Jun 24 2021

What Is the State of Government Competency Around AI?

Agencies are enthusiastic about artificial intelligence but need to invest in technology and technical staff to move beyond the pilot stage, recent research suggests.

Artificial intelligence technology has the potential to help agencies make better diagnoses and analyze reams of data more easily. While many federal agencies have piloted AI-based solutions, they are being pushed to go beyond trials and expand their use of AI, according to a recent survey and report.

The report, “From Pilots to Proficiency: Operationalizing Federal AI,” conducted by MeriTalk and underwritten by Dell Technologies and NVIDIA, found that while there is enthusiasm for AI and progress is being made, agencies face challenges in scaling AI projects. To do so, they will need to invest in a robust IT infrastructure and training, the survey indicates.

According to the survey, 71 percent say their agencies are struggling to take localized AI pilots or skunkworks programs and incorporate them into overall IT operations. The report is based on an online survey of 150 federal IT decision-makers familiar with their agencies’ use of, or plans for, AI in April 2021.

“One of the biggest challenges I’ve seen around incorporating AI into an agency’s IT operations is modernizing the compute infrastructure,” Larry Brown, solutions architect manager for NVIDIA, tells FedTech. “To move enterprise AI forward, agencies should take a holistic view of their infrastructure, modernize their networks, upgrade storage capabilities, invest in high-performance computing and expand scalable cloud solutions.”

How AI Is Being Used in Agencies Now

According to the survey, 87 percent say they see operationalizing federal AI as the cornerstone of a digital-first government. The majority of federal IT decision-makers surveyed say their agencies have more than 10 AI pilot programs, with 35 percent reporting between 11 and 15 pilots.

In terms of the AI capabilities agencies are adopting, 46 say their agencies are using AI at the edge, including applications in Internet of Things and sensor technologies. Another 45 percent say they are using AI for modeling and simulation, 43 percent are using it for robotic process automation, and 41 percent are using it for both natural language processing and machine learning.

According to the survey, 85 percent say the government needs to do more to embrace AI technologies at the edge. The most attractive edge applications include AI model training (47 percent), intelligent surveillance (45 percent), high-performance computing (45 percent), intelligent wearable technology (45 percent), facial recognition (43 percent) and intelligent video analysis (41 percent).

DIVE DEEPER: How are agencies making use of edge computing in the field?

Overcoming the Challenges to Wider AI Adoption

Despite enthusiasm for AI, agencies face technological and organizational challenges to further adoption. The survey shows that the top challenges to deploying AI models at the edge include data center-level security concerns (50 percent), power consumption/availability (43 percent), systems management expertise (43 percent), and limits on size and weight of computer nodes (42 percent).

“In addition to compute infrastructures, agencies also experience issues reducing data complexities and silos,” Brown notes. “While 87 percent say their agency has a comprehensive data strategy, just 35 percent are implementing agencywide data management and governance. AI technologies cannot work successfully without data, so it’s important that the data is easily accessible.”

Agencies cannot successfully incorporate AI into their overall IT operations without having staff with AI-related skills, Brown notes, such as data scientists or engineers to complete mission-related tasks. “These experts are in high demand, but hard to find,” Brown says. “Our research also found this to be true, as 49 percent of federal IT leaders noted that an increase in AI talent is among their top three requirements to continue moving AI forward.”

To overcome these challenges, Brown suggests that federal IT leaders should ask two questions: What is your mission outcome or objective, and what are the data sets that you have to work with?

If federal IT teams know how they would like to use AI to “solve a specific problem or develop a faster time to decision, they have to know the destination,” Brown says. “As far as the data sets, agencies have to find the needed data in those data sources and know where it’s going to come from with AI. We don’t just want static data but fresh data,” he adds.

AI initiatives are heavily focused on data science, Brown notes, which relies on high-performance computing functions and ample storage. However, he says, what often happens is that agencies make investments in software and application tools but not in their compute infrastructure.

“To support widespread AI, integration requires newer infrastructure capabilities than the traditional computing platform,” Brown says. “Investment should focus first on modernizing agency networks, upgrading storage capabilities, expanding use of cloud capabilities and strengthening the data center’s base compute infrastructure.”

RELATED: How is the U.S. Postal Service using artificial intelligence to help find packages faster?

“Once those framework capabilities are in place, the engineers, data scientists and developers can build their AI algorithms, because only then will they have the high-performing compute and storage functions to utilize AI capabilities to the fullest,” he adds. “It’s an important but often overlooked piece of the AI puzzle.”

Agencies also need to hire more engineers, data scientists and developers who can create the initial algorithms, plus cybersecurity personnel to keep all that data secure, according to Brown. “They also need to train people on data governance to ensure the data is being managed properly,” he says. “Agencies need executive support, both on the business and IT leadership level, as well as a ‘technical champion’ to support any required technology transformation.”

piranka/Getty Images