To Improve Your AI Adoption Strategy, Look At the Data

When it comes to AI implementation, government leaders find themselves in somewhat of a Catch-22: They need data to enhance AI programs and capabilities, but too much data can cause delays in the process. 

When left unchecked, incomplete datasets result in biased algorithms that discriminate against constituents. Add to this the genesis of new data privacy laws, like the California Consumer Privacy Act and Europe’s General Data Protection Regulation, and data can seem like more of a problem than a solution. But with a streamlined data management strategy, AI has the power to transform government systems, processes and services. Case in point are examples like the National Institutes of Health, which has developed a working group to explore and further the use of AI in biomedical research by tapping its vast networks of data. Meanwhile, cities are starting to tap their own data networks for everything from open data portals, which can more effectively disseminate information to constituents, to AI-backed tools that can predict potential threats and tackle congestion on roads. 

Still, when it comes to AI in government, we’re only just beginning to scratch the surface. That’s according to Kirk Kern, chief technology officer at NetApp, who appeared as a guest on Government Executive Media Group’s AI Tipping Point podcast along with Greg Smithberger, director of the Capabilities Directorate and chief information officer for the National Security Agency. They each shared their challenges with AI adoption and offered up some solutions government leaders can use to better organize and make use of data for AI projects. 

Below are the top takeaways from their discussion: 

1. Productive Data Management Requires Humans and Machines to Work Together

Contrary to what many people think, machines aren’t out to replace human intelligence. In fact, quite the opposite: AI is most effective when it empowers humans to do their jobs better. For the public sector, that means spending less time on menial tasks and more time improving services for constituents. 

The NSA, for example, has begun using Natural Language Processing to help linguists and analysts sort through mountains of data in various languages and dialects more efficiently.

“We can't scale to work the problem set with human linguists alone — we need the machines to assist,” Smithberger said. “Now, we’re using NLP to make foreign languages easy for our linguists and analysts to work with.”

This project is just one part of a large-scale initiative the NSA is undertaking to make its data more accessible. As part of this initiative, the NSA has overhauled its data environment, moving its data in 2018 to a “big data fusion environment” known as the Intelligence Community GovCloud. 

“The environment enabled machines to look at all the data we have and help tee it up to the humans, so that they can interact with the data as a single entity, as opposed to spending days sorting through it themselves,” Smithberger said. 

2. Data Compliance and Ethics Standards Are Paramount

Testing new technology often comes with risk, but the NSA is no stranger to managing sensitive data. According to Smithberger, the agency supports 800 to 900 compliance regimes. And, in many cases, there are legal and regulatory restrictions the NSA must adhere to. The challenge was finding a way to take advantage of automation, big data analytics and machine learning while respecting confidentiality and following legal and compliance standards. 

Smithberger and his team resolved the issue by building stricter access controls around information, labeling both staff and data to better understand who is authorized to access which information. 

“It really all came down to tagging the data,” he said. “That allowed us to easily understand who can see the data, how it needs to be handled and how long we can keep it in our system.” 

3. AI Innovation Requires Higher Quality Data and Training-Sets

Building better data and access infrastructures are only the start. To better interpret data, government IT leaders will also need to invest in a comprehensive data management strategy, according to Kern. 

“Fundamentally we still have humans in the background improving and conditioning these models for better performance,” he said. “Eventually we’re going to have to start constructing these systems with higher quality training datasets.” 

To get there, agencies will need to improve their existing data management techniques. 

“Data management in AI is almost as important as the technology itself,” Kern said. “There’s a significant amount of labor that goes into curating these training datasets and then labeling the data in the data lakes. That's where data management techniques become critical.”

He recommends adopting low-level data management primitives like cloning, copying, snapshotting, restoration and metadata management. Creating and maintaining a high-functioning data pipeline will help data scientists and engineers more seamlessly manage high volumes of data and move them across different formats and locations. 

4. Industry Partnerships Enable Smart AI Adoption

Achieving these outcomes, however, isn’t always easy, and government agencies are, rightfully, looking to industry partners for additional resources and expertise. 

Just this year, the Central Intelligence Agency announced it would be investing in a multi-cloud, multi-vendor environment called Commercial Cloud Enterprises, or C2E, which Smithberger said will broaden CIA’s pool of partners to drive more innovative products and services. 

The NSA is taking advantage of similar industry partnerships to complement the capabilities of its IC-GovCloud. The NSA will implement a high-performance analytic environment known as Hybrid Compute Initiative to address challenges like processing demands and high data quantities. The initiative will require a vendor to provide the supporting infrastructure and hardware. 

“Hardware-as-a-service will enable us to provide more scalability in partnership with the industry and help to expand these capabilities across the larger intelligence community and open the door for further innovation,” Smithberger said. “As we open up the doors for other platforms, you know, new possibilities emerge.”

5. Government Needs to Become a Larger Consumer of AI 

Many public sector organizations are investing in AI innovation, but Kern urges them to be faster and more aggressive in the adoption process. 

“The use cases that the public sector has invested in are too low impact,” he said. “Instead, we need a grand challenge AI project to push the scope and scale of what's possible. The impact of AI is so significant that it's worth taking a chance and developing a program that might fail.”

Smithberger agreed, but acknowledged that it can often be challenging for government IT leaders to take risks due to the nature of their roles. 

“The government needs to take the right level of risk,” he said. “We can’t be too risk-averse, but we can’t be cavalier either. That’s where you really see the value of working with other agencies and industry partners because it allows us to learn from each other and share our best practices so that we can go further faster.”