← The Daily Scoop Podcast
The Daily Scoop Podcast
How the CDC is using AI to revolutionize public health
00:32:47
The Department of Heath and Human Services has be…
Loading summary
Transcript57 lines
- [00:00]
A
Today on the Daily Scoop Podcast from the Scoop News Group brought to you by Databricks how the CDC is using AI to revolutionize public health. It's Tuesday, December 2, 2025. Welcome to the Daily Scoop Podcast where you'll hear the latest news and trends facing government leaders. I'm the host of the Daily Scoop Podcast, Billy Mitchell. Thanks so much for joining me.
- [00:24]
Now let's dive into the day's top headlines.
- [00:30]
The National Nuclear Security Administration is looking for information on potential AI uses for its mission, following an executive order to establish an integrated AI platform that will fuel scientific discovery. In a request for information posted on Monday, the Department of Energy subcomponent that oversees the nation's nuclear stockpile said it's exploring the use of the budding technology and specifically requested information about its use in classified environments, best practices for data curation and and how to approach, developing and enhancing AI models, among other things. The request comes just a week after the Trump administration launched the Genesis mission aimed at scientific discovery through AI. That effort will not only create an AI platform for such discovery, but it.
- [01:14]
B
Will also depend on the country's existing.
- [01:16]
A
Research and development infrastructure, including the Department of Energy and its national labs. To further the Genesis program, NNSA said.
- [01:24]
B
It'S proactively exploring the use of AI.
- [01:26]
A
For its critical operations to accelerate nuclear weapons development timeline, ensuring our deterrent remains responsive, effective and state of the art against evolving global threats. Per the request. The agency also said it plans to use AI in its nuclear non proliferation missions, ensuring robust monitoring of threats across the globe. The responses, it said, will be used to inform the plans for the Genesis program. NNSA said it's looking for information from businesses, institutions, cloud providers, think tanks, universities and others to learn more about market capabilities. Responses are due by January 23rd. Now, in other news, software company SAP inked a new agreement with the General Services Administration to offer federal agencies access to its services at significantly discounted rates, deepening its long standing partnership with the federal government. GSA announced the One Gov deal Tuesday, stating that the agreement offers up to 80% discounts on SAP's database, cloud and analytics services. The agency estimates that this will lead to $165 million in savings for federal agencies. Specifically, agencies will be able to access products related to SAP's database and data management services with an 80% discount. SAP's cloud services, including SAP Business Technology Platform, SAP analytics, cloud and HR payroll will be offered at a 35% discount. Other benefits, including dedicated support and waived data egress fees, are included in the deal, according to gsa. The agency also said the agreement is available for existing SAP customers for expansions, renewals or modernization projects. The deal makes SAP the latest major technology company to participate in GSA's OneGov initiative, which aims to work directly with IT manufacturers to consolidate IT purchasing and accelerate modernization across federal agencies. For more news at the intersection of the federal government and technology, make sure to visit fedscoop.com.
- [03:28]
The Department of Health and Human Services has been leaning into the use of artificial intelligence to drive better health outcomes for the American public, highlighted by the rollout of ChatGPT across the agency earlier this fall. In particular, the Centers for Disease Control and Prevention has been a leader in generative AI adoption since 2023, and Travis Hoppe, CDC's chief AI officer, believes AI innovation can continue to move the needle on public health operations. Hoppy joined me recently on stage at Fed Talks to share the latest on CDC's AI journey, how the Trump administration's AI action plan is guiding the agency's AI implementation and what's next. Now, here's that interview with Travis Hoppe.
- [04:10]
B
Travis, welcome to Fed Talks. I think it's the first time we've talked in a formal fashion. Usually we're just, you know, hanging out after a signal lines, you know, talking the simple stuff. But I'm excited to talk about AI. You're one of the federal government's chief AI officers in a really cool space. Public health's obviously a really important mission and there's a lot going on at HHS in that space.
- [04:30]
A
If any of you are readers and.
- [04:32]
B
You should be@fedscoop.com, we had a exclusive broke the news reporter Madison alder did on HHS's HHS ChatGPT. So let's start there and hear about the rollout of that in that the role that that's playing for CDC in its journey to adopt generative AI.
- [04:52]
C
Yeah. And thank you. And I'm excited to be here with all of you today to really talk about some of the stuff that CDC is doing in hhs. So Bill, you mentioned HHS GPT, and for those of you who don't know, the federal government inked a deal with GSA and they were trying to bring OpenAI and Anthropic to the federal government. HHS started that rollout and brought ChatGPT to everybody within the federal government, within Health and Human Services. We were able to take that. And at cdc now, every single member of staff has access to the latest models. This is cool. This is Amazing story was broken. But the cool thing is we've had this since 2023. So CDC was the first federal agency in the entire government to turn on ChatGPT for everybody. And so we did this with a lot of work. We wanted to make sure that we studied and looked and made sure what does this mean for public health? So we got there through a big combination of looking at all the data standards, looking at all of the risks that might be associated with generative AI. Since we've had it since 2023, we have all this telemetry. We could look at what everybody has done with the chatbot. And so we were able. As soon as I took this role about six months ago as a chief AI officer, I said, hey, we rolled this out. It's amazing. What have we done with it? What have we done with this generative AI? And then what does this mean for the rest of HHS? We looked at this. We had over 1.2 million chats. We looked at what was in the chats. We've saved about 41,000 hours based off the money that we've put in and the investment that we made. It is a 500% ROI on our initial investment. And so we look at this and the success of our CDC GPT models and say that's going to roll out to all of hhs. So it's like really exciting to see that happening at the department level. And we've been setting this up already.
- [06:47]
B
So it sounds like you guys are helping inform the road ahead for HHS as it becomes this scaled ChatGPT or HHS GPT. Excuse me, for the entire department.
- [06:56]
C
Yeah, for sure. It's.
- [06:59]
When we went through this, we wrote generative AI guidance and we were the first federal agency to do so. Shared it with all of our partners that are out here. I know we shared it with Commerce and a few others that ended up being the Office of OPM's Generative AI Guidance. They used ours as a model for it. Very cool, Very cool.
- [07:16]
B
That's exciting example of how things scale in government. Let's take it to the AI Action plan. A number of folks have mentioned it today and I'd love to dig in there because it's interesting to see how different agencies are looking to implement that. Particularly I want to call out the AI innovation aspect because there's quite a bit in there in terms of call outs for federal agencies. So how does CDC think about implementing that and sort of putting that into action?
- [07:42]
C
Yeah, for sure. So it really does go back to all of the boring fundamentals that we need. It needs the data monetization, it needs the interoperability, it needs all of these kind of core things. We have really worked hard on this. We started the one CDP platform that is the one CDC data platform. That is our modernization strategy. That is the way we will get to using all the fun AI projects. So that is kind of a core aspect of our data standards. That said, there's a lot of work you have to do when you have to look at all your schemas, look at all of your definitions, make sure they're all aligned. We are using AI to help with that process too. So we have to do our data standards to get to all this interoperability, to get to all this alignment, but we're also using AI to help us get to that spot.
- [08:34]
A
So let's dig, or I guess maybe.
- [08:35]
B
Take a step back and look at the public health space.
- [08:39]
A
And I'd love to hear.
- [08:41]
B
You're the acting Chief AI Officer for cdc, so you've got a nice perspective on this.
- [08:46]
A
But where is AI really moving the.
- [08:48]
B
Needle for public health operations as we think about the CDC's role in some other missions within HHS?
- [08:55]
C
Yeah, so that's a really fun question. Last year for the AI use case inventory, which is public, we had about 50 submissions in that. This year we're expected to be about double of that. So if you're wondering what we're doing, we're practicing the trust and transparency. You'll be able to see all these things. There's a few things that I really like. I like to think of AI in our organization, and I think it applies to all of your organizations as well as things like front of the house and back of the house. So we all have the same back of the house sort of work. This is the operational efficiency. This is like setting up a chatbot for people to do things, to review documents. All of this kind of like fundamental stuff, these business operations, those are all happening. And I think that's a shared common experience for all of us here. This back of the house stuff. CDC is the nation's largest public health agency. And so we have a lot of interesting front of the house work. And this is where I'm going to remind everybody that there is more AI than just chatbots and generative AI. And we've been doing machine learning, we've been doing fitting to models and forecasting for a very long time. Our center for Forecasting and Outbreak analytics has this really cool program called Flu Site, which they're using to look at hospitals, they're looking at providers so that we can have an informed decision one to three weeks ahead when a flu might be coming through. We have another one called Tower Scout, which I'm really geeking out about because it's not a generative AI project, it's a computer vision project. This is something that we've talked about a little bit before, but it's different than all the other things. We're looking at satellite imagery. And when there's a Legionnaires outbreak, one of the biggest causes of this are typically air conditioners. And in the modern day, there are these big industrial air conditioners. And this would take eight, nine hours when there's an outbreak to look through all the satellite imagery and say, hey, where should we focus our efforts? AI, the program that we set up can do it in a couple seconds. And so that's like an amazing improvement when speed is critical.
- [10:55]
B
Well, it's some really cool examples, and I think it sounds like we'll have some more coming, you know, in the future with your doubling of your inventory. So kudos on that and thanks for the transparency. It's helpful for us to be able to get a sense of that and follow along, you know, a lot of this. And I think you'll probably have some key information because you've, you've. As you said, CDC has kind of been ahead of the curve in adopting chatbots in AI. But the workforce, in words, other upskilling element, is really important. We've talked about this before, but I'd.
- [11:25]
A
Love for you to share a bit.
- [11:26]
B
More about how you think about that within your role as the Chief AI.
- [11:29]
A
Officer and making sure that the workforce.
- [11:31]
B
Is ready for these tools and using them properly.
- [11:33]
C
Yeah, absolutely. And I will Note that the CDC's workforce, we are a federal agency, but we have so many jurisdictional partners at the state, at the county, at the tribal levels. And that is part of the workforce environment that we, that we, that we think of. But just within cdc, we have a really robust community of practice, which I'm really proud. There's about 20% of all the staff are in one Teams chat. That is just wild. And they just talk and they share information about this. We have office hours, we run this little thing.
- [12:06]
Where we ask staff what their favorite prompts are to use the chatbot, and that engaged staff in such a fundamental way where they could see themselves at all different levels. So if you're thinking about workforce, if you're thinking about getting your organization to adopt generative AI. It really has to happen at both the top level as we give information out, but also kind of at the bottom level. Let people start their own communities of practice. Let them bring them in there and find a way to highlight them.
- [12:33]
B
That's great. That's great advice. We're quickly running out of time and I want to squeeze one last in for you because I think it's something that people often toy with in this space. When we think about balancing AI, innovation and risk, how do you navigate that? It's not an easy thing to do.
- [12:49]
C
So you asked me this question with about 15 seconds left on the clock. I think probably the best way we're looking at this is thinking about things like system cards and model cards. When vendors come to us, there's a standard set of information and we need to start collecting that and we need to standardize that in our governance model.
- [13:05]
B
Travis, sorry to give you a short one or a big question for a little amount of time, but you absolutely killed it. Thank you so much for your time up here. Let's give him a round of applause.
- [13:15]
C
Thanks everybody.
- [13:16]
A
For more on federal AI adoption, make sure to visit fedscoop.com.
- [13:23]
Also in this episode, Databricks VP of Public Sector Todd Schroeder joins SNG host Wyatt Cash in a sponsored podcast discussion on why agencies are prioritizing the use of AI that works across existing data environments, saving time and infrastructure costs. This segment was sponsored by Databricks.
- [13:44]
D
Federal agencies are entering a major inflection point as AI reshapes how government invests in technology to get work done. After decades of accumulating disconnected systems from mainframes to cloud and so on. Software as a Service tools Agencies are still wrestling with fragmented data and high operating costs. Modern data platforms and AI, however, are really putting agencies at a new inflection point, enabling them to access and govern and analyze their data across any cloud or any system without requiring data duplication or incurring the costs associated with moving vast amounts of data back and forth. And this enables real time data sharing with federal partners and reducing the need for billions in redundant infrastructure and accelerating decision making. I'm Wyatt Cash with Scoop News Group, and joining us to put these new capabilities into perspective is Todd Schroeder, Vice President of Public Sector at Databricks. Todd has been on the front lines of government technology for many years, having served as chief of the Department of Agriculture's Digital Service center before helping agencies implement a lot of leading edge digital strategies while working at Salesforce, Google and UiPath. Todd thank you so much for joining us and welcome to the program.
- [15:03]
E
Thanks for having me, Wyatt.
- [15:05]
D
So, Todd, I think you'd agree, agencies have made significant strides in modernizing their infrastructure and applications, and yet, you know, they're still struggling, you know, to get a handle on operating costs, et cetera. What would you say is the biggest challenge, challenge that most agencies are still facing?
- [15:23]
E
Yeah, I think, you know, as, as we just talked about there, there's a machine in place. The machine does what the machine does. And, and largely I think the way we've, we've talked about and achieved results is by a certain approach to technology transformation or mission transformation. And what I, what I mean by that is traditionally to accomplish something new, we had suggested you must build something new, you must take out what is there and put something new in and you will get new results. We saw this from data center to cloud explosion, that era, that, that kind of paradigm shift. The data centers weren't good enough. We needed infrastructure as a service to move faster, maybe reduce our reliance on custom code. And as PaaS and SaaS offerings grew in, in the number and volume of offerings, we saw more and more of that, that really kind of ripped our organizations apart from a data standpoint, from a process standpoint. So it's difficult to stitch together one program to another program to a hundred other programs that are ultimately part of the system of systems or how a mission outcome is produced for society or stakeholders external to the government. And so I think one of the challenges still remains is the mindset of how we change.
- [16:48]
So while that's sort of a kind of historic precedent of how we make decisions and then the process behind those decisions, new technology in this AI era is offering new techniques to suggest that that isn't viable path anymore. It's not the fastest path, it perhaps isn't the most cost effective path. And so that's where I see the, the conversation and, and the listening really going. Because we can't take three years to start over to get new results. What we do is steeped in decades of iteration and improvement. And so for me and the customers, we're engaging with the idea that you can transform in place, you can become more intelligent, you can have more empirical knowledge across your programs by connecting that data without changing how all of those programs work today. To get to those insights is empowering. I understand and can reason with my enterprise differently today then, you know, aggregating data from every system that I have to try to understand it so I can reason real time with how it's all connected. That reasoning is where we're actually seeing an increase in velocity or acceleration against doing something about that. And so when I can have empirical evidence from all of my systems without moving physically changing how they work today, and I can bring a level of security and governance to that data, I can bring every AI model in the world to my data. And I have a bevy of capabilities on the platform with respect to data pipelines, which will improve automation and reduce bureaucracy or increase efficiencies. I can deploy agents that have been evaluated for accuracy and cost that helps my workforce do more, faster, with higher quality or degree of confidence in the decision. So it's those capabilities on the other side of the data and intelligence platform and databricks that we go from finding what's happening to fixing or changing the outcome produced. And that's kind of a beautiful marriage between the way we operate today and adding iterative AI or machine learning or agentic approaches to actually improve what's in place today. So we can kind of bifurcate ourselves away from all or nothing wholesale changes in hopes that it will produce a different outcome. It also slowly and confidently reduces the reliance on software tier spend. What we're seeing is about 50% of most of the applications that the frontline uses to accomplish its job were done to support knowledge work. I need to check something against the regulation. I need to check if this person is a bad actor. Those knowledge checks and the software that support those knowledge checks can be automated with huge amounts of accuracy into the data pipelines. And so when we ask those individuals to adjudicate a permit to inspect something for a, you know, social health benefit, all of that pre work, all of that knowledge work has been done packaged together for that employee to make better, faster decisions without such a drag on kind of that task based, knowledge based work that actually helps us reduce software spend, simplify the process that we're asking our workforce to go through to accomplish the job, and presents opportunities to continue to simplify a very, very complex architecture in any one department.
- [20:43]
D
So I'd like to drill down just a little in terms of what's really changed over the last couple years. I think you alluded to AI and the power that it brings in the way of contextual reasoning. This idea that you can bring these AI tools and applications to the data instead of the data to the application. So that sounds like a pretty fundamental thing, but can you also explain briefly what's really changed over the last couple of years and what databricks data lakehouse approach enables agencies to do that allows them to arrive at decisions faster and more cost effectively. And. And why is it so game changing?
- [21:22]
E
Absolutely. So your, your first question, bringing AI to the data. I would remiss if I didn't say back to our first question. The approach that I've seen the last couple years with respect to AI and it's ever increasing, was to look at an AI tool set. Buy that tool set, start moving your data to that tool set. And so in principle, we're following that paradigm that I spoke about that we really need to break, right? If I go buy a tool set that does something with AI and I have to move all my data to it, I'm creating another island of data. I have hundreds, perhaps thousands of islands of data already that I need to organize together to get the most out of AI, no matter what you're doing with AI.
- [22:12]
And so that's what we saw the last couple years. Databricks is so unique in that it brings all of those models to your data plane so that you don't have to take your data pieces or parts of your enterprise to one tool or arguably 50 different tools. You get all AI models brought to your data in your security wrapper, completely contained inside of what we call unity catalog. But that is complete governance around your data and how that data is exposed to or leverages AI. That's both. From a technical standpoint, I can see what model scores, what accuracy with respect to my own benchmarks, industry's benchmarks, how accurately it can understand my policies or other policies. So I know I'm using the right kind of AI for the exact right use case based on how I balance accuracy and cost. Super, super important in terms of scale. If my approach is take my data to AI, I'm probably going to have an even more complex architecture that has more security challenges or vulnerabilities that costs me more over time. The inverse and what Databricks is offering is you should have complete access to any model in the world and you should know which model to use for what use case very confidently so you can scale those use cases into production and know confidently how they're going to perform in regulated processes. Model evaluation is kind of the crucible for scaling AI. Being able to build your own model evaluation capability so you know how that model performs inside of your organization is what will give you the confidence coupled with complete governance from data all the way into how you're using that data with these AI models. And so that is one of the reasons that databricks is so unique so kind of skating to your second question, Wyatt.
- [24:21]
Databricks invented several open source technologies, Spark being one of the cores ML flow Unity catalog. So these are open source capabilities that have grown in global fans, right? Best of breed technologies that the databricks data intelligence platform has now built robust enterprise features around the lakehouse is the only lakehouse in the world. And you know, others are picking up on this, on these words. But what we do uniquely at our at our core is you get access to any of your data without physically moving your data from any cloud environment. So back to where we started our conversation. You have infrastructure providers, data centers, PaaS offerings, SaaS offerings that probably hundreds and hundreds of different endpoint applications that all have data. To be able to see all of that data and organize all of that data without physically moving that data is very unique to databricks.
- [25:29]
Traditionally speaking, old school warehousing technologies would have suggested you actually have to physically move all of that data.
- [25:38]
Our competitors still still require you to physically move all of that data just to get visibility, access and utilization of it. It's spark the processing underneath the hood that allows us to do this at such performance. So if I can see all of my data across my entire enterprise, no matter where it's physically residing and I can actually map that data together and wrap all of that data with security and governance, I can, I can see what piece of data changed for all of time, who changed it, why change, why it changed that Unity catalog capability, our data governance capability scales through the rest of our platform. So now data and the governance of that data become an enabler of what, what I can do with that data as opposed to long costly integration of data with software to your assets. So now on top of that open data platform we have multi model serving or our AI capabilities that we just spoke about bringing all of the AI models to you. We have the model evaluation capability that helps you test and make pointed decisions on what model for what use case and put those into production pipelines. We also have a full complement of agent capabilities. Now I can train agents based off AI and data pipelines. So those tool sets are what I'm talking about in terms of retiring the knowledge worker applications. It is the data plus our AI capabilities that really allow us to retire software tier spend and improve the processing of how data moves through an organization to result in a decision. Right? So the other part of our platform, data AI now exposable through data products. So we think about these as the things we interact with to do our jobs. So these are natural Language based querying. I can immediately query my data using natural language and because of the semantics of the data, it understands how a procurement and an invoice are tied together. It will help me understand root cause challenges. We have assistant products that will help you run data science and understand things that normally a statistician might need to help you with. And we have composable experiences, vibe coding experiences that allow you to bring all of this data AI together to present more AI powered user experiences, both NLP based and things that look like normal traditional case adjudication type applications where you can see agents work next to the case backlog, perhaps that you need to adjudicate. All of those things because they're so tightly integrated on one platform mean that we are seeing time to value happen in days, not weeks, months or years. And we're just seeing the cost profile. I've seen agents go live in two minutes in one day as opposed to very long drawn out procurement cycles and things like that. So the escape velocity of just organizing your data to make mission change is so unique and so disruptive and I think it's something that databricks is really helping the federal government take advantage of.
- [29:03]
D
Well, on that point, can you provide some practical examples of where maybe some federal agencies are already seeing a tangible difference in infrastructure saving or speed to value currently?
- [29:17]
E
Absolutely, Wyatt, and there's too many to go into too much detail, but these are high volume, nationally important programs. CMS helping thwart fraud. Turning real time data sharing from states and counties into CMS to improve the cycle time between high quality data means policy reform meets updated benefits and payments back to the states. The United States Postal Service, a great customer, but they are using us in a ton of different areas as we approach our busy season here with the usps. But, but mail operations, how, how those packages are logistically maneuvering through the country to get to, you know, you and I's house. The IRS is another great example in terms of complex high volume, un unstructured data flowing through highly regulated approval processes is another area that we're helping. And I'll leave you on our program at Ivana. You know, over 900 disparate systems, all in the realm of financial management, procurement, you know, overall auditability for the Department of Fence. So it's again in these examples it's organizing your data gives you very unique insights and then the capabilities on the platform help you change what those outcomes are, remove fraud, accelerate processing, deliver on sort of the promise of public sector and kind of those impacts in local communities. And societies.
- [30:49]
D
Fantastic. As we wrap up, I know you're showcasing a lot of those stories in the coming weeks at one of your events. What would you want the government folks to know about what will be available to see and if they can't make it to that event, what else they might want to look into?
- [31:06]
E
Yeah, please do come to the event on December 11th at the Reagan Building. It will be full of your peers in government and your partners in government sharing their stories and the journey they went through with databricks.
- [31:21]
D
Well, certainly seems like the opportunity to reduce all that infrastructure investment and re channel it into getting better decision making is something we're very much closer to being on the cusp of than maybe we ever have before. So, Todd Schroeder, thank you for joining us. Sharing some of what you're seeing, what's different, and what federal agencies could learn from other agencies about how to capitalize on that. So thanks for joining us.
- [31:47]
E
Thank you, Wyatt. I really appreciate it.
- [32:10]
B
Thanks so much for tuning in to another episode of the Daily Scoop podcast, available on all podcast platforms. If you've already rated the podcast on your platform of choice, thanks so much. High ratings and good reviews of the show help more people to find it. The Daily Scoop Podcast is a production of the Scoop News Group in Washington, D.C. adam Butler and Carlin Fisher help put the show together and the entire.
- [32:31]
A
Scoop News Group team contributes.
- [32:33]
B
We'll be back tomorrow with more top headlines. Until then, I'm your host. As always, Billy Mitchell. Thanks so much for listening.