Generative AI Learnings from Anchorage, AK

Happy Friday, everyone. It's Joe Morris, Chief Innovation Officer with eRepublic. A solo episode today while Dustin's off on another assignment, but we're here to dig into some really exciting topics today around generative AI and all of the latest learnings and happenings that we're seeing across the public sector. And it's a pleasure to be joined with Brendan Babb, the Chief Innovation Officer and Anchorage IT Director.

Hi, Joe.

He's here to help me dig into what we're seeing all in and around city government as it relates to all things AI. So Brendan, thanks so much for joining me today.

Thanks, Joe. I'm very excited for the opportunity.

So before we dig into all of the bits and bytes of AI, maybe you could share a little bit with me on your background and your role with the city.

Yeah, I'd love to. I have lived in Anchorage, Alaska most of my life. I was born in Alaska and I love it here. And I'm lucky to be the Chief Innovation Officer for the Municipality of Anchorage. It's really cool to try to make services better for people I see in the grocery store. That's great. Then we're lucky to have an innovation team from Bloomberg Philanthropies. I have a team of 2 other people that do all the work that I take credit for. Yeah, it's great. I've been interested in AI for a long time. Sorry, I'm getting— sun is moving in and out here. Studied it in high school and wrote a term paper on it that I'm sure is probably not current or correct anymore. But also did some work on research on genetic algorithms and genetic programming, which is a form of machine learning with the university here and then also working with NASA JPL. I did some work on Mars rover image compression. But one thing that's cool about that is it's a black box. Like, you get improvements, but you're not sure exactly how it's working, which is similar to some of the large language models. I find that interesting. I've been here 8 years at the city and excited to figure out how to use generative AI.

It's so fascinating. I know at least for me in my career here, it seems like the generative AI adoption curve has and very much accelerated compared to what we've seen before.

Yeah.

Faster than cloud, faster than mobility. It seemed like not quite overnight, but effectively overnight, all of a sudden it came. And now whether it's through govtech.com or our Center for Public Sector AI, we rapidly moved from kind of this like shiny object to seeing some real-world use cases across our states, our cities, and our counties, Where do you think some of those most promising use cases lie for improving city operations?

Well, one thing I think that's unique about government is we have a lot of open data. We have data that we've already put out to the public. We also have municipal code, like laws and regulations of how high your fence can be. So we have data that we can use. We don't have to worry about it being private data like corporations often have to. to do. So some of the experiments I've done is Anchorage has land use code and building use code like most other cities, but it's often written in very governmentese. It's very hard to understand, or it's a 300-page document. So I took a custom GPT from OpenAI and just kind of fed the information in there and just asked it to respond in kind of plain language. So you could ask like, how how high can my fence be in, you know, in a residential area? And it'll search through it and kind of come back with references to the actual code, but also explain it in simple terms. So I think there's a lot of potential there. And then also, we have a lot of data about houses and other things, and it's easy to make those kind of public-facing. So I think, I think that's one advantage government has. The other challenge is just You know, we also have a lot of sensitive information, so just trying to make things privacy and security, and sometimes government can be a little slower in implementing those kind of setups.

As you kind of, you know, dipped your toe in that pool for a second here on that, on the maybe the slower nature or the risk-averse nature, where do you see some of those challenges or even barriers being as you look at You know, AI adoption within local government?

Well, I see like everyone's trying to figure out kind of— there's a couple things that are unique to government. We have— well, maybe not unique to government, but anytime we have records requests and public records, so having a meeting where a decision is made, that information has to be available to the public if they want to see what happened. Governments are trying to figure out what qualifies if you're asking AI to generate something. Is that a record that needs to be saved? Because you can have conversations with people and write stuff on the back of a napkin, and that's not necessarily something that you can FOIA or needs to be FOIA'd. People are having challenges there, and then also just trying to figure out how to roll it out to the whole company. I've seen a lot of cities work on Just allowing some people access to kind of experiment and just try to get a feel for it because it is so new and it's moving so fast. I mean, one great thing about being a consumer of it versus being a startup is any advances that occur, that's something government can leverage. If you're a startup and all of a sudden your feature is now part of the software, that's a challenge. But I think it's just going to take government a little longer to kind of figure it out. But I'm interested. San Jose has a GovAI coalition, and they've been an early adopter of getting governments to kind of talk together. And I was on this Teams call with like 200 other people, and they're kind of being very— they're articulating the steps, a roadmap they want to take. And they're doing a good job of kind of breaking into groups and also trying to figure out a way to talk to vendors and have consistent questions so they can vet vendors. Everyone's just trying to figure it out and it's moving so fast. I'll create a presentation and by the time I give it, 4 features have already changed in the presentation.

Yeah, I feel like there were, especially early on, and it's still true today, that if you took a weekend off and you went camping with no internet access, you come back and you're like, oh my gosh. The pace has been so quick. You know, I'll come back to that collaboration piece in a second, but I think it's relative— related to this question too. Whether you're in Anchorage or here in Northern California where I, where I am, some of these concerns are the same, right?

Yeah.

Whether privacy, ethics, government has a lot of similarities. So when you look through those challenges and considerations, what role does that collaboration that you just spoke of play into maybe understanding where some of those hurdles are and those barriers are and knowing that, hey, San Jose's, you know, maybe they're, you know, a step and a half ahead of you, or they've already got this, that'll speed us up. Does that also play into that, you know, accelerating your adoption or your consideration is understanding where some of those trouble spots may be?

Yeah, I think so. I happen to know the Chief Information Officer at Boston. We're friends, and Boston's done a good job of kind of just publishing their, their guidelines for AI just to the public early on.

Mm-hmm.

So we adopted a lot of what they did for our internal guidelines. And just listening to other cities and states, New Jersey is really doing a good job of just trying to figure out how to train all their government employees. I think Oklahoma or another state in the Midwest is interested in training their residents, just giving everyone a basic idea because this is something that's out of the box already, so everyone's using it. Like you mentioned earlier, it's the fastest adopted new technology ever, so everyone has some familiarity. It's been great talking to other— there's a City AI Connect from Johns Hopkins University and you have to have a government email address to join it. So the idea is just to have kind of local government people safely talking with each other versus having vendors in the space at this point, but trying to figure out what challenges they're having and share those with each other. And that's great because often people will think of stuff you haven't even thought of yet.

Yeah, that's one of my favorite dynamics of being in this space, right, is the ability to steal those good ideas and collaborate and push each other forward. You mentioned Boston earlier, you mentioned San Jose, you mentioned the guidelines, but like, how important is it that, that ability for cities, counties, just, just peers to come together and shape those best practices and, and, you know, steal those good ideas?

I, I think it's my favorite thing about government. Uh, I've worked in, um, I've worked for AMD, I've worked for academia, You know, when you're working for a corporate company, you, you really can't share ideas. You have to keep that secret. We had Intel right across the street, and we were very, like, protective. And then working in academia, you also, like, don't want to share your ideas until you publish them. But government, it's like I can see someone and say, like, hey, I like your suit. And they're like, oh, hey, let me make you a suit and tailor it for you, because everyone in cities wants to help people in their city, and they're really excited if it helps someone else in another city. So that collaborative environment, um, is just great because, uh, yeah, you can just steal— I mean, it's great for Anchorage. I've stolen ideas from tons of other cities and they're happy to like tell me how they do it. Like, uh, we've had a lot of snow recently. I was talking to Syracuse last week about how they're doing stuff with GPS on snowplows and they're just really happy to collaborate. And so that's kind of my favorite part is you don't it's true collaboration. You don't have to be guarded. You kind of— everyone wants the best outcomes.

Yeah, no, I love it. And I like the competitive spirit that does exist a little bit too. The neighboring cities, the neighboring counties, all kind of— there's a little bit there, but that does fade away a bit more than you do see in the private sector, to your point. You know, as you look at that partnership dynamic, and you hit on it a little bit earlier, there's that government-to-government collaboration, but so much of this is also on that collaboration with industry, right?

Yeah.

And so much of those innovations that you see on the daily are being led out of that community. So how do you see partnerships and collaborations coming together from some of the communities that you've mentioned, whether that is the tech industry, whether that is these academic institutions coming together to advance this conversation?

That's a great question. I think some of the large companies like Microsoft, OpenAI, Anthropic, Meta, they're all working on models and improving things all the time. One funny thing is government can be like, oh, we want to hold off on AI, but a lot of the products that we're already using are having AI put in them. So it puts a lot of pressure just to kind of come up with policies on the fly. I think right now some of the large companies have an advantage in terms of they've been working on it a while, but I think they're going to need to collaborate with the users as well. And then I think also residents need to be involved and just kind of figure out what they want or what kind of warnings they want, or just— I think people are okay if AI is being used if you just say like, hey, like yesterday I rewrote a bio for some conference using AI. But if you just say like, oh, I used AI to rewrite this or kind of have labels on it. But I think it's going to be kind of a collaboration between vendors. Because the weird thing is everyone's excited about it, but use cases are kind of slowly coming along. I'm experimenting with hypothetical use cases or just trying stuff, but But they're slowly coming along. So everyone's excited, but then when people ask you what exactly you're going to do with it— so I think that's something that government and vendors can figure out together, along with residents, because ultimately it'll be what do, what do people want to use it for.

And that's going to— you just made me think of another question that I'll pose to you in a second. But earlier you talked about like AI in the context of public records requests, and it made me think back to not that long ago when we were working through uh, that in the context of social media and trying to get— build the understanding that, yeah, you're— if you have an official government social media account, those are also subject.

Yeah.

Same disclosure. So I think it's just that same education process that we're now applying to a new, you know, to a new realm. But when— what you just talked about is fascinating to me in terms of like the resident aspect. Do you think that there's a role for city governments, state governments, just government at large in terms of helping residents also build up that AI literacy?

I do. I haven't figured it out for Anchorage. I'm following the lead of other cities and states. California is also talking about providing some basic literacy. I've been impressed Baltimore had something for their city employees just on data literacy. So just kind of like, oh, this is why data is important when you're doing this, this is what we're going to use it for. And I think this has come on so fast that it's just useful for educating your residents how to use it because I don't think it's going away. I think there's a hype curve, but I think it is helpful for residents to know how to use it and also to get feedback, to give feedback on how they'd like to see it used or what kind of warnings. And your social media example is really good. like, that's relatively new as well, and we had to figure out how we, you know, archive that and how we deal with that. But on the flip side, I think it's going to be great for residents for— I've been experimenting with— we have assembly, which is like our council meetings, and they're like 3, 3 and a half hours long, 4 hours long, and they're often— we have video of them, we're trying to be as transparent as possible, but it's hard to find stuff. But if you And then you can throw AI at that. You can get a transcript of the video and kind of search for certain sections. I did an experiment and you kind of link to the video, like when, when was there a discussion about this? And one experiment I'd like to try, I don't think we're quite there yet, is kind of color commentary. I saw someone do something where they had 2 commentators just on a soccer video game in real time, but I think you could do that with council meetings where it's like, okay, now they're voting on the amendment.

Mm-hmm.

to the amendment, and the other person could be the color commentator and be like, this is important because last summer this got voted down, or just kind of add more context for residents so they, they can more easily understand what's going on. Because I, I've spent 8 years on this and I'm still confused sometimes when we're in a meeting. So—

And I think that you're seeing that, that evolution of that, that ex— that experience already, right? And, um, we see that both in the, the priorities that we get out of our digital cities and county surveys, this focus on the resident experience and trying to remove some of that complexity from the process. And AI, it certainly can, you know, can be a force multiplier there to do just that. So you mentioned that, you know, like the kickoff of this interview, going all the way back to high school, that this was something that you had been looking at. So now with it being here and we're seeing, you know, real-world use cases today in the public sector, may I ask you to take out your crystal ball And maybe give us a little bit of a forecast in terms of how do you see this envisioning? You look at city government, I don't know if it's 10 years, that seems a little far out, but you look at like the next 3 to 5 years, where do you think local government is going to be when it comes to their adoption of generative AI and what impact do you think it's going to have?

I think some of the early use cases are object detection, kind of measuring assets, So you could have things driving by and seeing bus stops and saying like, oh, we need to fix this thing on the bus stop or potholes. So, and then OCR or just government has a lot of text, a lot of unstructured data, and the ability to handle that I think will be a big change in terms of— I saw one proposed use case is like, oh, we have utility bills from 5 different utilities and we want to know how much energy we're spending Can there be a way to scan that and bring that all together? I think summarizing translation into other languages, I think, is big. The fact that it can handle images and videos is crazy now. I tried to have it do Where's Waldo. It didn't do a good job of that. But the fact we have a really popular hiking spot here called Flat Top, and the parking lot gets full, but there's a camera. I asked it just to count the number of cars. I was testing Claude and Gemini. But the ability to work with images and then language translation and indigenous languages is something that I'm really interested in. I have this map behind me of Alaskan indigenous languages, and some of them aren't written down. I think there's some work there where people are able to use video and find patterns. to improve translation and make sure that we include those on the front end of generative AI instead of trying to fix it towards the end. I also think the models will get a lot smaller. I'm excited to see what happens with Apple having it just locally on the phone. I am embarrassed by how much time I spend trying to find an email or a Teams chat, at work, and the ability to search through all of that, I think, will be great. I think the other thing that I'm excited about is machine learning and some of the work from Google, like DeepMind, able to detect breast cancer 4 or 5 years earlier. Some of the really not predictive AI, I mean, predictive AI, not generative AI, but some of the breakthroughs there on materials or just saying constraints like, I want a battery for the car, but make it out of sand instead of like make it out of something plentiful. I think some of the complex problems we've had are just, from my own research, they just have so many variables, they're just impossible to hold in your head. So using AI to find these patterns, I think, can be great. But there are downsides, and, you know, we have to be worried about misinformation and a bunch of other things. I'm super excited about it, but I also and pragmatic that there are challenges.

I think I had another CIO say that the potential really is now that we can free people up to do the most rewarding work. I think that is the real upside. Much of what you talked about, we'll begin to see as these copilots come alongside us.

Yeah.

That'll be pretty cool to see how much is freed up, what does that mean, and what impact does that have to operations and service service delivery. And I think of myself, like, if I put myself in the— like, I'm a Health and Human Services caseworker, the opportunity, like, that would draw me into that space would be the ability to meet with people and make an impact.

Right.

Not to be bogged down in some manual process. So the ability to abstract a little bit of that out of my day and then put me in front of a place where I feel the most value and the most rewarding impact, that's got some real impact to the public sector workforce. And I think that that's going to be something fascinating to hopefully see come to fruition.

I totally agree. I think that's a great point because one thing we do on our innovation team is we go and talk to a department and talk to someone who we're like, what's the boring part of your job? What's the thing that you put off to the end of the day because it's just monotonous? And we want to see if we can automate that or figure out a process to have that go away. I think these assistants, I will ask ChatGPT to rewrite something 20 times. I would never do that to a colleague. It would just seem weird, but there's an infinite patience and the ability to do that. Ethan Mollick is, I think, in Pennsylvania. He has a book, Co-Intelligence. I'm halfway through that, but he has really interesting thoughts about having it augment your abilities. Then you mentioned Copilot. I can actually write computer programs, but I'm rusty on some of the new languages. Asking it to just write Python for me, or I was going to— I was giving a presentation, I wanted an interactive website, and it was the day of the presentation, and I'm like, oh, I'll just see how ChatGPT does. And it did a pretty good job. And it was, I think, in education or interactive stuff, it'll be great. And then one thing I'd love to see for government, I don't think we're quite there, is rewriting really old code like COBOL, and just stuff where people have left the company into modern languages. And I think we're pretty close to that. We need to be able to write reliable test cases. But if you ported all your software to newer software, I think it'd be easier to maintain. And you could also attract younger, younger workers that know Python or R or aren't necessarily in the COBOL or Microsoft stack. I think they'd have more fun with that stuff and you can maintain it.

Well, more fun indeed. This interview itself has been fun. I want to thank you, Brendan, for taking a little bit of time out of your day on this Friday to join us on In Case You Missed It and share your insights about where generative AI may be heading in local government.

Yeah. No, thank you so much. This has been great.

Where does our audience, in addition to the link that's on the screen, where do they go to keep up to speed with you?

They can check out my LinkedIn. They can check out the municipality of Anchorage. We also have an Anchorage iTeam LinkedIn account. I try to tag stuff there, not really posting a ton from there, but I would also recommend San Jose's Gov AI Coalition. They have something on LinkedIn. And then Johns Hopkins University City AI Connect. There's just lots of different groups organizing around it. Bloomberg Philanthropies is doing a lot of work. I went to CityLab last year in DC. They had a thing for mayors on generative AI and just helped bring everyone up to speed. Tons of people doing great work. I would say South America is doing some great work in Latin America. I was on a webinar with someone from Buenos Aires and she was fantastic. They had chatbots that were getting like tens of thousands of throughput each month, so they were able to kind of analyze that stuff. One thing she brought up that I hadn't thought about is benchmarking new AI because you want, you want the one that's the best. And how do you test that with stuff moving around so much. But yeah, those are— and then Santi in Boston is doing amazing stuff.

Exciting times indeed. So we'll drop those links in our show notes. But Brendan, thanks so much for joining us. And until we see you again, have a wonderful weekend.

Thank you so much. This was a blast. Thanks, Joe.