EP 81: Generative AI in Municipal Government
You've heard of "artificial intelligence," or AI, in one sense or another; we've been reckoning with the concept through books, movies and academic discussions since its earliest mentions in the 1950s. When you hear about it today, though, it's typically in the context of "generative AI," the rapidly evolving web-based tool that humans are using right now to enhance their worlds. Generative AI (like the popular ChatGPT engine, among others) writes up entire documents, draws up complex images, researches historical issues, drafts organizational plans, and even provides advice on complicated scenarios, pretty much on its own, hence the "generative" term; all you have to do is type in what you're looking for. It's pretty amazing. If you work in local government, you might be thinking about where and how it fits in; perhaps your city already has an AI policy on the books, or has used it to help draft or enhance language in public documents. But, like with any transformative technology, it comes with all kinds of cautions and ethics worries. Are we comfortable with it essentially learning how to do jobs we've always entrusted humans with? Already, we've seen trust in generative AI lead to embarassing, avoidable public blunders and messes. It brings privacy and authenticity issues, too. Did you know generative AI can sample a recording of your voice and learn how to speak as if it's really you, potentially faking out anyone who hears it? Take this episode of Municipal Equation, the podcast about cities and towns from the NC League of Municipalities, as a sort of primer on the generative-AI conversation in the context of local government at the moment. What's the potential? What are the dangers? How can cities and towns use generative AI safely and for the best? It's not going away. // Municipal Equation is a production of the N.C. League of Municipalities, https://www.nclm.org. Contact host/producer Ben Brown at bbrown@nclm.org.
Top Keywords
- generative 0.015
- using 0.007
- chatgpt 0.007
- government 0.007
- might 0.006
- prompt 0.006
- christie 0.006
- publicly available 0.005
- municipal 0.005
- technology 0.005
- tools 0.005
- available generative 0.005
Transcript
From the North Carolina League of Municipalities, this is Municipal Equation, a podcast about cities and towns. Hey. This is Ben Brown, and welcome to another episode of Municipal Equation, the podcast about cities and towns brought to you by the NC League of Municipalities. And today, we're covering probably the most obvious cutting edge topic out there right now, being covered by every podcast on verging issues or really any professional focus. And I'm talking about artificial intelligence or AI, which, yeah, everybody's talking about for its disruptive capabilities and how it's going to revise all of our careers, for the potential for fraud and fakery, and also for the positive aspects, the things we can dream up that AI might help us to achieve, and so on. There's so much to talk about with AI, And on this episode, we're talking about some intersections of AI and municipal government. But there's so much to talk about and pull apart and discuss that there's no way we're gonna be able to address it all here. In fact, we're just kinda getting started and going to have to acknowledge that we're still learning about an unfolding issue. But that still leaves us with a lot to talk about, and I think we could use this episode to peek in on some of the discussion being had about AI in the municipal space. And I just recently attended a conference where that subject came up. First, let's do a little history and definition setting. We've been using the term or phrase artificial intelligence for a long time, and it might mean different things in different contexts to different people. Sometimes, for example, it makes us think about a machine that seems conscious and can make its own decisions based on the input it gathers about a situation, an android like Data from Star Trek. But there are so many different ideas and levels of it of AI. Like, for example, you might consider any automated program or service on your phone to be some kind of artificial intelligence, like Google Maps or something, even if it works off of material and sources and possibilities from programming that comes directly from humans. According to a history from Harvard University, the concept of machines that could think as such is credited to a British scientist named Alan Turing, who in 1950 came out with a paper titled computing machinery and intelligence. But the computer technology wasn't quite there yet in 1950. In 1956, a computer scientist named John McCarthy was one of the people centrally involved in a conference that brought together the top minds in the field. They say the term artificial intelligence was coined at this 1956 conference. It's actually credited to John McCarthy, that computer scientist. And so this conference ended up viewing AI as something achievable, something to really work toward. The technology developed over the following decades, and it helped that computers were able to store more information and process more information more quickly as time went on. With certain milestones hitting pop culture, you might remember in 1997, the chess master Garry Kasparov being defeated in a game in which his opponent was an AI program from IBM called Deep Blue. Today, what we refer to as AI is all over the place. And what's mostly got our attention these days is the advancement of generative AI, which is what we're talking about when we talk about things like chat GPT or MidJourney or these programs or websites where you can enter prompt, like produce an image of a taxi cab driving on the moon. And seconds later, that AI will make that image for you, a picture that's just what you asked for. Or you can enter a prompt and get text results. Like, you can enter a prompt like produce a brief written history on whatever subject, like hot dogs. And that generative AI will generate a brief write up on the history of hot dogs within seconds. And, yeah, people are using this kind of AI these days to write their school papers or quickly come up with information that looks like it took a lot of time to research and compile. Technically, I could have gotten AI to write this episode of Municipal Equation for me. Maybe that would be a neat reveal at the end of the episode, like, and guess what everybody? This entire episode was written by AI. Isn't that impressive? Well, I promise that's not the case. That's not what's happening here. I am writing this episode myself. But to demonstrate the point, let's go to one of the more popular generative AI tools, and that's ChatGPT. So let's pull it up and ask it a question. So I've got ChatGPT up, and I'm typing in what are the pros and cons of AI use in local governments. And I've hit enter, and it is searching now. And within seconds, we have an answer from ChatGPT. Again, what are the pros and cons of AI use in local government? Here's what it says. It starts with the pros. Number one, efficiency. AI can streamline processes and automate tasks, leading to increased efficiency in local government operations. Number two, cost savings. By reducing the need for manual labor, AI can lead to cost savings for the government. Number three, improved decision making. AI can analyze data and provide insights that can help government officials make better decisions. Number four, enhanced citizen services. AI can be used to improve citizen services by providing quick and accurate responses to inquiries and requests. Number five, increased transparency. AI can help governments make data driven decisions and increase transparency in decision making processes. So those are the pros of using AI in local government as generated by ChatGPT. Now here's the cons that it says. Number one, job displacement. The implementation of AI in local government may lead to job loss for some workers who perform tasks that can be automated. Number two, privacy concerns. AI systems may collect and analyze large amounts of data, raising concerns about privacy and data security. Number three, bias and discrimination. AI algorithms can inherit biases present in the data they are trained on, leading to discriminatory outcomes. Number four, lack of accountability. AI systems may lack transparency in their decision making processes, making it difficult to hold them accountable for errors or biases. Number five, resistance to change. Some employees and citizens may be resistant to the adoption of AI and local government, leading to challenges in implementation. So those are the cons. So, again, what what I just read, those are pros and cons of using AI in municipal government, was totally generated on the spot by ChatGPT, one of the many generative AI tools out there. So some municipalities have already done what might be a common thing for, before long, and that's to generate a policy book of sorts on how that municipality is going to interact with AI. It's a thing. It's here. Let's write a policy is what a lot of cities are saying. Matter of fact, Government Technology Magazine in March took a look at how local governments in North Carolina are using AI and what that means for their policies. It pointed out that the town of Chapel Hill is using generative AI to help rewrite documents and policies so they're easier for the public to understand. Easier language, lighter on the government needs. So using AI to help with that. It also quotes the chief information officer for the city of Raleigh, Mark Wittenberg, as saying that it's, quote, important for us, especially as IT leaders, to really explore what the technology can do, and then be very mindful again about the community, the impacts to the community, and positive and negative impacts that it can potentially have. City of Raleigh chief information officer Mark Wittenberg. Raleigh is definitely a city that might come to mind when you think about technology and innovation. Of course, another US city you might think of is Seattle, production grounds for a lot of technology we use. Seattle does now have its own AI policy, and a press release on the city's website says, in quoting the mayor, Bruce Harrell, innovation is in Seattle's DNA, and I see immense opportunity for our region to be an AI powerhouse, thanks to our world leading technology companies and research universities. Now is the time to ensure this new tool is used for good, creating new opportunities and efficiencies rather than reinforcing existing biases or inequities. As a city, we have a responsibility to both embrace new technology that can improve our service while keeping a close eye on what matters, our communities and their data and privacy, end quote, which sounds a lot like what Mark Wittenberg from the city of Raleigh just said. The city of Seattle's policy was developed after a six month working period with what was called the generative AI advisory team and other sources of input. And so some of these existing policies might serve as templates for other cities, but everybody's kinda talking about it in their own context as they should. And and because this is something that we just have to learn about. AI topics have been popping up on all kinds of conference agendas, including that of City Vision twenty twenty four, the annual conference of the League of Municipalities. It was also on the agenda of a conference I just attended, and and that was of the NC Association of Municipal Attorneys, which was a great group to be discussing AI use in local government and what the ramifications might be.
What's new about generative AI, and that's really what's been exploding in the past couple of years, that's what you've been hearing so much about, is that this is a really different form of artificial intelligence. This is Christy Nikadem from the UNC School of Government. It's not just performed to it's not just trained to perform a discrete task. It's trained on massive volumes of data, whether that data is written text or images, and then trained to replicate and reproduce patterns in that data in a way that resembles human created content. Sometimes that content is going to be text, sometimes it will be images, sometimes it will be video. But what's new here is that it is generating brand new content. It's not just trained to perform specific tasks like navigation on our phones. It's generating something that appears creative and appears to resemble human creativity. Although, as we'll talk about later, it's not really truly intelligent, and it's not creative in the same way that we think about human creativity.
Right. So it's not human. It's not real human creativity. It it can sort of emulate what humans do, but it's artificial. But Christy pointed out that it already has done some impressive things in the context of human jobs, like practicing municipal law.
It has passed the bar exam. It has scored a one sixty three on the LSAT. It's not a shabby score. Scored a fourteen ten on the SAT. High high scores on the verbal section of the GRE and the, math section of the exam and the highest possible score on a number of AP exams. So it does have some serious capabilities. Right? Even though there are a lot of risks and limitations that we're gonna talk about later today, it does have extremely impressive capabilities.
To show what it knew, Christy brought up ChatGPT and gave it a prompt, an opportunity for it to show what it knows, what ChatGPT knows about municipal law. So if you have not experimented with ChatGPT or one of the other,
text generation tools, I wanted to give you an example, a quick example of how it works. So this was a prompt that I put into ChatGPT. I said, how can generative AI be useful to attorneys representing local governments? And within fifteen seconds, this is the response that it generated. It gave me a full list. It told me about legal document drafting, policy analysis, regulatory compliance, land use planning, Now I might push back on some of those responses, but it tells me it's good at data privacy compliance. I'm not so sure. As we'll talk about later today, I'm not so sure that's the best use for generative AI. So I might query it some more. I might prompt it some more. I might ask it to explain some of these bullet points in more detail. I might disagree with some of the responses. But nonetheless, this is a fairly impressive capability, right, to be able to generate that sort of response in less than fifteen seconds in response to a short plain language prompt.
Christie acknowledged how scary it might feel to have this technology running around out there available for us to use. Some might worry about the immediate impacts or misuse of AI
or the success of it when it comes to replacing jobs that humans usually do. There are lots of different tech companies that have been doing mass layoffs in part because they are pivoting more to AI, but in part because they're finding that AI is really good at coding, and it's very good at tasks that, lower level software engineers have traditionally been able to do.
They're finding that these large language models are able to do them now. Which a lot of people see as something to embrace as it's here. There's no putting it back into the box. Generative AI is here, and we've gotta figure some things out with it, which has given some jurisdictions pause.
Maine the state of Maine went so far as to ban artificial intelligence, in terms of their state agencies using it for six months. They put a moratorium on it because of the potential cybersecurity risk.
That moratorium from the state of Maine went into effect in June 2023. A summary from the government said, the state of Maine government must keep pace with a rapidly evolving cyber threat landscape that poses significant risks to the security of the state's network infrastructure, including the sensitive and confidential data that we are entrusted to protect for our citizens. This directive, meaning the moratorium, is in response to the unique security and privacy risks posed by the rapid rise in the breadth and scope of artificial intelligence systems, specifically generative AI, and establishes the moratorium while a main IT conducts further risk assessment, end quote.
There are tools like Eleven Labs and Murph AI that do voice cloning. With a very short snippet of someone's voice from a video or an audio recording, they can create a clone of that voice and have it speak the script that they want it to say. There are big implications for that, especially in the political space. And, of course, there's the image generation capabilities that also have people concerned. So what Chad GPT is able to do for text, these tools are able to do for images. You can type in a prompt. Let's say, I wanna see an elephant, balancing a donut on its trunk dancing in the rain. It will generate at least four different images of that of of your text prompt for you to choose from. It's not searching for those images. It is creating them on a full plot.
And, yes, video generation exists too. All of these tools are getting more advanced all the time. Have you heard about meeting chatbot AI?
AI that can attend a meeting in your place? There are a few different tools that work this way where it can generate you can send it to a meeting in your place. You can, which all of us have probably wanted to do that at some point or another. Send a chatbot to a Zoom meeting or a Teams meeting in our place. Literally, this technology
is something you can send to a meeting in your place. It will comment, it can interact with other people, and it can send you a summary of what the meeting discussed
afterward. I I mentioned this tool in part because I have heard of local government employees using this tool to attend meetings. And what does that do? It creates a public record of that meeting that was happening on Zoom or Teams that might not otherwise be a meeting subject to the open meetings slot. So
something to be aware of. Christie mentioned a number of emerging tools in the generative AI space and being incorporated in the space of municipal law. Some attorneys at the conference raised their hand when Christie asked if any of them had already tried some of these tools, like the kinds that can do basic searches and drafting. But the attorneys who raised their hand, or at least one of them, said there was still a lot of human legwork to do even with using AI. Christie explains some of the pitfalls of using AI for something like legal research. Even though these tools are getting better all the time, there's still wariness, like with a thing computer scientists are calling the black box problem. Meaning that after these tools get trained,
it's not very explainable or reproducible why they do what they do or how they do what they do. So if you were to type in an end a question to one of these tools and I was to type in a question and we do it at the exact same time, we're likely to get two different answers despite the fact that we're using the same tool. That's tricky when we're thinking about using this for government related work, right, or legal work. It makes that particularly tricky.
What stunned some people in the room was when Christie pointed out that some legal cases, like real stuff happening with law filings and courts and so on, were including problematic work done by AI, erroneous work, where an attorney or somebody at a practice used AI to draft a filing of some kind, and it included AI gathered information that turned out not to be good.
Many of you, I'm sure, have seen headlines around how, lawyers have been in trouble for using AI, in some of their cases. They make mistakes on the law especially if you are using a publicly available model like CHATTPT. It is not trained on cases that are in Westlaw or Lexus because those documents are behind a pay law. So it might have been trained on some free publicly available case law that you could get off the internet. Even then it was not really trained to do legal analysis and when it finds something that it does not know it will make up a citation in it. It displays. It makes up beautiful citations, perfect form, but they're they're false. They're just made up. And the same for facts. I was talking with a colleague who was asking it some questions about, about North Carolina. I think she was, I think this is about hotels and housing and habitability. She was asking some questions about North Carolina case law and it was just making stuff up left and right. Just false cases, fake citations, fake summaries of those cases. And they can look pretty convincing. Like if you knew nothing about that area of the law, you would think the summary looks convincing and the citation looks correct because it understands the format. But again it's just sort of predicting words that it thinks sound good together. It's not actually pulling from an active document or source when it's generating that answer for you. So not only can this lead to accuracy problems and sanctions for attorneys but it can also lead to defamation lawsuits for users and for these companies. So OpenAI has already been hit with its first defamation lawsuit from someone who said look a third party was using chat gpt. They asked about me. This was a radio host. They asked a question about this radio host. And in response, ChattGPT said that this person had been engaged in a lawsuit, that they have been sued, I think, for fraud and embezzlement. None of that was true. The lawsuit didn't exist.
So things like that have fueled the talks of moratoria and so on. Although a quick Google search will turn up lots of opinion pieces about whether it's wise to put a pause on things or whether the courts might have something to say about AI moratoria. Meanwhile, some states were leaning in, like Pennsylvania, which early this year announced it was partnering with OpenAI, one of the big AI companies, to deploy a first of its kind chat GPT product with enhanced cybersecurity to help state employees, quote, understand where and how generative AI tools can be safely and securely leveraged in their daily operations, end quote. From a press release from the Commonwealth of Pennsylvania, Here's a quote from the governor there, Josh Shapiro. I believe Pennsylvania can be a national leader in the safe and responsible use of generative AI in our government operations. And this first in the nation pilot with OpenAI will help us safely and securely learn from and use this important technology to serve Pennsylvanians and empower our workforce. Generative AI is here and impacting our daily lives already, and my administration is taking a proactive approach to harness the power of its benefits while mitigating its potential risks, end quote. Again, from the governor of Pennsylvania, Josh Shapiro. North Carolina is making moves on AI as well. With different levels of incorporation in different places, like the aforementioned Chapel Hill and Raleigh with examples like using AI to make complicated government literature easier for everyday people to comprehend or AI technology to analyze video footage like from traffic cameras. That's an older example actually from Raleigh. It's changing all the time. Draft policies are circulating on listservs or will be. The North Carolina Department of Information Technology, a state government agency, has assembled a list of links from around the web that can help educate and train people on the use of AI. In June, that department said they were taking the lead to identify risks and give guidance to state agencies so they can use generative AI responsibly. It also has a brief guide on its website when it comes to the ethics and so on of publicly available generative AI. The Department of Information Technology says that users must, among other things, never enter personally identifiable or confidential information into publicly available generative AI tools, review, revise, test, and independently fact check any output produced by publicly available generative AI, and be transparent and identify when content was drafted using publicly available generative AI, and so on. It also encourages users to document all uses of publicly available generative AI. And in January, the NC Department of Public Instruction released a guidebook for using generative AI in public schools. So we can expect more examples of best practices as we continue with AI. We can expect to see different kinds of regulation
as well. Certainly, I think more regulation to come. We just saw the FTC announce a new rule that prohibits, using AI to impersonate government agencies, but there's no private right of action so only the Federal Trade Commission is gonna be able to enforce this. Again, the FEC may regulate later this year and the FCC has just regulated AI generated voices. So we're seeing sort of this patchwork of of regulation at the federal level.
Christie said new bills to regulate AI are popping up all the time. If you're interested in tracking that, the council of state governments,
tracks emerging legislation, in all 50 states Has a really active database that you can, check out what states are doing in this area right now.
And so that's kind of where we are right now. Lots of possibilities, maybe a little regulation, but we're still very much figuring out how it's gonna be with generative AI, What the law might say at some point, what's just practically smart to do with generative AI, same way we were biting our nails about the World Wide Web coming about or maybe even when social media was becoming a thing, with all kinds of benefits, but also downsides and potential for regulation, things that keep us up at night, questions to answer. Experts say for right now to just keep looking out for new headlines, new examples of the kinds of work that generative AI can do. Look for example guidelines,
examples of people getting into trouble with it so you can avoid that kind of thing. Just stay informed. Make sure that you are staying educated on these developments. Make sure your client is staying educated. Again, a huge part of this is training. The genie is kind of out of the bottle. These tools are there. They're publicly available. Local government employees are gonna be using them.
So let's close this out. I'm back on the OpenAI Chat GPT website, and I will say that this chatbot itself, it says upfront with language. Like, it volunteers this language. It says right in the middle of the screen on Chat GPT that you shouldn't plug sensitive information into it, and you should check your facts when using generative AI in your research. Says right here, while we have safeguards, ChatGPT may give you inaccurate information. It's not intended to give advice. Says that right in the middle of the page. So okay. With that said, let's do a fun prompt to end this episode with. I mean, I wouldn't say we've been doing any doom or gloom here about generative AI. I think we've just been realistic, and it's pretty obvious how this technology, you know, which isn't going away right now, will help us with things we need to do in the future. But there's a flip side too, and we've covered that. So let's have some fun. I'm gonna type in who is the best podcast host of all time. Okay. And about three seconds later, I have my answer from chat g p t. It's giving me a response. Says, quote, it's difficult to definitively say who the best podcast host of all time is, as preferences vary greatly depending on individual taste and interests. However, some podcast hosts have gained widespread acclaim for their charisma, interviewing skills, and ability to engage audiences, end quote. And it goes on to mention five very popular podcast hosts as examples of who could be the most popular. But it fails to mention Ben Brown of Municipal Equation, so that's an example of how this technology can get things wrong. And, obviously, I'm just kidding. This is a cool research tool that gives you information, but also provides you context to lay behind that information. So it it makes the most sense. Again, check your facts if you're gonna be using something like this. And I do wonder how any of you listening to this episode, right now might already be using generative AI, whether professionally with the work you do or personally for any number of reasons. Let me know. I'd love to hear about that. My name is Ben Brown, and you can reach me at bbrown@nclm.org. Again, bbrown@nclm.org. NCLM stands for North Carolina League of Municipalities, working as one, advancing all. We're an organization that represents the cities and towns of North Carolina, more than 540 of them out there, all different sizes and styles. But we keep an eye on things like these technologies as they become increasingly accessible, as they're gonna get a footing across pretty much all of these different kinds of cities and towns in the future one way or another at some point. So let's keep following this together and circle back on a future episode, Municipal Equation, the podcast of the NC League of Municipalities online at nclm.0rg. We'll talk to you soon over another topic of interest to the space of cities and towns.