that's the necessity that we're looking at.

• Luke McCormack, former DHS CIO (moderator):

Let's jump over to CBP for a minute and talk about just technology integration. Right, you all are putting together a lot of capability there. You're layering on some technology. You're sending it out into the field, allowing them to make good decisions. Give us some lessons learned there, maybe some success stories about some things that have gone well. Maybe some things that you've learned that you know clearly need to be recalibrated.

• Martina Melliand, Director of the Enterprise Analytics Division, CBP:

Yeah, Luke, broadly speaking, what we did was we married symbolic reasoning with traditional machine learning. Now, historically, traditionally, those two things have been separate, but what is novel is the graph-based standards approach with symbolic reasoning, with machine learning, and we found that when we did that, data latency was no longer an issue. We were able to trace our data back and we were able to have data available to us in real time to help us make predictions. The problems that we're all facing, that you're hearing about, they're not new. The solutions that we're employing are not new, but what is new is how we're bringing these solutions together, and it's been successful.

• Martina Melliand, Director of the Enterprise Analytics Division, CBP:

We actually did this. Earlier this year. We had a field demonstration testing our capability to integrate several systems, automate pattern recognition and send an alert to a CBP handheld device and a CBP helicopter, which worked in a cost-effective way. That enhanced the situational awareness for our agents and pilots for their safety, and the even more amazing thing is that the pilot trusted the device when it alerted the pilot to a potential threat in the air. That's our first example of that real human machine teaming that I was talking about before.

• Luke McCormack, former DHS CIO (moderator):

Impressive and I'm going to get to you in a moment. But let me ask you sort of a similar question, a little bit of a different twist. As you all are looking to incorporate some of these technologies into your environment and then send them out to these agents, that has to be sort of fused with the rest of the work that they're doing, right, whether it's a new analytics capability, something to do with the case management system, whatever it may be. How does that work? How do you sort of think that through? What are the steps you're going through to make sure? From an agent's perspective, that's sort of a seamless experience to the greatest extent it can be yeah, and that's the challenge.

• Michael Prado, Deputy Assistant Director, Homeland Security Investigations:

Right and seamless is the operative word is you know agents, you know out in the field and again, we, we, we are so focused here and I've been at headquarters off and on for 10 years, minus some field assignments in between but when you're in the field, you're not necessarily thinking about those types of issues. You're thinking about the day-to-day making your investigation, connecting the dots and growing your investigation to ultimately lead to a prosecution. The last thing we want to do from a headquarters perspective, and certainly from a senior executive perspective, is hinder the day-to-day operations, the investigations that are taking place. So we try to do it in a manner that can integrate with what they're already doing. Right, and getting that feedback from the field is absolutely paramount. And asking the field, what is it that you all need? And having those town halls and whether it's myself or any of my senior executive colleagues going out to the field and actually hearing from the ground, from the troops, from the agents in the field, what is it that you need? And so I used to run our tactical teams for our DC office when I was in ASAC and you know working with those guys what is it that you actually need versus what is it that headquarters thinks you need, and that's a huge difference.

• Michael Prado, Deputy Assistant Director, Homeland Security Investigations:

Hearing it from the agents in the field as to what their day-to-day challenges are and then trying to again seamlessly integrate that into their ongoing investigations without creating any sort of undue delay in their day-to-day work is absolutely critical. It's a challenge and not always successful, especially when you roll out a new system. I was in the field when ICM came on board and we made the switch from tech to ICM and talk about taking a couple of weeks off from your investigations just to learn the new system. But that was necessary. But once you got it, once you got the handle and the hang of it, then you were so much more efficient as an agent, being able to upload your reports and being able to upload actual images and things to the reports as attachments. So sometimes there is a necessity to create a little bit of a stoppage if it's going to ultimately get a better return on investment, but we try to minimize that to the greatest extent possible.