Without good data, government AI tools will just be 'garbage in, garbage out'
Vertigo3d/Getty Images
Virginia’s chief data officer recently shared what his state is doing to ensure quality data and warned that artificial intelligence is only as good as the information it has to work with.
Artificial intelligence tools—including generative AI—hold great promise for states, but without good datasets to train those tools on, governments risk “garbage in, garbage out.”
That’s according to Virginia Chief Data Officer Ken Pfeil, who noted that applications, like the chatbot recently piloted in New York City to connect business owners with information, need good data. Otherwise, he said, leaders are just going to continually make those tools “stupid.”
Generative AI and other applications rely on large language models consisting of millions of data points, with governments using that data to answer residents’ questions and connect them quickly to pertinent information.
“You've got to have good foundational governance practices in place,” Pfeil said during the Data in Action Summit last week, hosted by Informatica and GovExec, Route Fifty’s parent company. “You've got to have great curation, you've got to have the right labeling, the right tagging. You've got to have both the blocking and tackling done to facilitate quality of data. AI scares me with bad quality data.”
That is why Pfeil is eagerly anticipating Gov. Glenn Youngkin’s executive order on AI and data management, which is expected soon.
In the meantime, Pfeil has been “hitting the accelerator” on key data projects in the state since arriving in the role of chief data officer in April 2022. His tenure is tied to Gov. Glenn Youngkin’s time in office, who is barred by state law from serving consecutive four-year terms.
The Office of Data Governance and Analytics, which Pfeil oversees, is a relatively new agency, having been formed in 2021 under Virginia law. It works with dozens of agencies across state government, as well as with individual localities on its open data efforts. Pfeil said a key priority has been decommissioning legacy technology and moving to the cloud. Pfeil sees cloud computing as the state’s new “foundational infrastructure.”
Other key data projects that Pfeil and his colleagues have tackled include Operation Bold Blue Line, a collaboration with the Virginia State Police and local law enforcement agencies to use data to help abate violent crime in 13 communities. The data collected has helped agencies better allocate resources and has fed into a predictive analytics model to try and forecast future trends.
The Office of Data Governance and Analytics also has worked closely with the Virginia Department of Veterans Services, which has the aim of identifying every veteran in the commonwealth and connecting them with the critical benefits and services they are entitled to.
Those and other projects are supported by the Virginia Open Data Portal, which allows officials to visualize data, filter it and create dashboards, among other functions.“The interesting thread in both of those examples is having that desired mission outcome to drive the data side, which I think is just important to stay focused on,” Pfeil said.
A crucial focus is breaking down silos and getting departments to work together. Pfeil wants state leaders to empower their data workforce to collaborate and play to each individual’s strengths, rather than having various state agencies carrying out the same tasks or ones that don’t suit them. That is particularly important as state governments look to stay competitive with the private sector for technology workers.
“Don't pigeonhole your people into doing something that's not rewarding,” Pfeil said.
Building relationships are key, not just across state departments and agencies but also with federal and local partners. That way, Pfeil said, agency heads can be effective partners in data projects. There is a lot of data to be ingested and analyzed, they need to know “what's in it for them.”
Pfeil also noted that virtually every state technology leader is wrestling with how to implement AI into government operations, and many are unsure of how to proceed.
A survey earlier this year of state chief information officers found that more than half said generative AI will be the most impactful emerging technology area in the next three to five years, while another 20% said AI/machine learning more broadly would be most impactful.
“No one has a cookie cutter solution to any of this,” Pfeil said. “It’s learning as you go, but trying not to make egregious mistakes as you get there.”
NEXT STORY: Generative AI could save governments nearly $1.8T annually