Enabling Responsible AI by Managing Data at Scale

Brad Corwin became interested in AI while working with embedded programming and flight simulators. A computer engineer by training, Brad joined Booz Allen after completing his master’s degree.

Today, Brad serves as a director for AI and data operations, and says his work with defense clients centers around helping them understand disparate data. Brad points to the Department of Defense’s (DOD) 2023 Data, Analytics, and Artificial Intelligence Adoption Strategy, which identifies quality data as the foundation of the DOD's “AI Hierarchy of Needs,” since all analytic and AI capabilities require “trusted, high-quality data to support decision makers,” according to the strategy.

Here, Brad discusses the foundational role that data plays in enabling AI for defense missions. 

For Booz Allen’s defense clients, it’s all about “bringing together the right data for the mission and then developing that into an analytic or leveraging AI to enhance a mission in a way that creates impact,” Brad explains. To do that, he applies a data mesh approach.

“Data mesh is a mindset shift for an organization’s culture,” Brad explains. “Users start to think of data as a product rather than just a dataset.”

“The words seem similar,” he acknowledges, “but there’s a difference. With a data mesh, you make sure that the subject matter experts who know the details about the data assets are adding information or metadata. They package that information up securely to make it easy to understand and share.”

Brad notes the term data mesh was coined by Zhamak Dehghani. She identified four core principles of data mesh architecture: domain-oriented decentralized data ownership and architecture, data as a product, self-serve data infrastructure as a platform, and federated computational governance.

“With self-serve infrastructure, teams can create data products,” Brad explains. “Federated computational governance enables the control over data while adhering to standards, and domain ownership incorporates the subject matter experts.”

“Ultimately, a data mesh is a way to think about data that makes it easy to consume,” he continues. “It's critical for DOD because of the pure breadth of the Department. It makes the data interoperable, which helps to deliver a common operational picture from the boardroom to the battlefield, faster decisions enabled by responsible AI, and increased efficiencies.​”

AI and how to use it when managing and accessing data at scale is a critical part of the data mesh conversation. “Everyone wants to find more and new ways to leverage artificial intelligence,” says Brad. “It comes down to how you’re enabling users to apply those insights.”

Trust is another matter. “How do users trust AI? A good model will only be built off trusted data, which can be empowered by the data mesh,” Brad notes. “We need the ability to access data quickly and as close to the source as possible.”

Security is also important. “You get into different levels of classification and different compartmentalization,” Brad adds. “You may not be able to quickly access that data. With the data mesh and federated computational governance, you can find ways to gain access securely and quickly by keeping the control with the domain owners.”

Organizations considering a switch to a data mesh approach should first consider whether it’s the right solution for them and then focus on the approach to take. There are differences between a fully centralized, fully decentralized, and federated mesh approach, and much depends on the organization’s size and complexity.

Brad reiterates the need for a cultural shift within the organization. “It’s crucial to change the mindset. Instead of thinking about how to access a specific data system, users should think about how to answer a question or solve a problem,” he notes. “From there, building a culture of trust across data products is critical.” 

In addition to this client work, Brad also emphasizes the importance of mentoring the next generation of talent at Booz Allen in applying technology and collective ingenuity to solve real mission challenges through events like hackathons.

Earlier this year, Brad notes, multiple Booz Allen teams participated in the BRAVO 11 Bits2Effects hackathon at a DOD AI Battle Labs facility in Oahu, Hawaii. Over the course of four days, nearly 600 participants developed solutions against over 100 DOD use cases. In the end, a Booz Allen team won the "Best Time-Saving Award” for their innovative solution aimed at addressing a transportation-related DOD challenge—a key consideration given the vast expanse in the Indo-Pacific region.

Reflecting on his own path to Booz Allen, Brad adds, “You see the end-to-end impact of your work at Booz Allen. Whether it's data or AI pipelines, you build out the programming, and you get a test set with the client. You can see the results of what you’re doing. What you do here matters, whether it’s saving money or saving lives.”