Generative AI Offers Federal Agencies Common-Sense Opportunities to Simplify and Improve How They Work

News coverage surrounding generative artificial intelligence (AI)—AI systems that produce novel text, images, and music from simple user prompts—often focuses on headline-grabbing applications, unproven existential threats, or classic tech panic. Federal agencies, however, should ignore the hype and panic and instead focus on how generative AI can make government operations and service delivery simpler, easier, and faster. Generative AI use cases in government range from the standard—creating a really good chatbot—to the more ambitious—designing an end-to-end digital service. But generally, sensible applications of this technology can simplify access to and organization of information, automate bureaucratic processes and data entry, and contribute to design and content production.

The Biden administration is working to develop a national AI strategy. Part of that strategy will include how federal agencies can broadly leverage AI to improve federal services and missions, as the opportunities and use cases for AI in government are numerous. These possibilities will expand as generative AI progresses. According to research from Accenture, 39 percent of the working hours in public sector organizations have great potential for automation or augmentation through generative AI. Generative AI not only helps agency staff with mundane or time-intensive tasks (e.g., entering, managing, and analyzing inordinate amounts of data), it can also act as a “co-pilot” or creative assistant in drafting memos, presentations, press releases, policy or user guides, and contracts.

What distinguishes generative AI is the technology’s ability to design and create original content from prompts. Generative AI chatbots won’t simply spit out rote, generic (and often unhelpful) responses to user inquiries. Rather, this new generation of chatbots utilizes large language models (LLMs) to access massive datasets to construct original and targeted responses even if the user’s questions are unclear or imprecise.

A virtual assistant of this kind could greatly improve the American tax filing experience, for example. The U.S. tax code is notoriously complex, and every year people have specific questions that can’t be answered by searching dozens of pages on the IRS website or diving into 1 million words of code. Indeed, many tax filers don’t know what they don’t know. How can they get an answer for something if they can’t articulate what to search for or where to look?

A generative AI virtual assistant can take data and information from the IRS and work with an individual’s questions and responses to construct answers and provide guidance that is tailored to that individual’s circumstance. Furthermore, this is a dialogue that builds and learns. An initial inquiry asking the chatbot “what is a dependent?” could evolve into a collaborative, constructive conversation to address that individual’s full filing needs. Such an assistant would be invaluable to both the tax filer and the IRS, which has historically struggled to respond to phone calls during tax season.

This technology doesn’t let agencies off the hook regarding needed improvements to their websites and other digital services, but it offers an exciting alternative in accessing critical information that doesn’t require users to navigate across multiple screens, interpret policy or legal language, and inevitably experience cognitive overload. Such a virtual assistant is a simple, powerful application of generative AI for an important government service.

Additionally, Congress has begun exploring the opportunities generative AI offers the legislative process. Automatically transcribing hours-long hearings is an obvious win, but generative AI can take this activity a step further by producing succinct, readable summaries. This capability prevents congressional staffers from engaging in an unpleasant, time-consuming task while also making the critical information available to the general public in a faster and more accessible manner. This same solution could be applied to any number of public meetings and events produced by federal agencies.

But generative AI can do even more. Returning to the IRS as an example, the agency recently shared that it will be moving forward with developing a “Direct File” service—a free, voluntary, IRS-run online filing system. Generative AI solutions can assist in the design of end-to-end digital services like this initiative. AI-assisted coding and testing would allow the IRS development team to get the service’s “scaffolding” in place faster and easier so that service designers and product managers can spend more time upfront engaging with customers and performing user research and software engineers and testers on the more complicated elements of the service. The result would be a cleaner, simpler website that maximizes its minimal features so people can quickly and easily file their taxes online (ostensibly the primary goal of such an effort). AI-assisted coding and design may be a more ambitious application of generative AI for federal agencies, but it demonstrates the breadth of capabilities generative AI offers the federal government.

Importantly though, federal agencies—or governments at any level—shouldn’t use generative AI because it’s the shiny new thing. They should consider using generative AI when it offers legitimate opportunities to simplify and improve how government works. A virtual assistant doesn’t need to look like anything more than a basic chatbot—which most people are accustomed to interacting with—but it should make the virtual interaction easier and faster in getting critical information and guidance.

Lastly, effectively leveraging generative AI in the federal government doesn’t mean just using ChatGPT as-is—after all, ChatGPT is a public service that Open AI designed to be able to create synthetic data and realistic responses, even if those responses aren’t accurate—federal agencies will need to integrate their data into LLMs to make them useful to the public. LLMs are, by design, trainable and tailorable according to specific industry needs and protocols, including the public sector. Integrating federal government data ensures that LLMs provide accurate responses based on the latest government data.

The federal government struggles with a variety of issues and challenges, such as slow services and backlogs, significant administrative burden and bureaucratic processes, and impending budget constraints. Federal agencies need to take advantage of new tools at their disposal, including generative AI, that help overcome these issues and make interacting with the government feel easier and simpler.