Equitable urbanism: How AI advances inclusive city planning and resource allocation

Latin America is one of the Earth’s most urbanized—and urbanizing—places, where 81% of people live in cities. In 1950, only 41% of Latin American people lived in cities, and the massive crush of people flooding into urban areas since then has generated vast informal settlements that exist outside of municipal access to basic services and utilities and even administrative recognition.

According to the United Nations Human Settlements Programme (UN-Habitat), 21% of the inhabitants of Latin America and the Caribbean, some 110 million people, live in these informal settlements, where they face severe problems with pollution, access to transit and services, security, and subpar infrastructure. It’s a long-simmering problem that leaders are looking toward new technologies, such as artificial intelligence (AI), to help solve.

AI-empowered photogrammetry engines can map informal settlements, the first step toward integrating them into administrative censuses and allocating resources appropriately. AI is playing a critical role in more developed North American contexts, as well, using publicly available data to evaluate and improve the quality of urbanism via policy solutions.

While the context and application can be very different, urbanists and data scientists are joining forces to use AI to develop more equitable cities. The hope is that a more granular understanding of how urban processes work, derived from publicly available visual data, can generate better policy solutions and the coalitions needed to win support for them.

But AI is not an unalloyed good in urban development. This technology is still early in its lifecycle and comes with critical issues of data access, lack of expertise, and scalability challenges, as well as potential threats to the general public such as a lack of data privacy, surveillance capabilities, embedded bias, and disruptions to labor markets. Collectively, these issues significantly impede people’s trust in AI. Regardless, AI tools will play a greater and greater role in planning cities worldwide; the question that remains is how.

Developed by urban planner and information technologistAntonio Vazquez Brust, Mapping with AI for Informal Areas (MAIIA) is a photogrammetry platform that maps informal settlements in Latin America. Funded by the Inter-American Development Bank, MAIIA scans patterns, colors, and contrasts to define and map the borders of informal settlements using public satellite and drone imagery.

Intended for use by municipal governments in Latin America and the Caribbean, the platform works best when users can plug in existing (even if outdated) maps or assumed borders of informal settlements and let the model train on this data for several days. Once the algorithm is familiar with this local context, users can create an iterative series of maps every few months to track growth.

Vazquez Brust says this process is “exponentially” cheaper than traditional surveying methods and “can tell you where and approximately how many people live without access to, for example, water.”

At this point, MAIIA’s understanding of informal settlements is hyperlocal and extremely context-sensitive, Vazquez Brust says: “We kind of start from scratch in every city. If you train the algorithm for the outskirts of Bogotá City in Colombia and test it with another Colombian city from the coastal areas, the algorithm gets confused. Our human communities are so different.”

MAIIA also has a rudimentary understanding of elements inside informal settlements and can, for example, pick out circulation routes, topographic context, or utility access. The current algorithm requires many hours of training and manual labeling to be able to understand individual elements of a specific neighborhood or even just a few blocks.

“Being able to discern what’s going on inside the informal areas is the next frontier,” Vazquez Brust says. And while the technical capacity to connect MAIIA directly to government services with automated reports exists today, he says, Latin American “local and state governments rarely have any kind of human resources infrastructure that could take on this responsibility.”

So far, platforms like MAIIA are “a resource for people to make decisions, but they are not automatic triggers for teams on the ground,” says Soledad Guilera, a lecturer at the University of California, Berkeley’s Goldman School of Public Policy who studies the use of AI in cites. “These tools are accelerating the way we process information and helping to make it more efficient in terms of targeting what needs to happen, but it’s still a human process, thankfully.”

Next up for MAIIA, Vásquez Brust wants to build functionality to count informal settlement rooftops—a proxy for individual dwelling units—and ascertain buildings’ structural integrity. Most important, however, will be overcoming data scalability issues with a more universal model that takes less manual and automated training to apply the tool to each new locality.

While MAIIA focuses on the view from the sky in Latin American cities, State of Place evaluates the quality of urbanism at the street level. This tool, developed by Dr. Mariela Alfonzo, helps to equitably optimize the social, health, environmental, and economic value of communities and is grounded in data and social science.

State of Place uses AI to rate the quality of the built environment based on street-level imagery. It uses computer vision to extract data on 127 individual urban design features and aggregates this data into a score from 0 to 100, known as the State of Place Index. The Index is divided into subindices that measure 10 urban design dimensions, which encompass both broad patterns of urbanism (density, connectivity, land uses, streetscape enclosure and continuity, and so forth) and granular features such as crosswalk markings, benches, bike lanes, and outdoor dining. The software visualizes this data spatially and graphically to help users assess their built environment assets and needs.

State of Place has also developed a series of forecasting models that tie the Index, and the 10 subindices, to various outcomes, including real estate values; chronic disease occurrence; heat index; and how much people walk, drive, or take transit. Via the software, users can then generate specific urban design recommendations most likely to help them achieve their policy goals or desired outcomes, such as increasing office premiums, reducing diabetes rates, mitigating crime, or boosting transit ridership.

Users can then simulate how recommended changes to the built environment might impact the overall State of Place Index and in turn aid or detract from their policy priorities or community value. For example, how does adding benches along a streetscape or planting trees alter the number of vehicle collisions or increase property taxes? Does it draw more pedestrians to the area, boosting retail revenue? These value forecasts—or predictive analytics—have helped public, nonprofit, and private sectors alike use compelling evidence to help “justify their investments or get community buy-in so that they secure funding or approvals for their proposed projects,” Alfonzo says. They also help users prioritize redevelopment proposals that optimize the values communities care about most.

Their AI models have already uncovered some curious connections between urban design features that buoy seemingly unrelated quality-of-life elements. For example, working with the City of Durham, NC’s Transportation Department, which was aiming to achieve Vision Zero goals, State of Place models showedthat for every 1-point increase in the Index, there was a 12.3% reduction in the likelihood of a vehicle collision, on average. Consequential as that number is, this was a relatively expected outcome. However, when digging deeper, Alfonzo also found that places that scored better on the Index’s parks and public spaces subindex had a far lower risk of vehicle collisions—with a 1-point increase in the subindex translating to a 26.5% collision risk reduction on average.

“It’s not something that traffic engineers tend to think about; it’s an indirect relationship,” she says. “What’s happening is that when you have a park or public space, you have more people, more walking, and people tend to slow down in those areas.” Understanding the connections between the built environment and quality of life doesn’t just help prioritize effective design strategies—it can literally be a matter of life and death.

The State of Place Index, as an objective measure of the built environment, does not include demographic factors, but that doesn’t mean there isn’t a connection. Demographics—including race, income, ethnicity, and other metrics that impact equity—are accounted for via State of Place’s predictive analytics models. For example, when working with the City of Philadelphia, State of Place foundthat places with lower State of Place Index scores had higher rates of chronic disease, crime, COVID-19 incidence, heat, and flood rates. The people who live in lower-rated areas were more often Black and had lower incomes and lower education rates than those living in places with a higher State of Place Index. By quantifying inequalities tied to the built environment—or spatial equity—users can more effectively advocate for more just, thriving places.

To that end, State of Place actually quantifies the gentrification and displacement risk tied to proposed built environment investments. For example, suppose a redevelopment plan will add a series of pocket parks for street festival vendors, new arts and culture incubators, and retail outlets, as well as more market-rate housing. The software not only quantifies how much the State of Place Index score will rise given these changes but also predicts how much rents, residential values, and property taxes are likely to go up.

Advocates and policymakers can now calculate the gap between what existing residents or neighborhood shop owners can actually pay compared to projected future rents or property taxes and then inform policies and strategies—like community benefit agreements, land banking, subsidies, and the like—which they can use to close affordability gaps. These new, proactive policies, crafted to reflect the model’s predictions, are required “to ensure that the increase in value actually accrues to the residents who needed that redevelopment effort to begin with,” Alfonzo says.

Trees are one index of urban quality that State of Place considers, but their maintenance and viability are the entire focus of Quantified Trees (QTrees), developed by Julia Zimmermann at Technologiestiftung Berlin, together with Birds on Mars and the Greenery department of Berlin-Mitte. Drawing on an extensive database of trees in Berlin that covers 860,000 of the city’s approximately 1 million trees, QTrees uses AI and data on precipitation, temperature, type, age, and existing watering schedules together with data collected by moisture sensors to determine the ideal watering needs for each tree. It integrates a 3D model of Berlin that tracks the shade cast by tall buildings on trees and the path of the sun across the sky to calculate shading levels for each tree, though the model is not automatically updated when new buildings are built or demolished.

“All of those little characteristics of one tree come together to calculate the soil tension for the next 14 days, the power trees use to suck at the soil to get the water out of it,” Zimmermann says. Soil tension is a key metric for how much water a tree needs, and though soil moisture is a more accurate measure in many cases, the process of monitoring soil moisture is impractical and unpredictable. It requires many soil moisture sensors to be effective, and moisture dispersal in urban soils is often erratic, as infrastructure elements—such as utility pipes, building foundations, and transit lines—are often coursing below ground, impeding the flow of moisture to trees.

Currently, Berlin has a network of more than 200 soil tension monitors collecting data. “The higher the tension, the dryer the ground is, and the more you need to water it,” Zimmerman says. Using a dashboard, municipal authorities can sort trees by greatest watering need, location, age, and more. QTrees runs on public data and is based on open-source code, which is why it can be easily adopted in other cities as long as they provide a tree database and moisture sensors as a reference to train the AI model.

In Berlin and every other city, trees are a powerful index of inequality. They are vital for biodiversity, aiding human health and increasing carbon sequestration and water retention. Where they’re absent, cities are hotter, less tranquil, less hospitable places. “If a city is green, the livability is better,” Zimmermann says.

Addressing the data-scalability issues inherent to QTrees would require deploying significant resources to assemble similar tree databases in other cities, and making this case runs headlong into the ROI demands set by the private sector that often develops AI and cities themselves. The long-term benefits of arboreal cities are clear; the short-term return is not. Here and elsewhere, pure market efficiency is not the best path to equitable urbanism. So how does one train AI to demonstrate human values beyond profitability? “This is why it’s so important to combine social science with data science,” Alfonzo says.

The ethics of AI is a critical regulatory issue that’s subject to debate in a still-nascent field. But arriving at the right policy necessitates inviting in the right group of stakeholders, namely one that’s as diverse as possible, says Guilera, that can understand the social and economic context of everyone affected by AI. “The more we rush into being efficient and quick with this, the more risks we take in terms of how we’re making this decision and who is being included in this decision,” she says. “I think one of the great roles that cities can play is leading the public conversation around AI, in terms of education but also civic engagement and setting a space for the conversation at the local level.”