Urban Artificial Intelligence: From Real-world Observations to a Paradigm-Shifting Concept
By Hubert Beroche, Founder at Urban AI
Cities are facing unprecedented challenges. The figures are well known: while occupying only 2% of the earth's surface, urban settlements host more than 50% of the global population and are responsible for 70% of greenhouse emissions. While concentrating most capital and human wealth, they are also places of systemic inequalities (Nelson, 2023), exacerbating and materializing social imbalances. In the meantime, cities have fewer and fewer resources to face those tensions. Increasing environmental constraints, combined with shrinking public budgets, are putting pressure on cities’ capacities. More than ever, urban stakeholders have to do more with less.
In this context, Artificial Intelligence has usually been seen as a much-welcomed technology. This technology can be defined as machines’ ability to perform cognitive functions, mainly through learning algorithms since 2012. First embedded in heavy top-down Smart City projects, AI applications in cities have gradually proliferated under the impetus of various stakeholders. Today’s cities are home to numerous AIs, owned and used by multiple stakeholders to serve different, sometimes divergent, interests.
I first apprehended this diversity of AIs through a Smart City world tour during which I explored 12 cities and met over 130 stakeholders, in 2019. During this project, I met worldwide scientists and practitioners who are using AI for very different purposes: real-time earthquake detection in Tokyo, urban biodiversity monitoring in London, infrastructure maintenance in Montreal. Not only did they use AI differently, but they also developed, governed, and implemented it differently depending on local cultures, policies, and capacities. Singapore-based H3 Dynamics startup is using drones to create self-repairing cities while CANN Forecast developed a sensorless data-driven solution for urban maintenance, echoing political and capitalistic differences between the Asian city-state and Quebec’s capital.
Based on this first observation, the massification and diversification of AIs in cities, I co-authored a report with 20 scientists and practitioners where we proposed the concept of “urban Artificial Intelligence”. At that time, in 2020, the concept aimed at naming a diversified range of AI applications under a common name and designated “a set of algorithms that learn from urban data sets and are used for solutions that are, or could be, deployed in a city” (Beroche, 2020).
Towards a systemic approach of urban AIs
After observing the high diversity of AIs in cities, the next step was to analyze their similar feature to understand which components unify those categories of AIs and make them share a conceptual space, more than a common name. To do so, we co-authored with Sarah Popelka and Laura Narvaez Zertuche, with the active participation of the Urban AI Community, an Urban AI Guide (Popelka, 2023). Based on an extensive literature review and eleven in-depth interviews with worldwide urban stakeholders, we first extended and improved the existing urban AI definition, now considering that “the term “urban artificial intelligence” refers to any system that incorporates data derived from the urban environment, which is then processed by algorithms, the result of which has useful applications in the socio-spatial nexus of the city”. The main shift from the first definition to the latest one is the systemic approach, which also represents our renewed ambition to unify the studies of urban AIs. At a conceptual level, such an approach also echoes Kate Crawford's Anatomy of an AI System thought-provoking work and, actually, OECD's latest definition of AI.
Still drawing inspiration from Kate Crawford’s methodology with her AI anatomy, we worked at transcribing an urban AI anatomy based on the literature review and the interviews we conducted for the Guide. This anatomy highlights the interconnected layers encapsulated in a common technological architecture. Highlighting such components has not only a conceptual interest — finding common features between urban AIs- but also an operational one. When talking about urban AI, people very often have in mind the extremities of this anatomy, data visualized dashboards, and physical sensors. Though, urban AI systems are much more complex and contain more layers, each of which needs specific expertise and resources. Understanding this whole systemic and integrated process is the first important step towards mastering and well-governing those technologies while avoiding unexpected side effects.
In addition to this common technological architecture, we also observed three features that characterize those typologies of AIs and justify special attention.
Urban AIs are hybrid
In other words, their mode of existence involves materiality, a presence in the physical world that sets them apart from various other forms of AI. This materiality is not only infrastructural but also, and more importantly, infra-ordinary. It reveals itself and unfolds in the banality of everyday life. Whereas the materiality of Facebook or YouTube recommendation algorithms is imperceptible — resting on a network of invisible infrastructures such as data centers, undersea cables, etc. — that of an autonomous boat, a drone or a Lidar sensor used for mapping a territory is apparent to us. This co-presence in the banality of everyday life has several very concrete implications. Among them is the fact that these AIs are critical as they have the potential to transform the lives of city dwellers. An error in the calculation of an AI embedded in a navigation software — let’s say Waze, for example — can quickly create traffic jams or even accidents. More broadly, the hybrid nature of urban AIs should prompt increased attention to their design. One example is the interface of urban AIs, which often is limited to the surface of our smartphone screens. However, this mode of interaction is not only anti-urban but also dangerous. The smartphone literally blinds and obscures us to signals from the urban environment, potentially leading to road accidents, among other things. Hence a need to invent other materialities — poetic, virtuous, frugal, empowering — for these urban AIs.
Urban AIs are political and politicized.
Firstly, because they unfold in the city, which itself is subject to a web of regulations. In this sense, urban AIs must — or at least should — comply with local regulations of Municipalities and Local Governments. This is expressed by researcher Stefaan Verhulst through the concept of “AI Localism”, which “refers to the actions taken by local decision-makers to address the use of AI within a city or community”. In other words, the governance modalities of an urban AI vary from one city to another depending on multiple local components, including political ones. This is evident, for example, with the emergence of AI registries in Helsinki, Amsterdam, and soon Barcelona as well as with the creation of an Algorithms Management and Policy Officer by the Municipality of New York. More radically, the recent ban on shared e-scooters in Paris, the prohibition of drone flights in dense urban areas in several countries where it is allowed in other regions of the world such as China, or the ban on facial recognition in many cities, reflects the localization of urban AIs. These examples demonstrate the rise of differentiated urban AI uses and developments, driven by local policies in complement to national, sometimes regional, regulations. This trend is so powerful that some scholars use the concept of “urban decoupling” (Ekman, 2020) to describe “the development of diverging types of cities and urban governance models in the future”, letting envision an urban AI fragmentation.
At the same time, urban AIs are also being politicized. As we’ve shown in the Geopolitics of Smart Cities report, co-authored with Ana Chubinidze and Lina Goelzer, such technologies can reshape urban behaviors and transform cities’ spatiality. By doing so, they can materialize and reinforce existing social norms, political powers, and economic systems. At the international level, this can lead to turning urban AIs into instruments of soft power. More radically, urban AIs can also be weaponized to attack or take control of urban vehicles (connected cars, drones), infrastructures (hospitals, airports), and dwellers (through targeted misinformation).
Being aware of the political dimension of urban AIs is crucial for anyone wishing to implement those technologies. It’s the first step towards governing this technology and not being instrumentalized by it. It also allows us to understand that instead of being a constraint or a threat, this political dimension can be leveraged to empower communities and amplify cities’ speech (Sassen, 2013). As all urban technologies should do.
Urban AIs are evolving in complex environments.
The city is a complex system (Batty, 2009). Urban order results from the interaction of a considerable number of independent yet interconnected agents. This order is inherently fragile, mutable, and dynamic. The latter element is particularly important as it justifies the existence of urban AIs. These algorithmic systems are relevant because they can enable us to better understand the evolution of this complex system that is the city. When used judiciously, urban AIs can enlighten individual choices and facilitate the understanding of urban dynamics. Having such knowledge, experts, and people can make better decisions for the city and their own lives. There are plenty of examples: Aretian startup using urban data to improve urban quality of life through better economic planning, in Boston, MIT Media Lab using AI and tangible interfaces as a tool of augmented and collective urban planning, in Hamburg, Berlin CityLab, and Telraam using AI to help people better know and take care of their city, respectively in Berlin with participatory tree management and Belgium with crowdsourced mobility data.
Furthermore, the complexity of the city itself is a factor that alters the fundamental nature of urban AIs. Take the example of an autonomous vehicle. While it travels on a highway, one can easily imagine that its automated navigation is activated. However, as soon as it enters the city, it becomes technically impossible to maintain this mode of operation. Unable to analyze and calculate the urban complexity in real-time (pedestrian crossings, traffic lights, construction works, etc.), the autonomous vehicle requires a switch to manual mode to avoid the risk of causing an accident. Another example is the dazzling of laser recognition systems in Hong Kong during political protests in June 2019. Here, population density coupled with human ingenuity jeopardized the functioning of AIs — except, in this case, it was a deliberate sabotage.
A first conclusion drawn from these failures could be to simplify the city, to reinforce AI systems by creating “Autonomous Car car-centric cities” or surveillance cities. Many places are moving in this direction. However, we already know from past failures where this trajectory leads: towards inhospitable cities.
Urbanizing Artificial Intelligence
Another possibility is to start from this complexity to imagine, design, and implement urban AIs. To develop AI systems that are situated, open, frictional, decentralized, meaningful, and ecological. AIs that embrace urban complexity, empower multiple intelligences — not only an algorithmic one — and make possible diverse futures.
At Urban AI, drawing inspiration from Saskia Sassen’s work, we name this process “urbanizing artificial intelligence”. Urbanizing AI means making those technologies fit into cities' ancestral, multiple, and complex lives.
Each time we talk about “urban AI” we are not only mentioning a tool or a concept but we are also expressing this collective vision. A future where cities’ speech thrive and, “poetically, humans dwell upon the earth” (Hölderlin). A future where AI tools serve flourishing and meaningful lives. People’s lives.
Reference List
Batty, M., (2009), Cities as Complex Systems: Scaling, Interaction, Networks, Dynamics and Urban Morphologies, UCL Working Papers Series (2008)
Beroche, H., (2020), Urban AI, Urban AI.
Ekman, A., Esperanza Picardo, C. (2020), Towards Urban Decoupling, ISS European Union.
Nelson, R., Warnier, M., Verma, T., (2023), Conceptualizing Urban Inequalities as a Complex Socio-Technical Phenomenon, Geographical Analysis (2023).
Popelka, S., Narvaez Zertuche, L., & Beroche, H. (2023). Urban AI Guide. Urban AI.
Sassen, S., (2013), Does the City Have Speech?, Duke University Press
Zhuang, Y., & Fang, Z. (2020). Smartphone zombie context awareness at crossroads: A multi-source information fusion approach. IEEE Access, 8, 101963–101977.