Four years after the pandemic overwhelmed Nevada’s unemployment insurance system, the state’s employment agency still has more than 10,000 outstanding appeals, and about 1,500 of them are from during the pandemic.
But this summer, agency officials are hoping to drastically accelerate the appeals process through a new tool: artificial intelligence.
In July, the Nevada Board of Examiners — a panel composed of the governor, attorney general and secretary of state — is set to approve a contract for the state to begin using Google’s AI technology to expedite how appeals for unemployment benefits are processed, said Christopher Sewell, the director of the Nevada Department of Employment, Training and Rehabilitation (DETR).
Under the arrangement, a transcription of a virtual appeals hearing would be transmitted to Google-run AI servers that are familiar with the state’s unemployment appeals policies. The AI would then issue a ruling on the appeal and send the decision to a state employee, who would check the technology’s work and ensure that the decision was correct.
Under DETR’s current processes, an employee would review the hearing and issue a written report based on state policies. This process takes an average of three hours to complete, while the AI technology can issue a ruling within five minutes, said Carl Stanfield, DETR’s information technology administrator.
“The time saving is pretty phenomenal,” Stanfield said.
Artificial intelligence — defined as machine-based systems that can make decisions, predictions or recommendations — is still in its infancy in Nevada state government, but some agencies have begun using the emergent technology to streamline processes and make services more user-friendly, officials said.
The Silver State Health Insurance Exchange, the agency that operates the state’s online health insurance marketplace, began using a virtual AI agent last November to handle more basic questions over the phone, freeing up employees to address more complicated user questions and lowering wait times. The Nevada Department of Motor Vehicles has also been using an AI-driven chatbot since 2022 to answer user questions, and has plans to unveil a more advanced chatbot in the future.
AI has boomed during the past couple of years, particularly with the advent of tools such as ChatGPT. Nationwide, governments have used the technology for wide-ranging purposes, including bill drafting, customer and worker assistance, transcribing, translating, processing unemployment claims and monitoring traffic. Some states, including Nevada, have instituted AI-related policies or created new bodies specifically designed for regulating AI.
But the tools have also been met with legislative concerns.
In the past two months, two interim legislative committees held hearings on the use of AI in government. Some legislators appeared wary of the technology, particularly about it becoming too powerful, as well as who would be liable in the event of malfunctioning technology.
Sen. Skip Daly (D-Reno) was particularly outspoken on the topic, remarking, “Are we out of our ever-loving minds?” after a presentation on how other state legislatures use AI.
In an interview, Daly acknowledged that AI is here to stay and there may be uses to improve government efficiency, but he stressed that the state should take a careful approach, such as by prohibiting AI from transcribing state depositions.
“I’m just dubious of the whole concept of overreliance on algorithms and computers,” he said. “I hope that we are cautious about it, and think before we just say ‘we got to be faster or better than the next guy.’”
Nevada’s Office of the Chief Information Officer is responsible for overseeing AI use in state agencies. The office reviews agency proposals of AI in a similar process that’s already done for approving large technology projects.
In November, the office released guidance on agency use of AI that said the tools must adhere to all data privacy laws, strict measures must be in place to limit access to AI tools and data in AI technology must be encrypted. The memo also said non-state operated AI tools should be used “with extreme caution.”
A more expansive policy is also in the works, though a date of completion is unknown, Timothy Galluzi, Nevada’s chief information officer, told The Indy. This policy will result from consultations with the private sector, higher education institutions and the nonpartisan Guinn Center, as well as a review of other states’ AI policies.
“Our goal is to walk a really fine line here, where we are protecting state data and infrastructure while still allowing the innovative use of these technologies to create a more efficient and effective state government,” Galluzi said. “There’s a lot of opportunities that these tools can bring us, but those opportunities don’t come without their fair share of risks.”
Although only a few agencies have taken the steps to integrate AI into their daily operations, state officials emphasized the technology is all but ensured to become more prevalent across state government.
During a late May meeting of the Joint Interim Standing Committee on Government Affairs, leaders from the Office of the Chief Information Officer (which is housed within the governor’s office) called on legislators to provide sustained support and funding to ensure AI-related policies are implemented effectively.
“This is going to be a pretty hot topic,” Galluzi said.
Other uses of AI
Silver State Health Insurance Exchange unveiled a virtual AI agent last November at the start of open enrollment. It was the first state-level health agency to receive approval from the Centers for Medicare and Medicaid Services to use an AI chatbot.
The agency contracts with a call center that typically receives two types of inquiries: basic questions (that may have to do with account access) and more complex questions, such as those addressing program eligibility that could require supporting documentation, said Russell Cook, the agency’s executive director.
The virtual AI agent is able to assist callers with basic questions, such as wanting to get in touch with an insurance agent, allowing employees to focus on the more complex inquiries. The agency’s quality assurance team also reviews the virtual agent’s interactions and can report problems to the vendor.
The virtual agent addressed 15 percent of calls during the open enrollment period. In the first four months that the virtual agent was active, the average customer wait time dropped by 20 percent compared to the year prior, while the length of conversation increased by 30 percent, which Cook said was likely because employees were taking fewer basic calls.
The virtual agent is particularly helpful in assisting callers outside of business hours, Cook said.
“We’re really looking primarily at AI as offering a supplement to our call center agents, which is not intended to replace customer service, but is intended to provide more responsive customer service to customers with basic needs,” Cook said.
The agency is also piloting two other AI initiatives: one to be able to automatically process documents and another for employees to be able to use a chatbot to answer basic procedural questions.
Three other initiatives are in the works, including a bot that will automatically listen to and score all call center interactions to track employee performance. The agency is also building out technologyrobots to make policy documents searchable and update those if necessary with new federal guidelines, as well as creating an interface to answer employees’ policy-related questions.
The Nevada DMV has used a Salesforce-run chatbot since December 2022. The feature allows users to ask questions related to anything from scheduling appointments to registering vehicles, and the bot can provide answers or connect users to webpages on the topic.
The DMV is also working to release a more advanced chatbot that will be similar to platforms such as ChatGPT, a spokesperson said. This feature will use generative AI — which can replicate human capabilities to produce original content — and will be subjected to more scrutiny because it is more advanced. The agency has been conducting extensive “stress testing” to test the bot and ensure it won’t give out inaccurate information.
Sewell, the director of the state employment agency, admitted that he was hesitant at first to use AI for unemployment appeals processing, a discussion that originated last summer.
“Were there a couple of sleepless nights after that decision was made? Of course, because this is a new technology,” Sewell said. “But it is going to work, and it’s going to work the way we want it.”
He implored policymakers to not be too “rigid” when it comes to AI policies and hopes the agency keeps using AI in the future, such as for call centers and to help people find jobs.
“This is our first go-around with AI,” he said. “But it’s not going to be our last.”
