When the AI boom hit a few years ago, there were, naturally, concerns that bots would replace human workers, but good news: Artificial intelligence is instead reshaping the way government workers do their jobs. It’s clear that AI can automate routine tasks and take the burden off thin-stretched state and local teams in a variety of other ways. But this is a work in progress, and there is still much to be learned, so maybe don’t pop the cork on the champagne just yet.
In addition to functional questions — what is AI actually good at? — the human-capital issues persist. People are less worried about AI taking their jobs than they once were, but the fear isn’t altogether gone.
HOW AI CAN HELP
The positive potential for AI is becoming rapidly apparent. In Maryland, for example, it is already changing the way people work.
“We’ve seen great examples so far of AI’s ability to assist with routine daily workflows: Things like drafting emails or summarizing content,” said CIO Katie Savage. “We’ve also seen a number of agencies stand up chatbots, which can take the burden off of some of our end-user support or customer service centers.”
Georgia likewise is reaping the benefits, using AI to assist with daily tasks and helping the state do the proverbial “more with less.” AI is supporting cybersecurity operations, for example, while virtual assistants are streamlining how unemployment applications are processed.
Government Technology/David Kidd
Overall, AI is proving that it can deliver “efficiency, accuracy and capacity,” said CIO Shawnzia Thomas. It’s freeing people from rote tasks, “enabling them to focus on judgment, empathy and innovation, the things that humans do best.”
New Jersey, meanwhile, has been running an AI sandbox for a little over a year, and about 20 percent of state workers actively use it. Chief Innovation Officer Dave Cole surveyed regular users and found that most use AI at least weekly, and some daily. “It’s saving them in the range of hours a week of time spent on tasks that they would otherwise have to complete,” he said.
People are using AI to draft and edit emails, memos and reports, and they’re tapping it to summarize documents. “Government has large inputs — public feedback on a rulemaking process, for instance,” Cole said. AI can “pull something actionable out of that, and do so in a consistent, fair way.”
All this must be handled with great care. In Maryland, for instance, the IT team moved early to block the use of third-party transcription software. “We are acutely aware of all of the opportunities for vendors to own or exploit our data if we’re not careful,” Savage said. To keep data secure, the IT team has a senior adviser for responsible AI on staff and developed a Responsible AI Policy.
As the benefits of AI are coming into focus, experts point to a number of other important caveats from a workforce perspective. AI can help HR sort through mounds of resumes, for example. But it’s also producing resumes at an alarming rate. That’s problematic, said Mary Guy, a professor at the School of Public Affairs at the University of Colorado Denver.
“The letters of interest look fabulous, the resumes look great, and now they’re finding that when they call in the most promising applicants for interviews, the reality is nowhere close to what was presented in writing,” she said. For HR teams, the AI is speeding up the processing, but slowing down the process.
And when it comes to ways in which chatbots impact the workforce, Samantha Shorey expressed real concern. She’s an assistant professor at the University of Pittsburgh and a fellow at the Roosevelt Institute where she co-authored its paper AI and Government Workers: Use Cases in Public Administration.
AI-powered chatbots can answer repetitive questions, taking out the routine labor. That’s nice. But it leaves government workers tackling more and more hard questions. Callers are frustrated by the time they reach a human, and the problems are harder to solve.
“They are handling the ones that are too complex for chatbots to navigate,” Shorey said. Routine work gets automated, but workers potentially burn out faster. It’s just one example of the ways in which AI as a workforce-augmentation tool remains a work in progress.
FILLING THE SKILLS GAP
Even as these concerns play out, it’s clear to many that AI can help fill certain state and local workforce gaps.
At a high level, “Government is facing budget cuts, hiring freezes and pressures to do more with less,” Thomas said. “AI can play a critical role in helping to stretch our limited budget, helping protect critical services, and reducing fraud and waste and abuse.”
In a tight fiscal environment, “we’re giving them some tools to relieve the burden, and allowing them to focus on the citizen needs,” she said.
AI can also help to fill the gaps in areas where human talent is hard to come by. For example: Legacy systems remain mission critical, but programmers skilled in COBOL, for example, are increasingly rare. “AI can really help in maintaining these legacy systems that still carry this old code,” Thomas said. Data scientists likewise are pricey and hard to come by. “We can’t afford to bring them in, and if we do bring them in, the private sector will steal them.”
AI helps here. “It can automate parts of the data cleaning, the categorization, giving analysts these actionable insights without them having to be data experts,” she said. “A lot of data science is just aggregating the data, cleaning it up, making it useful. AI can do a lot of that lift for you.”
Government Technology/David Kidd
In areas like cybersecurity, Savage is coming to see AI as a must-have when it comes to things like reviewing security logs. “Some of these data sets are so large, it would literally be not humanly possible to do some of these things,” she said. “It’s one thing to look at malicious actors here or there, but with AI you can start to put together a story over time. Where are these IP addresses located? What’s the frequency? Are they hitting other states?”
AI likewise supports Savage’s efforts to shed redundant software and streamline digital resources across agencies. In pursuit of an enterprise-level IT strategy, “I do need to understand those products,” she said, and AI can mine product documentation to help determine what stays and what goes.
Even when AI isn’t doing the super-human (i.e., scanning millions of security events), its ability to automate routine work represents a skills-gap win, in Cole’s view.
“In New Jersey, like in other states, it’s hard to get the funding to have the workforce that we need to meet the responsibility,” he said. “If these tools can help smooth out the parts that are a little bit routine, and then free [workers] up to do the higher-level work, that really helps to fill in some of those gaps.”
JOB-REPLACEMENT CONCERNS
Workers ought to be excited about all this, and the high usage rates in the New Jersey AI sandbox suggest that many are. But concerns linger.
The Roosevelt Institute paper on AI and the government workforce points to the moment when the federal Department of Government Efficiency (DOGE) cut 1,200 workers from the Centers for Disease Control (CDC) and Prevention rolls, suggesting they could be replaced by AI. With that kind of talk at the federal level, it’s likely that state and local employees will worry about AI taking away their jobs.
As Shorey sees it, people are still concerned about this, but their reasons are changing as the initial use cases unfold.
Initially, people worried about AI “having the ability to outperform human beings at no cost,” she said. Now they’ve seen that AI does not necessarily do the job any better — they aren’t being outperformed. They’re worried instead about AI being used to justify downsizing the workforce, just as DOGE did in the CDC example.
AI as a technology “cannot be separated from the way that we talk about AI, the hype machine around it, and the way that AI is used within larger social forces, to achieve other goals,” Shorey said. In the current political environment, where those goals may involve government shrinkage, employees fear “AI being used to legitimize downsizing.”
Whether and to what degree people still fear AI will depend to some extent on their job responsibilities. “The more sophisticated the skills are, the slower it’s going to be for AI to replace that worker,” Guy said.
“If a worker can really only do one thing, and it is the simplest, most programmable of tasks,” they may have good reason to worry — either that AI will replace them, or that they won’t have the skills needed to keep up with a changing role.
“The more capable a worker is, the more educated they are, the more they are able to look around the corner and find other tasks they can do,” and therefore the less likely they are to be concerned, she said.
In Georgia, Thomas has looked to keep the fear at bay by involving workers directly in the AI adoption process. “I’ve told my team: Look at the work that you do and please tell me something that you do on a repetitive basis every day, all day. Tell me what AI can help you with,” she said.
“When you put that in their hands … they’re relieved to say: You know what? I do need some help. I can use a tool to help me with my daily job,” she said. That buy-in is crucial. “You don’t want your folks fearful of their jobs being eliminated because then they start thinking about that, and the work they’re doing starts to dwindle down. They say: ‘Well, I may not be here.’ You don’t want that.”
Clear messaging helps too. If people are still worried about AI replacing them, “that’s something that leaders in any organization need to be very careful about,” Cole said. “It’s been clear from our leadership here at the state of New Jersey that that is not the purpose. That is not our mission. Our mission is to deliver better services and to do so efficiently and effectively. So we have not been focused on job replacement.”
To make that promise real, “we look for cases where we can fill in those gaps in the staffing that already exist — that we haven’t been able to fill just due to the realities of budgets and some of the funding dynamics,” he said. These use cases help to demonstrate leadership’s commitment to leveraging AI as a workforce augmentation tool, rather than as a replacement mechanism.
In Maryland, Savage takes much the same approach. “One of the experiments we’re doing is in the licensing and permitting space. It’s a very manual, not user-friendly process to get a business license or permit in the state,” she said. “That could be a really awesome opportunity to leverage a chatbot.”
If that chatbot could take some of the pressure off a state worker, without any threat of the bot replacing the human, she suggested, that can help alleviate any lingering fears.
WHAT’S NEXT?
As today’s early AI implementations come to life, state IT leaders and other experts point to early lessons learned that can potentially help guide state and local efforts going forward.
Guy suggests we are at an inflection point, where the workforce would be well-served if leaders were to pause and reset the expectations. “The people who were absolutely most enthusiastic about AI and could see it doing so much two years ago are now pretty disappointed with what they’re seeing,” she said. On the other hand, “the ones who two years ago said this is never going to happen, now they’re surprised because it is happening. There are more programs, there are more applications for it.”
As people recalibrate their expectations for AI, leaders need to proceed thoughtfully.
“AI will take on some functions and tasks, not the entire job,” Thomas said. “AI is best at tasks within jobs that are repetitive, rule-based, with heavy data. This means that jobs should not disappear overnight — but the roles should be redefined.”
As leaders work to redefine those roles, “workforce planning should be focused on task reallocation, rather than job elimination,” Thomas added. That in turn will likely mean more training. “That’s another workforce implication: You have to upskill and reskill.”
That means skilling people for new roles, and also training them to be effective with AI. “That’s an urgent task right now,” she said. “They need to know how to use the tools.”
That’s been a big focus in New Jersey. “One of the most important things for us was to not only provide tools, but to start with training and make sure that the training and the tools are highly coupled,” Cole said. To that end, the state developed an AI workforce training program with InnovateUS that is now used freely by about 25 states and localities. Some New Jersey agencies require the training, while others strongly encourage it, per an order from the governor.
The key is to make that AI training authentic. “Make sure that as you go through that training, there are real-life examples and prompts so that you can pull up the AI tool and try it out,” Cole said.
“That experiential learning part is really important because there are things that everybody using these tools needs to be careful of — around biases, hallucination, understanding how much information you can give it at a time,” he said. With such context-specific training, “we see a corresponding increase in utilization of AI-related tools.”
In the big picture, the impacts of AI on the government workforce will be determined not just by how state agencies handle the new tools, Shorey said, but also by the legislative guardrails.
There’s emerging legislation to determine how much decision-making AI can do on its own, without human oversight. That has direct implications for the workforce. Will human workers be seen as “the most important aspect of completing a task, because they’re the thing that a computer can’t do?” Shorey said. Or is that worker seen as “the final turn of a screw, just finishing work?”
In the latter case, that human employee potentially “becomes less valuable within the labor process,” she said.
One final thing worth noting for IT leaders is the ways in which AI is changing their jobs.
“I may be the last CIO to manage humans only, or the first CIO to manage humans and agents. I’m super excited about that part of it, being able to manage some agents along with humans,” Thomas said. And she added: “I am learning, just like everybody else. This is new.”
