AI is increasingly a tool of local government — but uptake may be slow across Allegheny County suburbs as local officials try to determine how this evolving technology might streamline the task of governance.
“We have so many local governments … that don’t have the capacity,” said Joy Ruff, executive director of Local Government Academy, an independent nonprofit. “They’re underresourced. There’s no plug and play right now. And so there’s a real need for local governments to acknowledge that it’s out there and then to come up with policies that govern it.”
Pittsburgh’s rich tech sector is pushing the city and Allegheny County ahead of the curve to adopt the technology, but some of the county’s 129 other municipalities don’t yet have policies regulating their use of the technology, and some confirm that they don’t yet have the guidance they need to safely use artificial intelligence tools.
Dwight Boddorf, borough manager for Tarentum, said that AI can be “tremendously helpful” for a small borough with limited staff.
“Smaller communities are often expected to provide the same level of communication, compliance, documentation and service as larger governments, but with far fewer staff. AI can help make us more efficient, but only if it is used with safeguards,” said Boddorf. He said he may use AI to review the borough website and flag potential ADA compliance issues or to compare permit forms with updated ordinances. He also praised the City of Pittsburgh for being “proactive” in discussing AI guidelines.
A statewide survey of government officials in Pennsylvania conducted in May and June indicated that 81% of respondents have no Generative AI policy at all, of which 62% agreed it would be valuable to have one.
A team at the University of Pittsburgh Institute for Cyber Law, Policy and Security conducted the survey via the Local Government Academy to determine how many local leaders, especially in Southwestern Pennsylvania, were thinking about responsible AI use. Respondents were from predominantly smaller municipalities outside the city, with most serving populations between 5,000 and 10,000 residents.
Beth Schwanke, executive director of Pitt Cyber, said that concerns about data privacy, lack of human connection, confidentiality, bias and legal implications came up in open-ended responses from survey respondents.
“A lot of places don’t have any guidance right now, or are even actively telling employees not to use AI, which I don’t think is a very effective tactic and frankly probably counterproductive,” said Schwanke, who is advising some local leaders on how to implement AI responsibly so as not to put residents at risk.
Small towns consider big tech
Allegheny County has the second most municipalities of any county in the United States after Cook County, Illinois. This makes the role of local government especially central in this region.
An Lewis, executive director of Steel Rivers Council of Governments, which aims to support municipalities in the Mon Valley, agreed with Boddorf that AI could make the lives of some local government officials easier. Yet she said that since it’s such a new technology and many officials have no knowledge of how it works, not everyone is adopting it at a rapid rate.
“The wheels of government tend to turn slower, even local governments,” said Lewis. She said most of her colleagues in the Mon Valley are probably using AI to a limited extent, but that it feels “foreign” because it’s a new technology.
Scott Craighead, manager of Elizabeth Township, said that while his small township didn’t have an AI policy in place he was curious to see what other similar sized governments might be doing.
Ryan Delaney, council president of Plum Borough, said that as of right now Plum does not have any official AI guidelines. He said the topic was brought up recently in a staff meeting because a lot of other governments are looking at it. He said he and his colleagues are taking the time to “get it right.”
Delaney said he views AI as “a tool but not the product.” He also emphasized human review, claiming “we have to keep demanding that human element.”
Schwanke said that while AI can be helpful in governance, officials must be ready to train their employees and have the right guidelines in place to protect their residents. She also wants to make sure there’s “room for experimentation and innovation,” with transparency toward residents on how AI is being employed.
To her, transparency looks like deciding to publish use cases or having an AI registry where county residents can know what AI is doing and how it impacts them.
“You should as someone in government have a higher burden, have a higher level of responsibility toward your residents,” said Schwanke.
The city provides training to smaller municipalities and boroughs
Crafting AI guidelines has been crucial to the work of Chris Belasco, chief data officer for the City of Pittsburgh, and Andrew Hayhurst, the city’s senior manager of innovation. They’ve shared their insights with other local municipalities through their Local Government Academy webinar. The webinar aimed to provide a framework for other municipalities to use AI responsibly instead of delaying adoption.
Boddorf said the webinar “helped frame AI as a governance issue, not just a technology issue.” He said it reinforced many of the themes from the State of Pennsylvania’s AI pilot program using ChatGPT Enterprise and the related report the state issued last year.
Lewis pointed to the webinar as a resource she encourages her colleagues in the Mon Valley to use when thinking about how to incorporate AI into their work. She said it’s the only training opportunity she’s seen so far for local governments in the region to learn more about AI. She explained that the trainings are specifically designed for Southwestern Pennsylvania where lots of small local governments formed along the rivers with the growth of the steel industry.
“That post-industrial legacy really plays out a lot in our cultures and in the needs and the challenges that our local governments face that are different from some of the places in the center of the state that are very rural,” said Lewis.
Pittsburgh: ‘Tedious tasks’ with human review
Beyond providing their framework to others for the Local Government Academy, the City of Pittsburgh is also part of the national GovAI coalition of local, state and federal government officials who are “united to promote responsible and purposeful AI in the public sector,” according to its website.
Belasco and Hayhurst pointed to cities like San Jose and Long Beach in California as leaders in using AI responsibly in the public sector.
Long Beach advocates for citizen voice in the AI implementation process through programs like the Long Beach Collaboratory which invites community members to work with city staff to come up with technological solutions.
“The programs that they do are brilliant. And so that’s probably the city that I look to the most and try to talk to the most,” said Hayhurst.
Blaine Kuehnert, a CMU PhD student focusing on responsible AI, said that while San Jose is “at the front of the pack,” Pittsburgh is “not far behind” in creating AI guidelines. He believes governments need to be intentional about implementing AI instead of using it as an easy fix to a problem.
Belasco and Hayhurst emphasized that AI would only be used for “tedious tasks,” not to replace city workers. They said that use of AI for administrative tasks must include human review.
Currently they are testing a model of a ‘Citybot,’ a version of 311 that would help city residents file certain requests. Hayhurst said it would “prompt them to get better answers so that we could get that information to our operations faster.”
They are also looking into using generative AI tools in the process of permit intake, emphasizing that they are not interested in replacing any plan reviewers.
“We’re trying to figure out how to enable their work to happen more productively. It’s not a replacement strategy,” said Belasco.
Belasco and Hayhurst also stressed the importance of being selective with the kinds of vendors they are using. They said one of their guiding principles when thinking about vendors is to protect city data and the privacy of city residents. The city uses Microsoft CoPilot to ensure all data is situated only within the city’s unit of Microsoft.
“If residents are trusting city employees with data, we’re asking our employees to use our enterprise tool so that it does stay within the boundaries of our environment,” said Belasco.
When they do disclose data, Belasco said it goes through a rigorous review process and the Western Pennsylvania Regional Data Center.
Allegheny County: Emphasizing transparency
While Allegheny County has an AI policy in place, they are working to develop an AI operations manual to help their staff better understand responsibilities and guidelines related to the technology. They expect to release it publicly within the next few months, according to County Manager John Fournier.
“If there isn’t a very clear case where this is going to make things better for people in the county, we’re not going to use it,” said Fournier.
Fournier said he hopes being transparent with residents about how county agencies use AI will help establish trust that “we’re doing good work and we’re making good decisions.”
He doesn’t want to see AI replacing human involvement in governmental processes, but rather views AI as a tool that can help Allegheny County employees do more in their work.
“Government has to involve people and I think the leadership in Pittsburgh and Allegheny County deeply understands that,” said Schwanke, who is advising the county.
The national stakes: Concerns are more than ‘friction’
Research and experience nationally has shown that relying on AI in a government setting can put citizens at risk.
A 2024 report by TechTonic Justice, for example, describes how most low-income people in the United States receive their benefits from systems that are highly reliant on AI. Residents in need may be falsely denied benefits they were entitled to because the decision was offloaded to an algorithm or AI with no human review.
The report lists examples of states where AI has harmed vulnerable citizens. In Arkansas, these systems cut the in-home care of 4,000 disabled people. In Michigan, 40,000 people were falsely accused of unemployment insurance fraud. In Texas, 4 million people would potentially lose health insurance because of a “labyrinthine” Medicaid enrollment system, according to the report. In each of these cases, according to the report, no human review was employed, leaving vulnerable citizens struggling to receive the resources they needed.
Kuehnert listed other examples such as a case in which a risk assessment tool, COMPAS, was used in the judicial system to inaccurately determine a defendants’ likelihood of recidivism.
He also pointed to a pre-AI predictive risk model that was applied to a family screening tool in Allegheny County in 2017. A paper from the ACLU stated that tools like this “are created with arbitrary design decisions, give families no opportunity for recourse, perpetuate racial bias, and score people who may have disabilities as inherently ‘riskier.’”
Alex John London, a professor of Ethics and Computational Technologies at CMU, pointed to cases like these as examples of where use of AI could be more harmful than helpful. He said that responsible AI depends on making sure the interests of the people are represented. Much of this depends on the goals of political leaders, he said.
“Whether we use those things well or badly is really a function of how well we understand our tools and the ends to which we’re committed. That’s less about the AI and more about what the goals of our leaders are and what ends they use these tools to pursue,” said London.
AI is unique in how fast it’s evolving and the scale at which it is being deployed, according to Elise Silva, director of policy research at PittCyber. It comes with cybersecurity risks that are unprecedented. Yet, she wants local governments to realize that big technology changes have always impacted governance.
“We have lived through great change and we have adapted. Some of the governance structures that we have in place can work now, but they need to be adapted to the unique risks of this particular technology,” said Silva.
London noted that with the invention of the internet, many government processes that relied on paper and pen became digitized. In many ways, this increased the speed of things. He doesn’t view AI as different from other innovations. He is more concerned with how people are thinking about the consequences of AI as “friction.”
“What frightens me the most is when people think that [discussion of] ethical and responsible AI is just friction — that we’re just trying to hold people back from the grand future,” he said.
“What I would say is ‘No, what I’m trying to do is ensure that the future that we reach is a future that we want.’”
Ada Perlman is an editorial intern at Pittsburgh’s Public Source. She can be reached at ada@publicsource.org.
This story was fact-checked by Cilia Catello.
MOST READ STORIES
Republish our articles for free, online or in print, under a Creative Commons license.
