State and Local Governments Can Leverage LLMs for Better Document Management

“LLMs are primarily focused on understanding and processing natural language, while generative models create new content,” he says. “Additionally, LLMs are often used as a foundation for building more specialized AI applications, which can then be fine-tuned for specific tasks.”

State and local governments can take advantage of the power of large language models to accelerate the completion of everyday tasks. 

How Can State and Local Agencies Use Large Language Models?

Overall, LLMs can be viewed as “very large deep-learning models that are pre-trained on billions of data points and can infer and generate humanlike responses,” says Kim Majerus, vice president of global education and U.S. state and local government at Amazon Web Services (AWS).

“They are one important component in powering AI and generative AI solutions,” she says. LLMs are designed to perform a variety of language-related tasks, such as answering questions, summarizing information and even generating new text.

RELATED: Three tips to mastering operational AI transformation.

“You can think of an LLM as a really smart assistant who can quickly read through mountains of text, summarize it and provide insights that can help a lot of governments and their officials make better decisions in a much quicker way,” says Matthew Dietz, senior global government strategist at Cisco.

As such, LLMs are “key to transforming public services,” he says. They allow government “to be more efficient and more responsive while also driving new and innovative services to their constituents, their communities and their employees.”

What Are the Risks of Adopting LLMs for Governments?

While LLMs offer much promise, they also come with risk, particularly around data privacy and potential bias. Data fed to LLMs needs to be kept secure, and the people using the models must be sure that their LLMs do not inadvertently incorporate or propagate any bias.

State and local entities can mitigate those risks “by ensuring strong data governance, by keeping sensitive information protected through local, private and secure LLM deployments, and by conducting regular audits to prevent biases and data corruption,” Dietz says.

DIVE DEEPER: Data governance strategies can help avoid AI biases.

Another risk comes with “over-reliance on this new technology, or assuming it’s a quick fix,” Gaudet says.

“While LLMs and generative AI offer tremendous value and potential, they still require human oversight, often referred to as human-in-the-loop. I prefer to emphasize the need for the right human in the loop,” he says.

“Ensuring that a skilled and relevant staff member reviews the data for errors and bias reduces over-reliance and maintains accountability at the appropriate level,” he says. “Ultimately, whether we create content independently, with a team or by using tools like LLMs, we remain responsible for the final results.”

What Are Some Government Use Cases for LLMs?

To demonstrate the art of the possible in state and local government, Majerus points to a recent real-world example: The Nebraska Judicial Branch is leveraging LLMs on Amazon Bedrock, a fully managed service that helps users build generative AI applications.

“Handling about 285,000 legal cases annually, the Nebraska Judicial Branch manages hundreds of thousands of paper documents, audio and video files, and various forms of evidence every year,” Majerus says. AWS helped the branch automate its document management system and introduce a generative AI feature that helps attorneys access data quickly from any case exhibit.

LEARN MORE: Intelligent document processing may be AI’s most palatable use case.

In addition, “contact centers and citizen experience applications are other spaces where LLMs can improve the conversational AI experience” for state and local agencies, she says. “LLMs can understand intent and sentiment, and can even provide sophisticated reasoning to engage in humanlike conversations and offer complex answers.”

In this regard, Gaudet points to the rising use of chatbots by state and local governments as a prime example of the potential for LLMs to elevate constituent encounters. LLMs supporting chatbots can be trained on website and publicly available data “to provide humanlike responses to common public inquiries,” he says.

Generative AI expands these LLM-driven applications. For instance, new tools can be trained on law enforcement data, “enabling report generation and identifying escalation trends,” he says. “Other applications include analyzing financial data and creating visualizations.”

Overall, “governments can leverage LLMs to enhance data analysis, decision-making and communication,” Dietz says.

“For example, an LLM can process complex legislative documents, helping officials make better data-driven decisions. Or it can help them to analyze healthcare reports to detect trends and improve public health strategies,” he says.

DISCOVER: Synthetic data supports state and local government AI initiatives.

How Can Government Agencies Scale LLM Use Cases?

While many state and local organizations will introduce LLMs in limited use cases at first, they will eventually need to scale up those deployments as generative AI expands into a range of government operations.

While artificial intelligence “offers government agencies cutting-edge technology to address mission priorities, they face the heavy burden of ensuring safe and responsible use,” Majerus says, adding that cloud providers can play a key role in the responsible scaling of LLMs.

“Amazon Bedrock already provides access to LLMs, which gives agencies flexibility to use different models that best address the use case and change as their needs evolve,” she says. 

Through various partnerships, AWS is elevating additional tools and capabilities to help state and local agencies “build and scale generative AI applications that are customized for their specific use cases,” Majerus says.

PREPARE: Demystify AI adoption for your organization.

The right networking infrastructure will also play a role in scaling LLMs in support of state and local uses.

“At Cisco, we provide the secure infrastructure and the network capabilities that make these AI deployments both feasible and scalable without compromising security, because security is always at the forefront of our minds, especially when it comes to AI,” Dietz says.

At the same time, “scaling LLMs is not just about the technology,” he notes. “It’s also about empowering government employees.”

In Maricopa, Gaudet likewise puts a heavy emphasis on the human and organizational components of scalability. “The best way to scale is to break down data silos so the agency can see the full picture,” he says.

“While government collects a lot of data, it’s rare for data from different departments to be integrated,” he says. “Encouraging collaboration and communication among team members across departments is key to overcoming this. If the team isn’t working together, the data won’t either. Agencies will go as far as their data will take them.”