How Federal IT Leaders Are Considering Effective AI Adoption

How Federal IT Leaders Are Considering Effective AI Adoption

It's critical to identify business use cases, sharpen data quality and governance, and work with end users to help improve AI adoption.

Federal agencies are increasingly looking to harness artificial intelligence within their core services. One of the major challenges toward adopting AI is successfully integrating it within core services, and advocates for automated technologies have put together key steps to help smooth the process.

Agencies are currently focusing a significant part of their AI adoption in low-value, administrative tasks so they can prioritize cognitive, creative and other high-value work across the federal workforce. Amid this sentiment, many AI advocates across government are identifying where AI can best fit business values and building initial efforts from there.

“It’s the tiny things that, at scale, make a ton a difference — making sure that a forum gets to the right office, making sure that we’re able to track and capture what kind of questions we get asked most so that we can constantly update our FAQs in our chat bots with those answers,” GSA Centers of Excellence Federal AI Implementations Lead Anil Chaudry said at a Digital Government Institute event Thursday.

But even with the business use cases in mind, Department of Health and Human Services AI Program Lead Sanja Basaric added that most AI projects tend to fail, largely due to bias in data, algorithms or the teams managing them. Basaric added that it is therefore critical to ensure that the data initially used in algorithms, as well as the governance that surrounds data, is shaped for successful AI adoption.

“Data is not information, and information is not insight, so when you understand where the data resides, who owns it, how to make it machine readable, how to standardize it, and then what are the data-sharing policies that govern it — if you look at a lot of the policies around data in the government, they were written before AI was even something we could fathom,” Basaric said.

Dynatrace Public Sector Chief Technologist Willie Hicks seconded these assertions, adding that preparing data for proper algorithmic application is essential. Without it, even the output of the most sophisticated and finely-tuned algorithms will remain faulty or biased.

“There’s an old adage, put bad data in, bad data out,” Hicks said. “If I’m analyzing bad data, I’m just going to get bad results, no matter how good my algorithms are. Then, I say start small. You can’t boil the ocean. Don’t start with some massive project. Start off with a small, easy-to-implement project, get to dip your toe in it, and then kind of move from there.”

Chaudry added that taking a “low-level building blocks” approach to starting projects around AI can also spot potential vulnerabilities or blind spots in advance. He said that most AI use cases in government right now are smaller projects that when scaled could significantly streamline and modernize an array of core business processes.

Basaric seconded these assertions while emphasizing the importance of education and training to get federal workers comfortable with using AI for lower-value work processes.

“This is going to require a major cultural shift across the entire government,” Basaric said. “We have to revisit enterprise data policies. We need strong foundations that consider the entire lifecycle and that define the business case before we even begin the AI, and a lot of this actually begins with training and education.”

Basaric touched upon the HHS Trustworthy AI Playbook, adding that if HHS and other agencies foster a culture that makes AI a “team sport,” this approach will make AI a transformational tool across across major business lines.

Since proving return on investment for different AI applications is another common challenge that agencies face, Basaric added that a culture shift and push for AI-forward education and training can help usher in further support for various aspects of AI development — from making data AI-ready to the development and maintenance of AI itself.