A panel of experts says when it comes to state government and artificial intelligence, the simpler forms of the technology are best.
The high-profile generative AI technology, which can create text, images, video, audio, code and other media, doesn’t have much of a role—yet.
Generative AI “can generate great images of a cat riding on a porpoise, if that’s what you want,” says Nick Mastronardi, founder and CEO of Polco, as he flashes that image on the screen during an NCSL Base Camp 2024 session. “But for most of us in our workflows, what we’re really interested in is getting great data about our communities to help us with strategic plans or apply for grants or allocate grants to our local governments.”
“We are still at a point where we are probably going to be in a state of inflated expectations for these technologies, and we’ll just have to be careful about separating the hype from the reality.”
—Shayne Kavanagh, Government Finance Officers Association
Mastronardi says using AI to guide that decision-making depends on having vetted data—and more is better.
“There is so much data out there that we can and should be leveraging to make great policy and programmatic decisions to help our budgets, to help our constituents, to deliver public services really efficiently,” he says.
But as usual when discussing AI, the praise is followed by a caution.
“There are things that we have to be really careful of on this journey,” Mastronardi says. “We need to be careful about where we put our data in order to leverage the power of AI with it, because if you put the wrong data in the wrong place, that can be an irreversible decision.”
Mastronardi notes that AI can draw on vast databases managed by states, cities, towns and the federal government to track all aspects of a state’s financial picture. It’s vital, he says, that states draw on reliable data and keep it secure.
Even with good data to work with, he says, AI can’t fully replace people.
“We’re getting close to being able to say, ‘Hey, AI, help build me a strategic plan or economic development plan, or build my budget,’” Mastronardi says. “It can help you get from zero to 70% or 80% for some of the most important elements in this policy life cycle. But a data scientist can really help you make sure you extract the right insights from it.”
Help With Cost Analysis
Shayne Kavanagh, senior manager of research for the Government Finance Officers Association, says AI can be used to assess fee structures so they can be set where a maximum number of citizens can afford to pay to limit uncollected fee payments.
AI can also track the quality of roads, bridges and other infrastructure to determine how much life is left in them and when it would be most cost effective to invest in repairs.
It can do that by drawing on visual data collected by cameras on state and police vehicles and by automated video inspection companies and compare that with the age and materials of the structures to come up with recommendations.
That can help budget managers, he says.
“Don’t spend too early but don’t wait until the deterioration is so great it costs five times as much as it would have,” Kavanaugh says.
Kyle Wedberg, senior manager for research and consulting with the Government Finance Officers Association, says if states do comprehensive statewide assessments of infrastructure it “could help localities that may be behind a learning curve optimize that sweet spot, find that place where the investment of dollars is maximized.”
University of Nebraska professor Craig Maher, director of the Nebraska State and Local Finance Lab, studied how AI could help four very small towns assess their financials and forecast their budget strengths and challenges. He used 15 years’ worth of the towns’ financial reports and examined tax revenue, debt and budget obligations, then asked ChatGPT to weigh in.
“You get some broad policy recommendations,” he says, including suggestions about stabilizing the budget, growth opportunities and enhancing reserves. The ideas “were relatively vague, but at least it offers a starting point for having these conversations.”
He reinforced Mastronardi’s view that humans are still critical to the process.
“Lots of institutional memory needs to be applied to these general themes that ChatGPT has found with these data,” Maher says.
Wedberg says AI can enhance research on the cost of proposals from legislators. He notes it’s a big investment of time and staff to assess financial impacts, but AI can digest large amounts of data to get a clearer picture.
“Thinking about how you assign revenues and derive taxes based on better information can be wildly reformative for state budget processes,” Wedberg says.
Kavanaugh says that so much about AI is in the unrealistic part of what’s been dubbed the “hype cycle,” a phase any new technology goes through. He cites the onetime prediction that every home would have its own 3D printer.
“We are still at a point where we are probably going to be in a state of inflated expectations for these technologies,” he says. “And we’ll just have to be careful about separating the hype from the reality.”
Kelley Griffin is a senior editor in NCSL’s Communications Division.