How one Pennsylvania department uses AI to handle cases
The update was a unique opportunity to view the intersection of government and expanding AI technology.
The UC chat system helps users through a daunting process, often at a stressful time when people are out of and looking for work. New features aim to improve that process with a more natural conversation style providing better information.
The enhancements allow users to ask questions in the same language they would when speaking to a human and receive the kind of responses they might expect to hear from a human. The system can draw from previous messages to create “contextual awareness” and improve answers.
When things go wrong, the new system can seek to clarify an issue or contact an L&I employee who can better address the question. It’s also able to better understand users’ typos, awkward phrasing, or incomplete questions.
For people seeking complex or immediate assistance, interactions with chatbots aren’t always a welcome experience. As AI improves and becomes more ubiquitous, so do its pitfalls, with users receiving AI answers — often with incorrect or incomplete data – as the top response on search engines.
Found in everything from school reports to federal reports, news of the errors abounds. That’s largely because generative AI gets its knowledge base from the existing information available on the Internet, irrespective of whether we got that information right or wrong when we put it there. Generative AI also adapts its behavior to communicate with the individual user, meaning it might tell two people very different things with the same prompt.
That’s why it’s important to understand that chat systems like the ones used by the commonwealth for UC don’t use generative AI to answer questions. Instead, L&I representatives told The Center Square, they employ “proven, mature AI features.”
That assurance is important, especially as the state and countless other organizations, from medical providers to businesses, rely on chat systems to handle sensitive user data. In this case, L&I developed the upgrades internally with the Office of Administration using Google DialogFlow CX, a platform that simplifies the creation of conversational agents.
So far in 2025, L&I has Paid out more than $1.1 billion in UC benefits to 691,564 individuals, answered 524,407 helpline calls and provided 9,025 UC Connect appointments. In June, its chat system handled 7,477 live sessions.