AI, Google Cloud and Sentiment Analysis

Happy Friday. Welcome to In Case You Missed It, GovTech's government technology news roundup, where we take a look at some of the biggest news from the past week and give you our thoughts on what it means to government, residents, and the market. I'm Joe Morris, and I'm joined by Jed Presgrove. Dustin's out today, but our special guest is Quinn Chazen. He's the Strategic Business Executive and the Head of Customer Analytics for the public sector with Google Cloud. He's here to answer our questions about how local and state leaders are using artificial intelligence and other emerging technologies to gather data and solve old problems in new ways. Today's episode is brought to you by Google Cloud. Google products, including Google Cloud, can help government agencies improve citizen services, increase their operational effectiveness, and deliver proven innovation. Google Cloud has helped governments worldwide respond to the COVID-19 pandemic. Their customers include the Chicago Department of Transportation, the City of Los Angeles, and the City of Chattanooga, the State of Arizona, and many others. For more information, visit cloud.google.com or read our show notes on govtech.com. So without any further delay, let's bring on Quinn. He's been at Google for more than 8 years now, but currently is the Head of Customer Analytics for the public sector within Google Cloud. Basically, if you're a state, local, or federal government agency looking to integrate with Google on a data project, he's going to be your point of contact. And here's a fun fact, when he's not busy at Google, he's also the founder of Meat Empire, a South African-style jerky company. But that's probably a discussion for another day. Welcome, Clint.

Thanks for having me, Joe.

So that's, uh, obviously been a busy couple years given kind of that outline there in terms of the pandemic and your efforts. But before we dive into that, can you tell us a little bit about yourself, your background, and your role at Google Cloud?

For sure, yeah. My name's Quinn Chazen, like you mentioned, and I've been at Google for 8 years, basically splitting half my time between the digital media advertising, ad tech, martech side of the house and the cloud side of the operations. Before being on cloud, I led all of our technical media engagements with the federal government, state, local governments. So through that work, I've been working in public health basically the entire time. trying to bring services to constituents and then kind of grown the aperture from there, extending to other government social services, DMV services, post office, you name it. You're engaging with your public sector agency. I think it's part of my role and my team's role to help make services easy as possible for you and make government come closer to your needs. And so that's really what we're focused on every day.

Quinn, thanks a lot for sharing those details about yourself. You know, going back to artificial intelligence, you know, this has been a captivating technology for state and local leaders, but often agencies struggle with the right use cases. Can you share how Google Cloud has helped operationalize AI to aid in the public health domain?

Yeah, absolutely. I mean, it's such a big field between cybersecurity and privacy and constituent services. Specific to my domain, I think when I think about AI, to me and the customer base that I work with, it's really about bringing the human touch, the personalized touch to— from your agency to the constituent, instead of having someone go through a laundry list of services to get what they need. How do we bring a conversational tone to a chat agent, to a call center, to bring the service closer to the constituent? How do we use AI to understand folks' reactions on our websites, through advertising, to to bring them a personalized service or recommendation directly to them. So when I'm thinking about AI, there's a part of the fundamental, fundamental basic process. We're integrating data from different sources that is difficult to combine naturally. We're trying to organize it that way. We're doing back office work to extract information from documents maybe that we need AI to pull entities out of, right? But I'm more excited on the front end about how do we use that confluence of information to take the higher-level advantages of AI to improve service outcomes without having to hard-code each step of the way in policy.

Yeah, very interesting, uh, Quinn. So I wanted to talk a little bit about sentiment analysis. I was wondering if you could talk about, you know, sentiment analysis, how it relates to AI, and maybe, you know, why is it such a critical data point for state and local agencies?

Yeah, and I think when we say sentiment analysis for Google Cloud, what we really mean is diving through the corpus of feedback that you are getting as an agency. It is used in a wide variety of areas. In a public health domain process, what we're really trying to do is take all the feedback you get in your natural language. This could be from call centers, it could be from search on your website or search from Google, surveys that your epidemiologists or other folks are running on the ground, bringing that all together, and using tools to say, hey, what is the topic of discussion happening in real time? How is that relating to the policies and programs that we're putting out? And how is the public reacting to that at a local level in multiple languages in real time, right? And so when we understand sentiment analysis, it's sort of on those 2 vectors. It's sort of what the conversation is, what it's about, and then generally who is being impacted and is it a positive or negative feedback? People really happy about this topic, really unhappy about this topic. So I think it's a general term that we use, but behind it can be a really powerful mechanism to understand the confluence of touchpoints that your agency has with constituents that might not be in places that you would expect, but places that are really, really valuable and useful for agency missions.

That's great. You know, you look at our national benchmark surveys across states and cities and counties and pre-pandemic, even post-pandemic, we've seen the rise of what we call the government experience, right? The experience that you have as a resident, that I have with our resident across all government. services. So with government leaders being more aware of that concept, you can certainly see the use cases for sentiment analysis growing. What has the impact been for agencies that have tapped into it specifically? I know in your domain of expertise it's public health, but maybe you can share with our audience some of those specific use cases.

Yeah, absolutely. You know, we have an example here of some work we've done in the state of California. during the COVID-19 pandemic. I won't get too deep into the weeds here on exactly all the pieces that's going on, but you can see each line in here represents essentially, for lack of a better term, a data pipeline between different services to integrate them to help tackle the vaccine distribution effort that California and the rest of the world had to go through. And so what we're really doing is bringing together data that is very difficult to ingest in real time, but making it available to make those hypergranular decisions. In that example, not only were they using it to understand the local sentiments and feedbacks at a very local level of people's reaction to the public policies, non-pharmaceutical interventions like mask mandates, vaccine distribution efforts that were going on, and who that was impacting, but then taking that information and making it very actionable to folks really at the front lines of that, of the issue, is something that we're really focused on. And so all of this is abstracted away from you, and our goal is to really help provide you the output of that, which is really a policy decision device, right? And so when we're talking about what it's used for, a lot of times it's— I think about it as logistics decisions, right? You have a limited set of resources, you have a limited set of public sector folks, like what is the most effective place to put those resources today? in a way that might change tomorrow. And so whether it's digital media properties or whether it's actual, you know, boots on the ground for services, the, you know, understanding the reaction that constituents are having to those policies and then giving that closed-loop reporting back to the public sector stakeholders is, I think, key to making progress in any public sector initiative, public health or otherwise.

You know, anytime you mention any emerging technology, artificial intelligence or, you know, natural language processing, robotic, you know, any— anything, you get a little bit of anxiety sometimes in frontline staff. But you also see those positive benefits that you just talked about, the improved decision-making. How are you seeing employees responding to having this type of data and outputs available and accessible to them?

Yeah, I mean, I think we are in a very different place than we were even 2 or 3 years ago. The outputs that we try to make from these processes, maybe 5, 7 years ago, would have to be someone with the research understanding that can hypertune a machine learning model in TensorFlow. Today, I feel like we've gotten to a place where most of our engagements, we're often serving up the end results of those models into the same tools and systems that public sector agencies are using today. We have an example here of 3 output use cases from that example I gave before where These 3 groups and bubbles represent basically an advertising media firm, local health jurisdiction and counties which do not have advanced analytics capabilities, and rapid public health outreach teams, boots-on-the-ground teams, none of which are familiar with building and maintaining AI models. But the media team is familiar with media tools. We can get the results of those models plugged into tools like Google Analytics and advertising to do public health information campaigns. We can share the results of dashboards, what sentiment changes at a local level to local counties and jurisdictions that can plan their own work from there in simple outputs. And we can plug into mobile applications just to help plan, you know, turf coverage for groups that are doing rapid development and rapid distribution of public health goods. In COVID cases, the vaccine, but, you know, in opioid context, we've done the same with Narcan and other efforts. So you can see how, while it's very complicated work upfront to build those AI models with the teams that we're working with, You know, Google makes that process really easy and really scalable. And most importantly, we're trying to make it— the outputs in a way that, you know, you're not diving into and tuning each piece of the model. We're getting that information into the tools and services that you use today, so you don't have to sign out and sign into something totally different. You know, if you're comfortable working with the tools that you're in, we want to bring that data to you. And so today I think it's an entirely different place, and you really shouldn't think of AI or automation machine learning as something threatening. We really want it to be a force multiplier for the work that you do today.

Quinn, going back to the idea of like someone maybe having reservations about this kind of technology, you know, many agencies that have been interested in the emerging AI technologies like natural language processing, which allows for sentiment analysis, sometimes they struggle with understanding what the implementation process looks like.

Yeah.

Could you share some insights into this for those that struggle with that?

For sure. I think the good thing about Google Cloud and natural language processing is it's the same techniques that happen under the hood for our own 9— 1 billion user products, you know, Google Search and YouTube. So it has to be lightweight and scalable. And for us, what that really means is typically the typical approach is we'll come in and it'll be a managed service of quick 6 to 8 weeks to pull together the data sources, right? And then we're sitting with you to help you understand the outputs that you need. And so you can almost think about this in the same way you have— you would have done a classic ETL problem before. We need to take data from one place and make it work with a different system. We need to transform it to, to play with our semantic database. Now I think you can do similar work with Google and our partners when it comes to AI and machine learning models. Hey, I need to understand XYZ about my constituency, and I like to use CRMs as my interface. I like to use dashboards as my interface. That's really the information that we need in order to just help you get started and go down that path. From a maintenance standpoint, you know, a lot of this work on the backend is actually either no ops, totally serverless environment you can do with Google Cloud, or if you have some, you know, really intense 100 million terabytes of information that you're working with, you know, it's a managed service supporting that kind of infrastructure work. But I think with Google, not only can you manage this, I think, serverless and fairly easily. But, you know, if you have simple datasets you're looking to work with, our data warehouse BigQuery, you can do it with simple SQL. And so we're trying to lower the bar for difficulty of working with AI and machine learning tools to something that, you know, a traditional database specialist could work with today. And so I think getting from that to production is, is much easier than, than you might think.

You know, earlier you talked about you know, AI being a force multiplier, and we've certainly seen that pop up in our conversations, but it also feels like not too long ago we were talking about big data in the public sector. There was a lot of, you know, fear and doubt, like how the heck am I going to hire a big data scientist when I can't fill these other vacant positions? And to your earlier point, it seems like, you know, solution providers like Google have filled that gap by building these, you know, tools that are much easier to use so that way you can kind of get over that hump, if you will, or hurdle in terms of hiring these AI experts. But what would your advice be to agencies that don't have employees that are well-versed in AI or machine learning, but they're looking to build that expertise, they're looking to build that capability, they're looking to take that first step into artificial intelligence deployment? Where should they start? What would your first steps be?

Yeah, you know, I mean, I think from Google's perspective, the focus on the mission is, is always somewhere that we recommend starting. You know, if you think about just the ability to pull together information and focus on the areas that are constituent-related, I think when we're trying to bring that information together and sort of teach you how we've created the models, how we've gone through this work, right, you kind of have that back and forth relationship. You know, it's something that we can teach you how to kind of stand up and add more datasets as you go along. But I also think in a way, you know, when we were talking about big data 5, 7 years ago, you know, it was on the beginning of the mobile digital revolution, and the amount of data coming at us is doubling, tripling, quadrupling every year as we've offloaded so much of our public decision-making processes as constituents into the digital domain. And so that glut of information It's just really expanding faster than we've ever seen it before. And so, you know, you don't, even if you have an AI expert, they may be somebody, you know, who's worked in a field adjacent to what you're actually trying to do, right? If you're someone in social media and digital, but you actually need AI machine learning to dive through medical records, right? AI is, I think, not something as unified or thought of in that same way as much as it used to be. So I recommend starting with the mission outcome, working on that and sort of working backwards from there. Rather than saying, hey, I'm gonna hire 10 generalist data scientists who have to deal with just streams of data coming at them for the rest of their lives, right, as the digital ecosystem keeps expanding. And so teach to fish, I guess, is the way I would think about it, but work backwards from the mission is always our perspective.

I love those 2 points. How also has the technology become more accessible, more user-friendly?

Yeah, I mean, from— and like I was saying before, from a build perspective, the ability to do this in simple SQL, Python, things like this without having to go in, in, into the managed server capacity has gotten much simpler. I would say Google has sort of 3 flavors of machine learning and AI capabilities now. There's the one-click button for things if you have structured data and you, you know, for example, You know, how you're digging into your call center and you're trying to compare it to CRM data that's both very structured. These are things that you can literally just click a button and the tools will do a lot of the AI work for you. There's the second layer of simple SQL you can build. In Google, we have some Vertex AI and AutoML, these capabilities, contact center AI and translate services that assist you in developing those models. The natural language processing tools out of the box will get you 40-50% of the way there. Immediately. And so I think from a user-friendliness perspective, we're always trying to abstract that difficulty up to make you ask the more specific questions without spending 80% of your time putting data together, cleaning it, organizing it to prepare for machine learning. You know, we're trying to abstract that issue away in the backend to allow you to spend more time sort of on the model development or tweaking the development of the models that, that we kind of spit out, out of the box. And so You can see more things coming from us. We— I know like Vertex AI is sort of the managed ecosystem from Google if you have a synthesis of a lot of machine learning AI models together that you're trying to harmonize or trying to make work in production. But having that in a unified plane, so even in a serverless environment, you still have that same feel of an infrastructure as a service, same feel as sort of the management capability without having to do that legwork, I think is an automation step that, you know, classic machine learning folks would Quinn, I know you watch this show.

I know that. And as you know, and in case you missed it, we're always looking to the future, you know, what's on the horizon, what's coming down the road. So what is next for sentiment analysis in the public sector?

Yeah, you know, as we were talking about before the show, it's, it's been crazy to see the development of this work throughout COVID-19 and how many different applications of AI/ML we found through that process. And that Google supported with our Intelligent Vaccine Impact solution, from sentiment analysis, understanding vaccine release, to forecasting and modeling efforts, to translation services and getting people care and support that they needed. I think public health now has a playbook to take that to other areas. We work very closely on the opioid epidemic with other groups, right? We've also worked very closely on back-to-work initiatives and social services and labor. So all those areas that were immediately impacted with COVID were the first users of a lot of these services. Which ironically were some of the areas where you would think would struggle the most with it, right? So I think that those parts of the government have a lot to teach other core services that have, I think, experimented less with AI and machine learning but are very central, from food benefits to DMV services to transit and departments of— departments of transit and climate change. These are all areas that we've started having conversation recently but I think have a lot to learn from this process and have a lot of upside in the potential to do good. So I'm really excited to see those kind of fields expand, and our team is definitely working on that every day.

That's great. Where can our audience go to learn more about Google Cloud and the capabilities that you've outlined here today?

Absolutely.

You know, cloud.google.com is where all the fun stuff will continue to be posted. We'll link out to the public sector blog, and if you follow the path there, we can kind of see what we're doing in the public sector as you go along. But feel free to check out examples from Oklahoma and opioid epidemic in California and as it relates to COVID-19 for some examples like this, or New York State's Excluded Workers Fund are all great examples to poke through. And those sites will, will lead you to a plethora of other, of other capabilities that we're working on.

And where should our audience go to stay in touch with you, Quinn? Any social platforms that you're on they can connect with you?

Yeah, absolutely. LinkedIn is always a great one. Feel free to reach out anytime. And if you're in the public sector, happy to connect and chat over email and set up some time to chat, or even if you're interested to learn more about AI and machine learning, I'm always interested in talking about this stuff. So this stuff's fun.

Well, great. Thanks, Ann and Quinn, for joining us today. That's all the time that we have for this episode. We want to thank you all for watching, and we want to invite you to tune in next week at 12 o'clock Pacific. You can tune in on LinkedIn, YouTube, and Twitter. For more In Case You Missed It content. Until then, have a wonderful weekend.

Thanks, Joe.