What is a chatbot, anyway?

After months of working alongside governments to help them responsibly integrate generative AI, our team members have some thoughts about chatbots.

Partner:

On this page
Pandemic exposes systemic inefficiencies
Developing a human-centered approach to language access
‍
Quantifiable improvements in benefit accessibility

Leveraging generative AI to scale Language Access

A model for inclusive government services nationwide

By: Carrie Bishop, Krista Canellakis, and Amanda Rucker

As generative AI (gen AI) advances in governments, so does misunderstanding of how to use its powers. At U.S. Digital Response (USDR), we’ve been fielding requests from governments big and small for ‘chatbots’ in hopes of solving a range of issues. Similar to requests for ‘apps’ and ‘dashboards’, requests for chatbots are usually nebulous and start with a solution rather than a good understanding of user needs.

From community members to first responders, there are countless use cases that many hope ‘chatbots’ will assist — but we argue they’re usually the wrong solution to the problem.

It is critical to start with user needs, and not force a technical solution.

We know these requests are rooted in a desire to offer something to users in hopes of responding quickly and accurately to their problems. As the chatbot requests poured in, USDR interrogated what our partners’ hopes were when they asked for a chatbot. From unpacking the problem, a few user needs emerge:

  • I want to access information from a range of sources with one request
  • I want to make my request in natural language
  • I want the information I get back to be formatted in a digestible way
  • I may have follow-up questions based on the response I get to my request

The user needs aren’t the only thing we need to keep front and center when solving these problems, the context of their needs is also important. Is this for a resident, who needs to know what a benefit’s eligibility criteria are? Or is it for a policymaker who wants to query every meeting agenda to find relevant decisions? Is it for an emergency responder who needs to know some real-time information in a critical situation? Or is it for a permit expediter who wants to know whether a new build will get approved?

Gen AI is a powerful tool, but not a magic wand.

Here are three major realities with the idea of a chatbot as a solution to these problems:

  1. Chatbots cannot do a good job of giving accurate information. They frequently hallucinate, and it’s hard to think of a public-facing situation where the government giving out inaccurate information is acceptable. Even internal use cases risk decision-making on bad information.
  2. ‘Chat’ interfaces can be hard to use. A Q&A interaction is not always the best way to navigate content, especially for residents who might be trying to find information on the go, don’t have time to type a full sentence into a website, or don’t want to speak their request out loud. Context-specific user testing is crucial before deploying a ‘chatbot’.
  3. It’s the wrong paradigm. Increasingly we will see the rise of AI assistants — the next crop of iPhones and devices will certainly have LLMs on the device, and much better versions of Siri and Alexa that will go and find information for you, based on your own personal needs. Government agencies need to get ready for AI to navigate their content, rather than providing their own AI interfaces. Eventually, AI assistants will be tasked with filling in forms and accessing services on behalf of users too.
  4. There are better ways to solve these problems, and they start with having accessible, well-organized content. Government agencies should continue to prioritize organizing public information clearly and in plain language. Gen AI can be a huge help in creating better web content and simplifying information, which is a great use case for the technology. To get started, try putting some of your harder to read web content into an AI tool of your choice (like ChatGPT, Gemini, or Claude) and ask it to re-write the content to a 5th grade reading level. It might not get it right first time, but you’ll see how powerful it could be with some more prompting. Or ask it to read some of your web content and tell you what appears to be the most important information on the page. You’ll quickly learn whether your information hierarchy is correct. Gen AI is also great at coming up with user interface ideas, and coding. Think of gen AI as having a super-charged digital services team that can help you get all your content, interfaces, and digital services in order. Of course, you need humans in the mix — conducting user research, writing great prompts, checking accuracy, running usability tests, and more.

Gen AI is a transformational technology that will have a huge impact on government services.

Chatbots aren’t the silver bullet, but there are plenty of great uses for gen AI. USDR can help you navigate it all by providing a safe space for experimentation from a product-agnostic viewpoint. We can help you decide whether you should use gen AI, prototype solutions (thanks to support from Google.org), and empower you with the technology to address pressing challenges in your communities.

If you’re grappling with gen AI, get in touch and let’s work together to make sure we’re using it to solve the right problems in the right way.

‍‍About U.S. Digital Response‍

U.S. Digital Response is a nonprofit, nonpartisan organization that works alongside governments at all levels to ensure they have the capacity to support the critical needs of the public. Founded in 2020, USDR connects governments with technology, resources, and support, leaving them better equipped to deliver services and support to millions of people nationwide. To date, USDR has built a dedicated community of over 9,000 volunteers driven to serve in a time of need and has partnered with 400 government and nonprofit partners on more than 400 projects. See more at usdigitalresponse.org.