Pennsylvania is pushing for expanding the use of artificial intelligence tools in government agencies, and local governments will follow suit.
For local and state agencies, it is a way to mitigate workloads by directing artificial intelligence to complete routine tasks that otherwise take hours to cost tens of thousands of dollars.
As Pennsylvania leaders aim to set the state as a hub for AI and data centers, the federation is one of the first states to consider using generative AI across state governments.
The state launched its year-long generation AI pilot program last January, with 175 employees from 14 agencies incorporating ChatGpt Enterprises into its work.
“The most common use of ChatGpt among pilot employees was tasks like assistance, research, brainstorming, and summarizing large amounts of text or data. All of these can be widely applied to our entire workforce.”
The state paid $108,000 for the CHATGPT Enterprise license, training and support for the PILOT program.
Openai cited the Pennsylvania pilot program in a press release announcing the ChatGpt Gov, which was specifically designed for government use.
The majority of pilot employees reported positive experiences, saving an average of 95 minutes per day. The results encouraged the state to make more artificial intelligence.
“We want to provide more employees with responsible access to AI-powered tools if it helps employees work more effectively and efficiently,” Egan said. “We are currently looking for ways to expand access to these tools to our employees.”
He added that training on “safe and responsible AI use” is a prerequisite for state employees to use AI tools.
Double check AI work
Experts warn that AI, particularly generative AI (where the video is output based on text, images, and even user-entered prompts), poses many risks and ethical considerations.
Generation AI tends to create falsehoods and present them as facts. This is a process known as hallucination.
“We need to understand that generative AI can be incorrect in particular,” says Kolgesner, responsible AI program manager at the Technology and Society Block Center at Carnegie Mellon University. “Do you have to treat me like a summer intern? You need to reaffirm the work.”
The Block Center has advised the state on AI implementation in its pilot program and has worked with government leaders in Pennsylvania.
Currently, the Human Services Bureau is considering using AI.
“The Human Services Bureau is exploring opportunities to use AI responsibly to promote efficiency in promoting employee-first approaches,” said DHS spokesman Brandon Kwarina. “Like new technologies, we have policies and oversight in place to guide responsible use and we continue to appreciate any progress that may support our work.”
Currently, Commonwealth's AI policy prohibits employees from making decisions using generated AI, requiring users to review and verify what has been created by generated AI. It also prohibits private data or information in the generation AI tool.
All state IT policies are reviewed annually, and Commonwealth's AI policy is expected to be reviewed in March.
Allegheny County is working on policy
The state has policies in place, but most local governments do not have specific guidelines on how to use AI.
Allegheny County is the exception. This year, they are developing AI policies in the hopes of finalizing it, said county spokesman Abigail Gardner.
Gov. Josh Shapiro signed an executive order in 2023 to establish standards for the use of generated AI in the state government. We also created a Generation AI Management Committee to oversee policy and implementation.
In March he created a generative AI labour and management collaboration group that brings input from employee and worker representatives when considering AI use in government.
“The Shapiro administration consistently emphasizes that generative AI is not a substitute for the knowledge and experience of federal employees,” Egan said.
Additionally, while the statewide movement to incorporate generative AI is new, AI has been used by governments for many years. Climate modeling, for example, relies on machines that find patterns and analyze large datasets, said Cole Von Glahn, senior technology policy analyst at the Centre for Environmental Policy Innovation.
“More of what we do in the world relies on deep learning techniques that have existed for a very long time,” von Grahn said. “I think the generative AI side of things is much more experimental. You can see ways to help in government. Now I'm not sure those methods outweigh the risks.”
From permits to housing applications
For some elected officials, artificial intelligence is a promising tool to streamline the bureaucratic processes that are particularly permitted.
“If you consider applying for a permit, it's going to be the same 12 boxes that everyone has to fill out. There are rules on how these are analyzed by people,” Von Gran said. “You can translate that into a way of looking at the machine.”
R-Cecil Representative Jason Ortitay hopes to launch a pilot program at the Environmental District office to test how AI is used to process permit applications.
“It has to be maneuvered, tested before being adopted on a much wider scale, and then trusted,” he said. “What you don't want to happen is that you implement it and it causes all sorts of chaos because that's not the whole point. The point is to make it better, streamline it and make it more efficient.”
He will discuss it with DEP and Shapiro's offices once the state budget is finalized.
The city of Pittsburgh housing authorities are also considering using AI to streamline processes, particularly in residential applications.
HACP has finalized a contract with Bob.ai, an AI company affiliated with public housing authorities, and has implemented AI to process the recertification. HAPC has around 5,100 tenants, most of which are recertified every two years.
Hacp Senior IT Director Monty Ayash said the one-year pilot program is likely to launch this year. HACP approved a payment of approximately $160,000 for AI in June, estimated it would reduce processing time by up to 50% and backlog by up to 75%.
HACP also launched a one-year pilot for Google Gemini in April for 60 employees. So far, employees have mostly used it to optimize communications, such as drafting internal documents and suggesting improvements to emails, Ayyash said.
“We're adding a very reserved AI component,” he said. “We don't want to stop getting the job done in place of AI learning as we monitor all sorts of behaviors that could really show we can really challenge the actual daily business operations our employees are used to doing.”
Abigail Hakas is a reporter for the Next Generation Newsroom, part of Point Park University's Media Innovation Centre. I'll reach her abigail.hakas@pointpark.edu.
NGN is a local news service focused on government and enterprise reporting in southwestern Pennsylvania. Find out more about foundation and corporate funders here.
