The next wave of connected and intelligent technologies, including sensors, 5G, and artificial intelligence (AI), holds great promise for improving the energy efficiency of many systems, including urban systems. Building automation systems can automatically monitor, control, and optimize a building’s heating and cooling, lighting, and other mechanical systems. Real-time traffic data coupled with smart traffic lights can reduce energy use. Digitalization is also enabling integration of previously isolated systems: Grid-integrated buildings provide demand response to the grid; and smart electric vehicles (EVs) shift their charging times to off-peak hours.
At the same time, cities are increasingly making their own climate commitments and looking for ways to reduce their own emissions—and the emissions of businesses and residents who live in cities. Alliances such as Climate Mayors, a network of 465 U.S. mayors, and the Global Covenant of Mayors for Climate and Energy, which includes 172 U.S. cities, represent the growing movement toward local action on climate change.
By embedding smart technologies in the grid, buildings, and transportation systems, cities can reduce their energy use and emissions. A 2018 McKinsey report finds that a city deploying smart city applications “to the best reasonable extent” could reduce its total emissions by 10 to 15 percent. Similarly, Microsoft and PwC found that AI-enabled decarbonization technologies could reduce the carbon intensity of the global economy (figure 1). These applications help cities plan and govern more efficiently, reduce their energy use and emissions, attract and support businesses, and discover new sources of revenue.
Figure 1: Carbon emissions intensity in a “business as usual” scenario compared with AI-enabled decarbonization
But cities are facing revenue shortfalls as a result of the COVID-19 pandemic, which is stalling smart city investments. Even the most capable cities struggle to evolve into smart cities, because cities are ill-equipped to overcome the key challenges limiting smart city development. The first challenge is research in the underlying technologies for smart cities is a public good. Few want to bear the costs of “going first,” when the benefits mainly accrue to others. Second, few cities have the tools to share data with one another, which hampers the development of accurate AI models. Third, cities have little incentive to bear all the risk of failure involved in adopting technology fueled by emerging technologies. Finally, without a federal data privacy law, cities struggle to address unchecked privacy fears.
Smart cities offer an important opportunity to address both infrastructure needs and strained state and local budgets at the same time.
The federal government should play a role in helping U.S. cities overcome these challenges. It is able to provide funding and coordination on a larger scale than cities working individually. While the federal government has undertaken an array of activities to support the development of smart cities, these efforts have mostly been uncoordinated, and the government has had no strategic vision for AI research, development, and deployment (RD&D) of smart city technologies. Smart cities offer an important opportunity to address both infrastructure needs and strained state and local budgets at the same time.
This report examines the specific subset of smart city technologies that use AI to generate insights for cities, businesses, and individuals in order to reduce energy use and emissions. It begins by describing the range of potential AI energy applications in smart cities, including intelligent transportation systems (ITS), smart buildings and electric grids, and city operations. The next section identifies the challenges cities face in implementing these applications and includes a cross-national survey to learn how national governments around the world are helping cities address these challenges. The report then presents an AI maturity model that reflects the degree to which a city has adopted AI and provides a framework for scaling adoption across cities. Next, the report summarizes current U.S. federal government initiatives in AI-enabled smart city systems. It concludes by proposing a set of policy recommendations for Congress and the Biden administration to expand research in AI energy applications, increase resources for cities making smart city investments, and work with cities to demonstrate real-world AI solutions.
Smart cities are those that use sensors, data, and analytics to tackle important citywide issues. Most smart city applications are built around the Internet of Things (IoT)—objects embedded with sensors and connectivity to enable them to send and receive data. The massive amount of data generated by IoT devices is the foundation of a smart city, but data alone does not make a city smart. That data is analyzed and acted upon, often through AI applications, to generate valuable insights for cities and their businesses and citizens.
AI systems are computer systems that perform tasks such as learning and decision-making. There are many potential applications for AI in smart cities, and many more will arise as AI technologies continue to develop and improve. This section examines AI applications in ITS, building energy systems, electric power grids, and city operations, all of which have direct implications for a city’s energy use and environmental impact.
Intelligent Transportation Systems
Technology enables elements within the transportation system—vehicles, roads, traffic lights, signs, and more—to become intelligent by embedding them with microchips and sensors and empowering them to communicate with each other, decreasing congestion and emissions, and helping cities simultaneously cut costs and create new revenue sources. A 2015 McKinsey report analyzing over 150 IoT use cases estimates that IoT transportation applications could be worth over $800 billion per year for cities around the world.
ITS include a wide and growing suite of technologies and applications, which have the potential to reduce greenhouse gas emissions by 3 to 8 percent. Applications with the greatest potential for reducing emissions are those that would either lower the number of vehicles on the road—such as vehicle sharing, public transit, and financial incentives to carpool or use alternative methods of transportation—or increase the percentage of EVs. Reducing congestion and increasing efficiency also lead to reduced emissions.
Connected and Autonomous Vehicles
Connected and autonomous vehicles (CAVs), which integrate information and communications technologies into their systems in order to assist drivers, are poised to transform the way people move, providing cost savings for drivers and new revenue sources for cities, as well as safety, efficiency, and environmental benefits. By reducing accidents caused by human error, communicating with each other to travel closer together safely, and communicating with the roadway to establish more efficient traffic patterns, CAVs can reduce congestion, fuel use, and emissions.
Connected vehicles incorporate intelligence and sensing capabilities to give vehicles more data, connectivity, and interactivity. This increased connectivity gives drivers traffic information and provides critical safety communications between cars, enabling increased safety and reduced traffic congestion. Autonomous vehicles (AVs) include all the features of connected vehicles in addition to the ability to operate with little to no driver input. They rely heavily on cameras, radar, and Lidar to collect information on the surrounding objects, including their shapes, speeds, and distances. AI collects and makes sense of the data from these sensors, allowing the vehicles to “see” so they can avoid pedestrians and other vehicles, obey road signs and signals, change lanes, make turns, and do everything else a human driver would do.
Fully autonomous vehicles present new opportunities for ridesharing and carsharing, potentially moving from the current system of individual car ownership to a system of cars-as-a-service, wherein individuals subscribe to a carsharing service that they then use as needed to get from place to place. This would be a potentially cheaper solution than individual car ownership as multiple riders share the cost of fuel and maintenance. It would also be a much more efficient use of capital: Currently, cars are parked 95 percent of the time. Carsharing services would get more use out of each vehicle, especially a fully autonomous vehicle that could drive itself to pick up passengers.
The energy and emissions impacts of CAVs are highly uncertain, depending in large part on users’ behavior in response to CAVs. As automated vehicles eliminate the inconveniences of driving (e.g., driver stress, more productive use of travel time), overall passengers and vehicle miles traveled could increase, along with emissions. On the other hand, shared and autonomous transport could reduce the need for vehicle ownership, facilitate vehicle right-sizing, and better integrate with mass public transit. A 2016 report finds that automated vehicles could reduce total U.S. road transport energy use by as much as 50 percent (in an optimistic scenario) or increase it by 100 percent (in a pessimistic scenario).
Traffic Management Systems
Intelligent traffic management systems can help smart cities reduce congestion by making traffic flows more efficient, so vehicles can reach their destination faster. Technologies such as intelligent traffic signals eliminate a major source of congestion: vehicles stopped at intersections. Meanwhile, different road-pricing schemes such as congestion pricing and high-occupancy toll lanes can bring down the total number of vehicles on the road, thereby further reducing congestion. By lessening the amount of time vehicles spend on the road, traffic management systems reduce commute times, fuel consumption, and greenhouse gas emissions.
Intelligent traffic signals work by collecting information on traffic conditions from connected vehicles, roadway sensors, and other signals to adjust their timing and get drivers through intersections as expediently as possible. Unlike signals that change at predetermined time intervals or react only to sensors in the intersection they control, intelligent traffic signals are equipped with AI that collects data on overall traffic demand in the area and operates all the signals in an area as a network to identify patterns and synchronize their activity.
Congestion pricing charges vehicles a fee for driving in congested areas during the busiest times of day. The goal is to incentivize drivers to carpool, use alternative methods of transportation, or drive at other times. Modern toll roads already use information technology to reduce congestion. For example, electronic toll collection enables drivers to pay tolls automatically via a device or tag placed on the windshield, such as E-Z Pass in the United States. ITS can collect real-time data on traffic conditions and the number of vehicles on the road and adjust tolls based on the amount of traffic. More-advanced congestion pricing algorithms can use AI methods, such as reinforcement learning, that can continuously optimize dynamic toll prices to minimize road congestion based on changing driving patterns.
Smart Public Transit Systems
Another approach to reducing congestion and emissions is increasing public transit ridership. Per passenger, public transit is more energy efficient and less polluting than commuting by car. An individual with a 20-mile commute can save 4,800 pounds in CO2 emissions a year by switching to public transit instead of commuting by car—a 10 percent reduction in the amount of greenhouse gases produced by the average two-adult, two-car household.
Advanced public transportation systems can make public transit a more attractive option for commuters by making the process more efficient, convenient, and accessible. In a smart public transit system, all of a city’s buses and trains would communicate with each other through vehicle-to-vehicle communications—sharing data on their location, arrival and departure status, and overall timeliness—and using AI algorithms to coordinate their schedules to maximize efficiency and avoid delays. This real-time data collection would allow cities to construct an accurate view of the status of all assets in their public transportation systems.
Advanced public transportation systems also benefit riders by sharing information online—through a website or app, and through signs posted at bus stops and train stations—so riders can plan their commutes and get to their destinations faster. According to one study, up to 70 percent of the time people spend commuting is “buffer time,” or time riders spend waiting for their bus or train to arrive, and reducing that buffer by accurately tracking buses and trains, adjusting their schedules, and sharing real-time information with riders could lead to time savings of over $60 billion per year globally.
Smart cities can further reduce their emissions by reducing the number of gas-powered vehicles on the road and encouraging EV adoption among consumers. Meanwhile, as EV sales increase, cities can turn to smart solutions to minimize the stress EVs put on the electric grid and maximize energy efficiency.
To encourage consumer adoption, many cities are investing in charging infrastructure. In a study by Volvo, 49 percent of total drivers cited the low availability of charging stations as one of the top barriers to purchasing an EV. There are a number of factors to consider when deciding where to install charging stations, including EV range, the number of EVs in an area, and the ability of EV drivers to travel from every part of the city to every other part of the city. Researchers in Hong Kong developed algorithms that could determine ideal charging-station placement in a city; and in the future, cities could use AI or machine learning to run similar algorithms.
Finally, cities can turn to smart charging to minimize the stress EVs put on the electric grid. Increased adoption of EVs will lead to increased electricity demand. Given that many drivers would plug their vehicles in to charge at around the same time of day—for example, in the evening after getting home from a typical 9-to-5 workday—the demand for electricity would skyrocket during those times. But through smart charging stations, which enable EVs to communicate with the grid much like connected vehicles communicate with the roadway, cities could manage EV charging to improve grid operations by shifting charging times to off-peak hours.
Traditionally, the United States’ electricity distribution system has only worked in one direction: Electricity flows from power plants through power lines and substations to customers (figure 2). But the rise of smart grid technologies—the digital hardware and software embedded within the energy system, including sensors, controls, intelligent appliances, and more—is changing the way consumers and businesses distribute and consume electricity, and allows for greater informational awareness and control of energy flows. Smart grids enable decreased reliance on fossil fuels and increased use of cleaner energy sources, as well as increased energy efficiency, reliability, and security. They also provide opportunities for consumers to lower their energy bills and for cities to reduce their overall environmental footprint.
For the more than 2,000 U.S. towns and cities served by a public power utility, city governments have a direct role in grid modernization and transitioning to the smart grid. The American Public Power Association in 2018 released its smart city roadmap for public utilities, noting that public utilities are well positioned to lead in smart city programs and integrate with other city services such as transportation. Cities served by investor-owned utilities have less direct control over grid operations but can work with the local utility and regulators to pilot smart grid applications.
Figure 2: A traditional grid designed for one-way power flow (top), and the smart grid that enables two-way power flows and greater integration of electricity end use (bottom)
Smart meters and AI are unlocking new sources of previously untapped demand flexibility, and are enabling new applications including dynamic pricing and demand response. Smart meters provide real-time information on electricity consumption and enable two-way communication between utilities and customers. The explosion of data provided by smart meters has opened up new applications for AI in actively monitoring, managing, and optimizing electricity use.
Dynamic pricing can provide greater demand flexibility. Unlike traditional flat-rate pricing, in which customers are charged a flat per-kilowatt-hour (kWh) rate, dynamic pricing refers to a rate structure in which utilities set variable prices for electricity that account for real-time generation costs and local grid congestion costs. By charging higher rates during peak times that more accurately reflect the cost to generate and distribute electricity at those times, utilities can encourage consumers to reduce their demand temporarily by shifting non-time-sensitive demand to off-peak times when prices are low. AI can help both design real-time prices and enable consumers to respond to those prices. While previous work has focused on reducing peak demand and minimizing grid build-out, similar techniques could use AI to create prices that optimize for greenhouse gas emissions reductions based on the real-time carbon intensity of generation.
Smart meters enable consumers and businesses to respond to real-time price signals by shifting electricity use to off-peak periods. Smart appliances—those that connect to the Internet, allowing customers to control them remotely—equipped with AI would take this one step farther, collecting data about their own usage and communicating with a customer’s smart meter to automatically run during off-peak times. In one case, Oklahoma Gas & Electric observed up to a 30 percent peak demand reduction for customers enrolled in its dynamic pricing program. In another example, ConEd was able to avoid $1.2 billion in new grid build-out to meet rising demand by investing in demand-side solutions, including demand response and distributed energy investments.
A 2018 McKinsey report analyzing the potential impact of smart city applications in three sample cities finds that dynamic pricing and demand response could cut emissions by up to 5 percent. Significant untapped flexible demand exists, but further deployment of smart meters and additional development of AI applications will be needed to access it. In order to provide flexible demand and avoid the large capacity additions that would otherwise be needed in its absence, more electricity consumers will need access to the greater control afforded by smart meters and dynamic pricing. As of 2019, about 40 percent of the nation’s electricity meters—39 percent of residential meters, 42 percent of commercial meters, and 46 percent of industrial meters—did not have advanced two-way communications capability that would enable visibility of real-time prices and support flexible demand.
Generation Forecasting and Optimization
AI can be used to forecast and optimize electricity generation from renewable energy, for example, by incorporating hyperlocal weather data to better predict energy production from wind turbines, or adjusting the position of solar arrays based on sun and cloud positions. The Electric Power Research Institute (EPRI) has identified generation forecasting and optimization as one of the key applications of AI in the electric power sector.
Since electricity supply must match load (i.e., demand), fluctuations in generation from VRE (variable renewable energy) and demand must be forecast ahead of time both to optimize real-time electricity dispatch and inform longer-term system planning. For example, Xcel Energy, a Colorado-based utility, partnered with the National Center for Atmospheric Research (NCAR) to develop a wind-power forecasting system using AI that combines NCAR’s data from local weather stations with sensor data on Xcel’s wind turbines. The model has reduced the wind-energy forecast margin of error by 40 percent. Similarly, AI has been used to create short- to medium-term forecasts of solar power, hydropower, and other generation technologies. But while many current efforts have used generic AI, future research is needed to develop hybrid physics-plus-AI techniques that incorporate climate modeling and weather forecasting to account for how weather patterns shift over time.
AI can also optimize generation output from renewables. One of the next grand challenges in wind energy is management of wake interactions between wind turbines to minimize energy loss and manage turbine-to-turbine interactions. AI can model turbulence in the wake of leading turbines, anticipate impacts on following turbines, and optimize positioning of the turbines on a wind farm to maximize generation output. Similarly, research is underway to improve solar photovoltaic (PV) plant operations, using AI to optimize preventative maintenance activities and diagnose underperforming equipment.
Distributed Energy Resources
Traditional electricity distribution systems have struggled to maintain reliability and resilience in response to disruptions such as severe weather events, technology failures, or sudden changes in demand. A 2020 economic analysis by EBP US for the American Society of Civil Engineers estimates the annual cost of power interruptions at $85 billion.
Distributed energy resources (DERs) solve some of these problems by taking a decentralized approach. They typically rely on renewable energy sources and energy storage such as batteries to produce, store, and distribute energy to a small, nearby area. These localized systems are connected to the larger grid and can serve as backup power sources in case of power interruptions or outages. Households or buildings with DERs—for example, solar panels or wind turbines—can also sell power back to the grid at peak times. According to Siemens, DER operators reduce their costs between 8 and 28 percent and see a return on their investment within three to seven years.
Groups of DERs can form microgrids, which are connected to the larger distribution system but can also operate separately, providing emergency power to small areas such as neighborhoods, university campuses, or military bases. Like independent DERs, microgrids can produce and store energy and sell power back to the larger bulk power system. Microgrids enable smart communities to meet climate and clean energy goals while also improving their resilience to outages on the bulk power system. For example, the Fort Collins Microgrid in Colorado is part of the district’s goal to create as much clean energy as it uses.
Similarly, virtual power plants (VPPs) are networks of DERs that are independently owned and operated but linked to a central control room in order to store and sell electricity and relieve the load on the grid during peak times. The control room uses AI or machine learning algorithms to accurately forecast the demand for electricity using relevant historical data, weather forecasts, and data from the grid, and then increases or decreases production accordingly. VPPs can also use AI to track electricity prices and automatically sell energy when prices are high.
AI can also enable more decentralized methods of buying and selling electricity, such as smart contracts that make it easier for households, buildings, and communities to take advantage of DERs. Two devices would automatically execute a smart contract if one party wanted to buy renewable energy from a DER at a certain price and the other were selling energy from their DER at that price. The transaction would be automatically logged on the blockchain, which acts as a transparent and secure digital ledger that both parties can access but not change. AI can enable smart contracts by negotiating and agreeing to terms on behalf of the parties.
Residential and commercial buildings account for one-third of global energy demand and 55 percent of energy consumption. These numbers are even higher in the United States, where buildings are the largest electricity-consuming sector, accounting for 71 percent of the nation’s electricity usage and 32 percent of U.S. greenhouse gas emissions. Digitization of building energy systems can greatly improve buildings’ energy efficiency by ensuring energy is consumed when and where it is needed and improving the responsiveness of energy services such as lighting and air conditioning.
Smart buildings gather data on energy use and other daily operations and then automate various processes, from lighting and security to heating, ventilation, and air conditioning (HVAC). Sensors enable building managers to predict, measure, and monitor real-time energy performance and identify where to achieve energy savings. Active controls can lower energy use within a building while also enabling better integration within the power grid. And AI can be used to analyze sensor data to generate insights on how to optimize a building’s performance. Benefits include lower energy costs, greater consumer choice, improved reliability and resilience, improved integration of DERs, avoided electricity capacity build-outs, and reduced environmental impacts.
Smart Appliances and Building Energy Systems
Smart appliances and building energy systems (e.g., lighting and air conditioning) can greatly improve energy efficiency by reducing waste and ensuring energy is consumed when and where it is needed. Connected appliances and sensors enable homeowners and building managers to collect real-time data on the energy performance of appliances and systems and remotely control equipment. AI applications can process energy data, identify where energy savings can be achieved, predict future energy demand, and automate energy systems.
In U.S. homes and apartments, cooling, heating, and water heating together account for nearly two-thirds of total final energy demand, with lighting, appliances and electronics, refrigeration, and clothes drying rounding out the remainder. Commercial-building energy use is more evenly split between cooling, ventilation, lighting, refrigeration, and office equipment, with each accounting for about 20 percent of the whole (figure 3).
Figure 3: Total and peak-period 2018 electricity consumption of major energy end uses by building type
The potential energy savings from AI applications in appliances and building energy systems is large. A 2019 review of AI applications in HVAC systems identified maximum energy savings of up to 44 percent. However, out of 18 AI tools developed for HVAC control, only 3 functions—weather forecasting, optimization, and predictive controls—have become widely adopted.
Whole-Building Analytics and Automation
Whole-building analytics and automation integrate multiple energy systems within a building using a single whole-building energy management system to enable greater integration and systems-level efficiencies and provide greater control. Sensors and controls are the backbone of a smart building: They can be integrated into all of a building’s systems and monitoring operations, and automatically makes adjustments to optimize their performance. Building analytics is the process of converting the data collected by smart sensors into insights that can improve a building’s performance, thereby saving energy and money. AI analytics automate this process, collecting sensor data and generating insights without the need for human analysis.
One of the benefits of building analytics is the ability to perform predictive maintenance on building systems as sensors detect performance issues early on, before they cause serious problems. In addition, sensors can detect when people enter or leave a building, where people are at different times of day—and which areas have the most traffic; can inform heating, cooling, and lighting needs; and guide decisions on how to increase occupants’ comfort while reducing energy use.
Building automation uses insights from sensor data to connect and control buildings’ HVAC, lighting, security, plumbing, emergency alarms, elevators, and more. When integrated with building automation systems, AI can optimize a building’s energy use and performance. AI software can also identify where energy is being wasted and generate recommendations for building managers to reduce their overall energy use and shift their electrical load to off-peak times.
The use of AI and building automation can lead to myriad benefits, from reduced energy consumption and costs to increased security and comfort. A McKinsey 2018 report finds that building automation systems alone can lower emissions by approximately 3 percent if most commercial buildings adopt them, and by an additional 3 percent if most homes adopt them. The DOE Pacific Northwest National Laboratory (PNNL) in 2017 considered a broader set of smart energy efficiency measures, finding that integrating smart sensors and controls throughout the commercial building stock has the potential to save as much as 29 percent of building energy consumption through high-performance sequencing of operations, optimizing settings based on occupancy patterns, and detecting and diagnosing inadequate equipment operation and installation problems.
Grid-Interactive Efficient Buildings
Grid-interactive efficient buildings (GEBs) use smart technologies and on-site DERs to provide demand flexibility and better integration with the electric grid. GEBs have the ability to dynamically manage their electricity loads to help meet grid needs and minimize electricity system costs, while also co-optimizing DERs such as rooftop solar, battery and thermal energy storage, and combined heat and power with building energy systems. AI can optimize building energy systems to meet occupants’ comfort and productivity requirements, while also responding to signals from the grid to provide ancillary services (e.g., frequency modulation) or demand response.
By communicating with the grid, AI-enabled smart buildings can engage in automated demand response, which enables adaptive algorithms to keep track of energy prices and automatically run more energy-intensive operations during off-peak times when the cost and demand for energy is low. In comparison with energy efficiency strategies that are insensitive to timing, and primarily aim to reduce cumulative building energy consumption, demand response focuses on shedding or shifting electricity demand during peak hours, when electricity is usually the most expensive. GEBs also have the potential to balance electricity supply and demand autonomously on a second-to-sub-second scale to maintain power quality (e.g., frequency) (figure 4).
Figure 4: Changes in building electricity demand as a result of four demand-side management tools
For the customer, load shifting has the benefit of reducing electricity costs. For utilities, demand response avoids the build-out of additional generation, transmission, and distribution capacity that is only used during peak times, potentially saving ratepayers billions in unnecessary grid build-out. GEBs that participate in demand response can also enable greater integration of variable renewable energy from wind and solar by shifting energy loads to periods of renewable availability, thereby avoiding the need to curtail, or waste, renewable energy. New analysis from the U.S. Department of Energy (DOE) finds that GEBs could reduce peak demand by 177 gigawatts (GW) (24 percent) in the summer and 128 GW (22 percent) in the winter.
Building codes and appliance standards can also unlock potential energy savings by requiring new buildings to be demand-response ready. For example, the New York State Energy Research & Development Authority (NYSERDA) publishes a voluntary “stretch code” for new building construction, which specifies that “building controls shall be designed with automated demand-response infrastructure capable of receiving demand-response requests from the utility, electrical system operator, or third-party demand response program provider, and of automatically implementing load adjustments to the HVAC and lighting systems.”
City Operations and Infrastructure Maintenance
Many smart city applications that improve city operations and maintenance also lead to greater energy efficiencies, reduced waste, and fewer greenhouse gas emissions. A McKinsey report finds that deploying a suite of smart city applications in waste and water management could reduce average water consumption by 20 to 30 percent and reduce the amount of unrecycled solid waste per person by 15 to 20 percent, while also reducing greenhouse gas emissions.
Many large waste systems have coupled connected trash cans with AI-enabled intelligent waste collection optimization to reduce emissions from garbage collection trucks. Boston University was able to reduce its fuel use and greenhouse gas emissions by 80 percent with the use of BigBelly Solar Compactors—solar-powered trash receptacles that compact trash and communicate when they are full, enabling the campus to reduce trash collection from an average of 14 times per week to 1.6 times per week. In Japan, Mitsubishi has partnered with Groovenauts to develop an optimized waste collection and transportation route between 26 designated sites using AI and quantum computing. The optimal route has reduced the total distance traveled by waste collection vehicles from 2,300 km to 1,000 km, enabling a 57 percent reduction in greenhouse gas emissions and a 59 percent reduction in the number of vehicles needed. Finland is also employing AI for smart recycling by managing waste using a robotic waste sorter.
Similarly, AI-enabled smart water infrastructure can help ameliorate leaks and reduce the cost of maintaining water and wastewater systems. Syracuse, New York, uses AI to analyze its aging water infrastructure to identify leak-prone pipes for repair before leaks occur, allowing the city to make the most of a limited budget for water maintenance. Water consumption tracking, which pairs smart water metering with digital feedback, can give customers greater insight and control over their water usage and nudge people toward conservation. When paired with AI-enabled real-time prices that reflect water availability, smart water meters could reduce consumption by 15 percent in higher-income cities. Similar to dynamic pricing for electricity, AI can help both design real-time water prices and enable consumers to respond to those prices. Levers that curtail water demand—whether by reducing leaks or enabling customers to bring down their water consumption—can reduce electricity consumption in the water sector, resulting in greater efficiencies and fewer greenhouse gas emissions.
Most advanced countries are deploying AI applications in their cities. For instance, Australia, France, Germany, Japan, the Netherlands, New Zealand, Sweden, Singapore, Spain, South Korea, the United Kingdom, and the United States have all taken some steps to develop and deploy AI-enabled smart city applications. A number of developing countries, notably India and China, are also deploying increasingly sophisticated AI systems in their cities. But given the significant benefits, why have AI-enabled smart city technologies not been deployed more broadly? One reason is there are a number of challenges involved in researching, developing, demonstrating, and deploying these technologies that cities are not equipped to address but national governments are. Many countries are already overcoming these obstacles, and their initiatives are instructive for the United States.
Research and Investment in Smart City Technologies Are Public Goods
Cities will underinvest in research for AI solutions that support smart cities because they would shoulder all the costs for only a small portion of the benefits. Though the entire global smart city ecosystem benefits from public AI research, it is not reasonable to expect any one city to foot the bill just so every other city can reap the benefits. At the same time, while several companies have invested in developing smart city applications, the market can only play a limited role in the adoption of these technologies because many opportunities are strongly tied to areas of public sector activity (such as environment and transportation). While the market may eventually be able to establish effective interdependent systems, it will take longer and happen much more slowly than it would with government support to overcome detrimental chicken-and-egg dynamics and encourage mutual adoption of these technologies until market forces can take over and drive full deployment.
National (or in the case of Europe, supranational) governments are best suited to address these barriers, making long-term investments in research and demonstration that cities and the private sector are simply unwilling to fund. These investments marshal resources and drive incentives so that state and city actions benefit the entire nation.
For example, as part of its Horizon 2020 program, a comprehensive funding program for research and innovation, the EU has established a three-year AI4Cities project that brings together leading European cities looking for AI solutions to accelerate carbon neutrality. The EU is providing €4.6 million ($5.4 million) in funding to be split between Helsinki, Amsterdam, Copenhagen, Paris, Stavanger (Norway), and Tallinn (Estonia) to encourage companies and developers in these cities to come up with AI solutions for mobility and energy challenges that will reduce CO2 emissions. For example, they will be asked to come up with ways to better use AI to support demand response, improve energy efficiency, and aid in the development of renewable energy. As part of the mobility challenge, the partnering cities will ask companies to come up with innovative ideas that include using AI to optimize traffic flow and the transportation of goods. By challenging industry to develop innovative solutions for public sector needs from the demand side, the EU is offering up its partnering cities to be successful first customers, increasing market demand for nascent clean technologies and enabling companies to create competitive advantage on the market.
The Australian government is supporting the development of smart city projects through its Smart Cities and Suburbs Program, an AU$50 million (US$35 million) competitive grant program. Projects are co-funded by local governments, nonprofit research organizations, and businesses. One project is the Energy Data for Smart Decision Making project, which seeks to develop a tool that enables property owners to predict the potential electricity they can generate from solar panels installed on their rooftops. Led by the University of New South Wales and Australia Photovoltaics Institute, the tool is now being used in 33 municipalities. Although this tool does not use AI, machine learning offers an opportunity to make it even more effective. Researchers from the Technical University of IaÅi in Romania have shown that machine learning techniques can better monitor PV panels for degradation that air, water, and impurities cause—which is important because degraded PV cells convert less solar energy into usable electricity. Creating more accurate degradation profiles for PV panels can inform stakeholders including utility companies, integrators, investors, and researchers, who are making decisions about improving PV lifetime.
Difficulty in Sharing Data Inhibits Development
Developing accurate AI models requires access to large pools of data. Models that can analyze the pools of data that many cities generate will be able to extract more actionable insights and greater value than those limited to a single city. However, few cities are equipped to develop interoperable systems and share data across their jurisdictional boundaries. Additionally, while cities benefit from analyzing other cities’ data, an individual city itself has little incentive to share data and may even enact policies that limit data collection and sharing, perhaps due to fears about privacy or cybersecurity risks. Finally, even if cities could and wanted to share data in a national pool, in most nations no mechanisms (e.g., urban data trusts) have been established to make it easy for them to share and use data.
As such, national governments have a coordinating role to play in developing common policies, programs, and standards for AI technologies that encourage interoperability and data sharing to increase the effectiveness of AI-enabled smart city applications.
South Korea enacted the National Transport System Efficiency Act in 2009, mandating local governments, regional administrations, public highway authorities, and agencies responsible for privately financed highways to collect and share traffic data in a standardized format with South Korea’s National Transport Information Center (NTIC). NTIC then collates, processes, and analyzes this data from 66 agencies and disseminates traffic information to citizens free of charge through various channels, thereby enabling the national government to analyze the impact different policies have on their cities and communities, and empowering cities with similar challenges to learn from one another. For example, when researchers from Ulsan National Institute of Science and Technology successfully developed an AI system that better predicts real-time traffic conditions using public information disseminated by a broadcasting network, the system was quickly supplied to other Korean cities, including Gwangju, Busan, Daejeon, and Incheon.
The United Kingdom, on the other hand, is facilitating sharing of non-public data that would not otherwise be made publicly available due to its proprietary or sensitive nature, but that has high value. Policymakers there have recognized that creating high-quality datasets that are properly formatted, complete, labelled, and corrected of harmful biases is time consuming and expensive, which means companies and researchers often have to make do with bad data and thus unreliable or inaccurate AI tools. To overcome this barrier to AI development, the United Kingdom established a model for data trusts, a type of data access and stewarding model inspired by legal trusts. These are institutions that enable companies to share data with each other, with data governance decisions made by “trustees” with fiduciary responsibilities. Without a coordinating body such as a government agency specifically devoted to developing and supporting these models, it is unlikely organizations will develop them on their own. National governments can therefore facilitate AI maturity for smart cities by experimenting with data trusts and other models to make existing high-quality datasets more widely available.
Risk and Uncertainty Limits Adoption
Because smart city technologies, including AI, are still emerging, many municipal governments perceive investments in AI for smart city initiatives as risky, making it harder for them to justify the spending without evidence of the return on investment these technologies can offer. Economists refer to this challenge as excess inertia or, more commonly, “the penguin effect”—in a group of hungry penguins, no individual penguin is willing to be the first to enter the water to search for food due to the risk of encountering a predator. Yet if no penguin is willing to test the waters, then the whole group risks starvation.
National governments can adopt policies and practices to demonstrate the value of AI technologies for cities and help address early-stage challenges. For example, the United Kingdom has created the Connected Places Catapult, a government-backed center of excellence for innovation in mobility and the built environment. To tackle barriers in scaling commercial drone products and services, Connected Places Catapult partnered with the Department for Transport, the Department for Business, Energy and Industrial Strategy, and the Civil Aviation Authority on the Drone Pathfinder program. Together, these agencies developed five “first of a kind” demonstration projects, including projects using drones for infrastructure inspection and Coastguard search and rescue. In each case, Connected Places Catapult matched technology providers to priority industrial use cases, and worked with regulators to enable real world demonstration.
The center also enhanced partnerships across industry and government to commercialize innovation. For example, to help leverage the United Kingdom’s position in air traffic management markets, Connected Places Catapult developed the Open Access Unmanned Traffic Management Framework Program, in partnership with the Department for Transport and leading U.K. companies, to test unmanned traffic management in flight trials. The outputs from these trials will give U.K.-based industry a head start on global export opportunities.
Singapore has established AI Makerspace, a national AI platform that provides start-ups and small and medium-sized companies with access to plug-and-play AI resources, such as tools that predict future demand based on historical data. This initiative is part of AI Singapore, the city-state’s impact-driven, national research and AI innovation program launched in 2017.
One of Singapore’s most ambitious smart city ventures is Virtual Singapore, a digital twin of the island that can serve as a test bed for government agencies, businesses, and researchers. Singapore has invested SD$73 million (US$53 million) in developing the platform, which will enable different sectors to develop sophisticated applications and tools that solve emerging and complex challenges. For example, Singapore’s Geospatial Specialist Office is using the digital twin to explore the impact of implementing solar panels in Yuhua, an older housing estate the city has identified for green-style upgrades (figure 5). Using the platform together with energy data from local authorities, the team of urban planners can analyze the buildings that have a higher potential for solar energy production.
Figure 5: Singapore uses a digital twin—Virtual Singapore—to analyze the potential for solar energy production