AI for Government: Tips for Using ChatGPT in the Public Sector [Includes Prompts!]
AI for government holds the promise of automating data analyses and rote tasks, helping us supercharge our productivity.
But be careful. Tools like OpenAI’s ChatGPT and Google’s Bard are still in their infancy and can produce inaccurate results—wildly so sometimes.
Keep reading for a deep dive into when to use AI for government, when not to, and some best practices to keep in mind.
Don’t Trust AI to Produce Data or Facts
One thing we know clearly in how to use AI for government is that you should not trust data or facts produced by ChatGPT or other AI-powered tools.
Why? Because ChatGPT has hallucinations, a phrase that describes what happens when AI fabricates information.
If you ask ChatGPT a question about an event in history it may produce an answer—even if the event never happened. So don’t trust it to tell you about the history of your local government, unless you know that history well enough to fact check the results.
On the history hallucination topic, when a reporter at the New York Times asked ChatGPT how James Joyce and Vladimir Lenin first met—a meeting that never happened—it produced this response:
This example is harmless. But there are other examples of hallucinations that show just how dangerous it can be to trust AI for producing facts, such as in creating your budget book, highlighting strategic priorities, or explaining rules and regulations.
For example, ChatGPT recently named a real law professor in a list of legal scholars who had sexually harassed someone, citing (Yes, we said citing!) a Washington Post article and including realistic details about the event, including that the harassment happened on a trip to Alaska.
But none of it was real.
The story and the supporting Washington Post article were both hallucinations, and the professor had never been accused of harassing anyone.
ChatGPT has also been known to hallucinate data and statistics.
The takeaway? Don’t trust ChatGPT with facts or data. Simply imagine trusting the AI tool to build a budget book with finance information, data on performance measures, or more. It would be so easy to miss a halluncinated “fact” that you then publish for your Council and residents.
Also, consider avoiding software that uses ChatGPT unless you can understand exactly how it’s being used, since it’s too new to be trusted as a reliable tool for work just yet when it comes to automation (that is, a human or two should always review the work AI produces).
How to put this advice into practice:
Don’t ask ChatGPT open-ended questions. These kinds of questions require it to draw from events, statistics, or other data—and instead of finding that information, it might just make it up.
“If you don’t know an answer to a question already, I would not give the question to one of these systems.”
– Subbarao Kambhampati, Professor and Researcher of AI at Arizona State University
If you’re doing work for your local government and trying to find data to use in a report, or historic information to use for strategic planning, don’t trust the information ChatGPT gives you on face value. Instead, give it information or data you have already vetted, and then use its outputs to streamline your work. (But even then, double check to make sure its outputs are valid.)
And if you want to get a little meta, you can ask ChatGPT to tell you about the dangers of using ChatGPT for work in local government . . . keep reading to see what we learned.
[Note: ChatGPT is constantly being updated. As OpenAI, Google, and others work to make improvements, hallucinations may become less common. But we’d still advise anyone using AI for government to proceed with caution.]
7 Dangers of AI for Government
Hallucinations aren’t the only potential danger of using AI in public sector work.
Here are seven things to look out for when using AI for government work.
1. Data Hallucination
As we’ve already covered, ChatGPT can sometimes “hallucinate” data or details that it hasn’t been trained on.
This means it might generate information that seems plausible but isn’t based on real data. This is an important factor to keep in mind when using ChatGPT for data analysis. AI is generally best used to help analyze and summarize large amounts of textual data, rather than generating data itself.
2. Privacy Concerns
Since AI systems process a large amount of data, there might be potential privacy concerns, especially if sensitive or personally identifiable information is involved. Users should ensure data is handled in a manner consistent with privacy laws and regulations.
For local governments, if you’re using a free AI account it’s important to know that all of the inputs you provide, including any of your data, belong to the company that made the tool you’re using.
Here are some privacy tips for using AI in government:
- Avoid this problem by using a paid account, so you retain ownership of your data.
- Turn off data sharing as best you can.
- Do not ever input private data or personal data, such as emails, birthdays, or social security numbers, into ChatGPT or similar AI-powered tools.
3. Dependence on Input
The quality of ChatGPT’s output is highly dependent on the quality of its input. If the information fed into the system is inaccurate or biased, the output could also be flawed.
Another way to put this is, “Garbage in, garage out”—an old adage familiar to anyone who works in data science.
To address this danger, work carefully on your prompts and the information you give ChatGPT, and you’ll improve the quality of the outputs you get.
4. Static and Hard-to-Audit Reporting
Even if you control all the data you put into ChatGPT for reporting—thereby avoiding the pitfalls of manufactured data from AI hallucinations—you should still be cautious about its outputs.
If you’re using ChatGPT to process large data sets for your budget book and other types of reporting, keep in mind that the report you get won’t update as your data updates.
This means the report is static. You won’t be able to update it as your data changes over time, making your report a single snapshot taken at a single point in time—unlike the kind of responsive, real-time reporting you get with a budgeting tool designed for local government, like OpenGov Budgeting & Planning.
Also, the process ChatGPT uses to create reports is a black box, allowing little insight into whether its work is correct. Since you can’t review the process used to make the report, you may have trouble understanding where all the data came from, which could be problematic for audits.
Despite its sophistication, ChatGPT might not fully understand the context or nuances of certain inquiries, which could lead to misunderstandings or inaccurate responses.
Remember, ChatGPT is just a tool. Its outputs require human interpretation and understanding to be useful, and should never simply be taken as is or in a professional setting without review.
6. Lack of Ethical Judgment
AI models like ChatGPT don’t have an understanding of ethics or moral values. They might generate content that is inappropriate or offensive unless certain safeguards are put in place.
Again, it’s up to you to review the outputs you get and make sure they meet your standards.
7. Lack of Specificity in Language
Policy and strategy is chosen specifically based on your community’s needs— and allowing a general language model to shape your communications can result in the wrong message.
Although you can prompt ChatGPT to do some impressive things when it comes to writing style and voice, its language model is still based on generalizations. This means the writing you get from ChatGPT won’t be as carefully worded and specific as the writing you get from a person in your community.
10 Ways to Use AI for Government [with Prompts]
So how should you use AI for government, specifically for local government?
Despite all the warnings above, ChatGPT can be an incredibly powerful tool, making your work easier and more efficient.
Here are 10 ways to use AI in your public sector work.
1. Public Service Automation
ChatGPT can act as an interactive tool to answer resident’s queries on a 24/7 basis, freeing up staff for more complex tasks.
It can provide instant, accurate information on topics like council meetings, waste collection schedules, local events, or zoning regulations, improving both efficiency and user satisfaction.
The only caveat is to make sure you train your AI to pull from correct sources on your local government website for sharing information.
2. Documentation Assistance
ChatGPT can aid in generating reports, proposals, meeting minutes, and other documents. The tool can provide first drafts or suggestions to improve written content, saving time and enhancing the quality of local government paperwork.
For meeting minutes, try using a prompt like:
“You are a helpful assistant trained in local government work. Take these rough notes from a recent City Hall meeting and turn them into an organized, well-written document with a list of actions and who is in charge of doing them.
[INSERT ROUGH TEXT OF MINUTES]”
3. Civic Engagement
ChatGPT can help solicit and process resident feedback, opinions, and ideas.
To do this, feed it a large amount of raw information collected via surveys or some other method, and ask it to produce a report on the findings.
By parsing through and summarizing large volumes of public input, local governments can use AI to better understand their community’s needs and concerns.
4. Policy Explanation
Complex policies and regulations can be translated into simpler language for public understanding. ChatGPT can provide explanations for common queries regarding local laws, taxes, or policy changes, helping ensure transparency for residents.
The key to doing this kind of vocabulary translation is to clearly tell ChatGPT the kind of language you want to use.
Try a prompt like:
“You are an experienced City Manager who has been working in local government for over 20 years. Your job is to take a new City policy and rewrite it for a broad audience. In your writing, you will translate any technical or legalistic language into language that anyone with a fifth grade education could understand.
Here is the policy that I would like you to rewrite:
[INSERT POLICY TEXT]”
5. Training and Onboarding
ChatGPT can serve as an interactive learning tool for training new employees or updating existing staff on new policies and procedures. Its interactive nature can create engaging and effective training modules.
You can train the system to help employees walk through certain scenarios, having it respond in different ways depending on the action the employee takes so they can learn policies in an interactive way.
6. Emergency Communications
During crises, ChatGPT can be used to provide quick, consistent, and accurate information to the public. It can also help manage large volumes of emergency-related inquiries, ensuring people are informed and guided properly.
Remember, it’s important that you feed ChatGPT the raw information you want it to shape and rewrite so that the outputs it creates are accurate. Again, it’s important to review for “hallucinations,” especially in emergency communications.
Try a prompt like:
“You are an experienced PR professional with over two decades doing PR for local governments. The City of [NAME] has just gone through a hurricane and now needs to let residents know what is safe to do/not do and share what the City is doing to provide emergency services and restore basic functions, including electricity.
Use the rough text below to create a clean, professional piece of writing as an emergency communication announcement for residents:
[INSERT ROUGH TEXT]”
7. Accessible Services
ChatGPT can help improve the accessibility of local government services for people with disabilities.
For example, it can convert written information into spoken words for visually impaired individuals.
8. Multilingual Support
Local governments in multicultural and multilingual communities can use ChatGPT to provide support in multiple languages. This will ensure information is accessible to all residents, regardless of their primary language.
When you compare ChatGPT’s translations to Google Translate’s translations, ChatGPT is typically the preferred option because it uses natural language and can translate idiomatic phrases to maintain their intended meaning. (That being said, Google Translate is a reliable option for professional communication that doesn’t use idioms or phrasing drawn from regional dialects.)
But make sure to be as specific as you can about the language you’re translating from and the language you’re translating to when using ChatGPT for translation.
Also, ideally you would have a native speaker review the translation for accuracy. If you don’t have one on hand, use ChatGPT to translate the translation back into English and see if there are any glaring errors.
Try a prompt like this:
“You are an experienced translator who is fluent in both American English and Latin American Spanish. Translate this public notice written in American English into Latin American Spanish so that it will be generally understood by a broad audience. Make the translation you produce in Latin American Spanish comprehensible for a fifth grade reading level.
Here is the text to translate:
What Should Local Governments Do about AI?
Given how new AI is, many local governments are issuing individualized guidelines on how employees should use it.
The Cities of Boston and Seattle have each issued guidance documents with best practices, and others are sure to follow. (Meanwhile, other local governments are issuing blanket short-term bans until more information can be gathered.)
“Whenever there’s an opportunity of delivering government services better, I think that it is our obligation to also learn about it, and if there’s risks, understand those risks.”
CIO Santiago Garces, CIO for the City of Boston, MA (source)
If you work in local government, you may want to consider creating your own best practices or simply adopting a list from another municipality.
Here are some resources to help you consider how to use AI for your government work, and to guide you in crafting your own guidelines:
By the way—we used ChatGPT to make this blog post.
That’s just one more example of how AI can help you do your work more quickly and efficiently.