Introduction
States, cities and counties across the U.S. are doing fascinating things with artificial intelligence and machine learning (AI/ML). Sensor networks fuel learning models that fight traffic congestion. Digital assistants help residents navigate agency websites. Remote cameras flag wildfires to protect forests.
This growing array of use cases has compelling implications for state and local governments. But for all the value of AI/ML technologies, you can have a hard time choosing what’s best for your agency — especially when new options arrive on the market just about every day.
This go-to guide will help you make sense of AI/ML opportunities. We share advice from industry experts and public sector IT veterans to help you find the right fit for your community. You will learn more about:
- Creative AI/ML use cases.
- Emerging issues like the rising role of edge computing and crucial factors that impact AI/ML performance.
- Best practices for planning and implementing AI/ML solutions and iterating to drive long-term success.
Exploring AI/ML Use Cases
In 1921, the Nobel Prize-winning historian Christian Lang famously observed that technology “is a useful servant but a dangerous master.”
“We’re converting technology back into becoming a servant,” says Michael Shepherd, Senior Distinguished Engineer with Dell Technologies. Learning algorithms teach AI applications to be more helpful to the public and to create predictive insights. Autonomous agents relieve workers of everyday chores. “It’s hyper automation that takes us to an entirely new level of human productivity,” Shepherd says.
This is happening in places like Amarillo, Texas, where an AI assistant named Emma talks to residents and guides them to city services. In San Jose, California, AI chatbots simplify the process of reporting problems or seeking information through the city’s 311 system. Maricopa County, Arizona, uses AI/ML to untangle traffic congestion in the Phoenix metropolitan area.
Use cases like these are popping up in communities across the U.S. Agencies are creating new services by tapping structured data from devices like traffic signals, and unstructured data in written reports and live audio or video feeds. They’re deploying generative AI (GenAI), which uses large language models (LLMs) to create text, images, sounds and other content. And they’re turning to predictive AI to analyze datasets, identify useful patterns and project the statistical likelihoods of certain outcomes.
While GenAI apps like ChatGPT have garnered massive media attention, predictive AI apps are showing promise in areas like public safety, traffic management and utilities monitoring. In the years to come, AI/ML solutions will most likely fuse generative and predictive AI as agencies find innovative ways to deploy learning automation.
Interactive self-service
GenAI is simplifying and streamlining civic engagement. “If you go to a city website, the last thing you want to do is figure out what department is going to solve your problem,” says Dustin Haisler, President of e.Republic, parent company of Government Technology magazine. A GenAI chatbot can take you where you need to go without navigating through page after page of website links.
“I really like what California is doing with its revenue and tax call centers,” adds David Fletcher, Senior Fellow with the Center for Digital Government (CDG). An AI agent listens to a call and gives the call center agent facts on legislation or regulations relevant to the caller’s intent.
GenAI can also personalize government services, which is especially helpful in communities with large non-English-speaking populations.
Natural resources
AI applications can help protect forests and control wildfires. In the Western U.S., for instance, the ALERTWildfire network relies on AI to spot blazes in Washington, Oregon, California, Idaho, Nevada and Arizona.
“AI is analyzing images from cameras on mountaintops and in remote locations around the clock,” says Sean McSpaden, a CDG Senior Fellow. “There are so many images and cameras that it’s not possible for a human to analyze all that data,” he says. The data also can help responders determine how many people and machines they need to manage a fire.
Physical infrastructure and utilities
AI can monitor and measure changes to bridges, traffic signals and other infrastructure. “We have a city that puts cameras on municipal vehicles,” says Jumbi Edulbehram, NVIDIA’s Global Business Development Leader for Smart Cities and Spaces. “When they drive around, they automatically detect and classify road conditions.” Potholes and other issues are reported immediately so they can be fixed before anyone even files a complaint.
AI algorithms also support predictive maintenance that reduces costs, boosts efficiency and extends the useful life of critical infrastructure.
Transportation and urban mobility
AI algorithms pull data from interconnected sensors on traffic signals to generate real-time insights that let agencies optimize traffic flow across a metro area. “It totally changes the game and unlocks all kinds of new efficiencies,” Haisler says.
For example, an AI-driven program in Phoenix, Arizona, and surrounding Maricopa County cut traffic congestion in half. AI tools can also manage curb space, guiding rideshare drivers to available parking spaces and billing delivery drivers for double parking.
Public safety
Traffic cameras and surveillance video feeds produce hours of visual data that may contain only a few minutes of useful evidence. “GenAI allows you to search large quantities of video and data in real time to identify suspicious activities and respond to them faster,” Edulbehram says.
McSpaden notes that the San Francisco Police Department uses AI to analyze crime data, identify patterns and predict potential hotspots. Cities can also use GenAI to translate incoming 911 calls and give dispatchers context on callers, neighborhoods and other vital data. These tools add up to faster assistance and opportunities to save lives.
5 AI Insights from the City of Amarillo
Amarillo is the largest population center in the Texas Panhandle, with more than 200,000 residents who speak English, Spanish and many more languages. The city’s AI-driven digital assistant initiative underscores best practices for state and local governments.
Give AI a human angle
Emma is an AI-powered avatar and digital assistant who can converse with city residents. People can visit the city’s website and ask her questions in sentences and phrases of their own choosing. She offers replies and points users to helpful pages on the website.
Solve everyday problems
“I have one middle school speaking 62 languages and dialects,” says Rich Gagnon, Amarillo’s CIO and Assistant City Manager.
Start with safe data
Emma uses GenAI based on large language models (LLMs). A familiar concern about GenAI is that users might unintentionally insert personal data when asking questions, and that their data might be used to train the LLMs. “A common first step is to avoid dealing with highly secure data,” says Dell’s Shepherd. Emma draws upon public data published on Amarillo’s website, so there’s no concern about security or privacy breaches.
Take time to clean up your data
GenAI and LLMs produce much better outputs when they’re based on clean, accurate data. A city website could have pages that are out of date, inconsistent and littered with errors that degrade the performance of a chatbot or digital assistant. “People often find out that their data is way worse than they realized,” Shepherd cautions. Devoting time and resources to cleaning up data as early as possible will pay huge dividends in the long run.
Expect pleasant surprises
Every city wants feedback from residents, but they often find that people don’t want to fill out constituent satisfaction surveys. With Emma, however, Amarillo residents are eager to share their opinions. “When you put a conversational system in production, you’ll find people are very vocal about what they like and don’t like,” Shepherd says. Amarillo is reaping the benefits of this phenomenon. “They’re shocked at how much data they’re getting back that they never could get through surveys,” he adds.
Key Decisions: Technology, Partnerships and Performance
Successful AI/ML adoption requires a mix of bold experimentation and cautious deliberation. With AI-related computing moving to the network edge, agencies need to rethink their data center infrastructure. Government IT leaders will need partners and collaborators to help them adopt best practices and avoid common missteps.
1. Move computing to the network edge
Network edge devices like sensors, cameras and smartphones have more storage and computing power than ever before. Many also include AI accelerators that enable them to handle heavier computing loads. These trends make edge computing increasingly relevant.
“You want to bring AI to the data rather than data to the AI,” Shepherd says. AI-powered computing devices can run millions of calculations in parallel. From a user-experience standpoint, it makes far more sense to perform those calculations on an edge device instead of transmitting all that data over a network and running the compute centrally or in the cloud.
For example, sensors in utility meters that measure water consumption should do most of the computing at the network edge where it’s fast and economical. Running AI workloads in a data center or a cloud instance can be more expensive and have lower performance because of network latency.
“We’re already seeing localized AI processing in next-gen 911 and other justice and public safety systems,” says Haisler. He also expects edge AI to grow in use cases for monitoring and managing infrastructure like roads and water/ sewer utilities. “I think you’re going to see AI architected and placed at the edge to deal with the potential ramifications of power outages, malicious acts and cybersecurity incidents,” Haisler says.
Growth of edge AI will increase network complexity. Remote devices will need to be configured to perform most computing tasks while delivering essential data in small bites to a central IT operation. “Now that I’ve got edge devices with storage and processing power distributed across the network, I have to think about not just security but also data and device management,” Haisler says. System availability, backups and restoration will require advanced automation.
In addition, agency IT leaders will need to consider upgrading to AI-ready PCs, smartphones and tablets. “People are going to see that these new devices have an order of magnitude improvement in capabilities,” Shepherd says. “They’re going to notice poor performance with older devices that rely on the internet and bandwidth.”
2. Form partnerships and find collaborators
Partnerships across government and industry will be essential for deploying and scaling AI solutions. Technology hardware vendors like Dell and NVIDIA are teaming up to provide computing devices to support AI.
“We work with all these partners because you really need to optimize the software and hardware stacks from the edge to the cloud to make sure you’re getting the most out of the technology at the least cost,” NVIDIA’s Edulbehram says.
Cloud vendors are a crucial link in AI/ML partnerships because they have the capacity to store massive datasets and the scale to match just about any computing requirements. “States and larger localities may want dedicated connections to their cloud partners and then support for software-defined networking,” says CDG’s Fletcher. System integrators and solution providers with deep experience in the public sector can help agencies find the right balance between on-premises and cloud solutions.
Agencies can also turn to information-sharing organizations for guidance on AI/ML solutions. The Government AI Coalition in San Jose, California, for instance, provides a starter guide for AI initiatives and hosts AI summits that attract government IT leaders from across the country.
Effective collaboration can shrink AI development timelines. “We have to learn from our colleagues in other state and local governments and education entities,” Fletcher says. “If we share our findings, people will know it hasn’t just been tried once, it’s been tried a half-dozen times.” Agencies armed with proven performance data may be able to move faster to early deployment or production of AI/ML solutions.
Agencies can crowdsource data quality issues by inviting the public to correct inaccuracies.
3. Optimize AI performance
AI/ML solutions should yield measurable results that advance an agency’s strategic goals. That’s not always easy given the complexities of data, hardware and software. Consider these key AI performance issues:
Reliability
People must trust AI automations. That means going the extra mile to double check the reliability of information from AI algorithms. GenAI systems are vulnerable to producing inaccurate, unfair and insulting outputs. “Often, there’s a lack of high-quality data that can lead to the models not being robust or performing poorly,” Fletcher says. Data should be inventoried and classified before being exposed to training models. Agencies can also crowdsource data quality issues by inviting the public to correct inaccuracies.
Compatibility
New AI solutions need to integrate with existing technology stacks. “Most of our applications reside on legacy infrastructure, and our data centers aren’t particularly positioned for expansion of AI solutions,” McSpaden says.
Productivity
Agencies want AI applications that drive the biggest benefits at the lowest cost. They also want solutions that automate everyday processes and free staff from time-consuming manual processes. Data sharing can boost productivity considerably.
“An enterprisewide API platform can give you common ways to share data operations,” Fletcher says. Agencies can also create centralized, consolidated data repositories to avoid repetition and standardize IT processes. “Make sure it’s not just one agency getting all the benefits,” he adds.
Sustainability
Agencies must consider the energy demands of AI solutions. “Think about solutions like graphics processing units (GPUs) that reduce computing power requirements while delivering enormous computing power,” Edulbehram says. Partnering with cloud providers that use high-efficiency data centers can also help agencies offload a portion of their AI-related power demands.
Scalability
During procurement, make sure you understand AI workloads and can build a scalable, flexible technology ecosystem. Cloud solutions can help you scale up early in the life of AI applications, but you may need to move workloads on premises to control costs as your AI solutions mature.
Best Practices for AI/ML Planning and Execution
These tips from our industry experts and government IT leaders can help state and local jurisdictions succeed with AI/ML initiatives.
1. Create a strategy
Your AI/ML strategy should include planning for acquisition, implementation, feedback and long-term maintenance. Key points to consider:
- Collect ideas and build buy-in. Convene key stakeholders to identify high-level objectives. Confer with consultants and professional colleagues to understand obstacles and opportunities. Nurture leadership buy-in by keeping executives and elected officials in the loop on your plans and progress.
- Prioritize. Base your goals on the most pressing needs for improvement. Before choosing an AI solution, make sure you understand who will use it, how they will benefit and any risks you might create.
- Target use cases. Identify about a half-dozen outcomes you’d like to achieve for things like digitizing citizen services and automating agency operations.
- Seek simplicity. Zero in on one or two simple, low-risk pilot projects you can build upon. Avoid projects that use sensitive or private data or raise safety concerns. Avoid overcomplications: The more variables you add in the planning phase, the more difficulties you’ll raise throughout the project.
2. Map out your implementation
- Understand cost. Analyze solutions by looking at the total cost of ownership. Make sure you understand how much computing power you need for AI workloads and the potential costs in areas like software licensing and cloud resource consumption.
- Account for infrastructure. Identify infrastructure needs such as storage, networking, security and business continuity/backup.
- Measure and monitor. Build mechanisms to monitor results and provide actionable insights. This data will be essential to strengthening leadership buy-in.
- Establish oversight and transparency. Create processes to ensure humans are involved in AI-related decision making. Keep the public informed about choices you’re making and establish transparency standards and practices.
3. Anticipate challenges
- Create an inventory. Assess your infrastructure and technology stack and identify likely areas that might require investments in hardware, software, automations and skills.
- Address resistance. Encourage adoption by appointing champions who will help colleagues understand the value of AI applications. Reassure agency staff and the public that AI is to augment humans, not replace them.
- Assess your technology skillsets. Note that many GenAI applications do not require advanced data science or programming skills. Agency staff can be trained to use GenAI apps and improve performance.
- Break down data silos. Some states have passed legislation requiring data sharing across all state agencies. This is a good model in any organization. Find ways to encourage data sharing among departments or agencies.
4. Update policies and governance
- Create living documents. Governance frameworks and policies need to evolve as new use cases pose new questions. Revisit policies and governance regularly. Avoid technical jargon and strive for clarity.
- Focus on fairness. Test GenAI apps to ensure they’re not generating biased or offensive results in chatbots and other tools.
- Disclose AI usage. Agencies need to be transparent about how they are using AI and GenAI.
- Address cyber risks. Establish security protocols to prevent leakage of personal data into AI training data sets. Security teams will need training on how AI can be used to infiltrate systems.
5. Cleanse your data
- Acknowledge data challenges. A GenAI app trained on an agency’s public data might encounter multiple documents covering a similar issue. Inconsistencies across these documents could cause inaccurate outputs. Agencies may be surprised to discover the scope of data accuracy issues.
- Commit to the value of clean data. Cleaning data may require significant investments in skill, time and effort. But AI applications will operate far more effectively — and deliver more value — if you make this commitment and see it through.
- Crowdsource the cleanup. Ask volunteers to contribute to AI apps in the development phase and put them to work finding errors and suggesting corrections.
- Let AI help. GenAI queries can help identify data issues and suggest fixes much quicker than people can do on their own.
6. Iterate and elevate
- Accept good to great. Seeking perfection can paralyze your progress. Start with a project where AI can deliver documented value and build on what you learn in the early stages. Once you’ve worked most of the bugs out of a solution, apply your learnings to new iterations of the app.
- Gather feedback and continuously improve. GenAI apps can be excellent feedback tools. You can use them to ask users for recommendations on current services and comments on the performance of everyday operations. Conversational interfaces encourage people to provide information they would rarely offer in a survey.
Conclusion: AI/ML success comes down to people, data and technology choices
Successful AI/ML deployments involve multiple decisions. Agencies will need to address data quality and accuracy as they deploy GenAI. They’ll also need to think through the ramifications of pushing AI computing to the network edge.
To get started, focus on low-risk use cases that can produce measurable value. Start small and grow these initiatives as you gain experience. Find strong partners with proven public sector track records to help your organization navigate AI-driven changes in culture and technology.
Remember that effective use of AI/ML starts and ends with people. Your staff needs to understand how AI can help them handle growing workloads and solve complex challenges. There’s also no substitute for guardrails, governance and disclosure to ensure responsible and ethical use of AI technologies.
“Public trust can be eroded pretty quick if people think AI systems might be operating without transparency or accountability,” Edulbehram says. “So having humans in the loop and human oversight right at the start is super critical.”