Synthetic Nights: Machine and Human Learning in 24-Hour Cities

Urban governance is largely day-centric. Most policies governing cities have a focus on the daytime and traditional business hours, not necessarily capturing the specific nocturnal challenges people and areas encounter after dark. Throughout the late 20th century, digital technologies changed the way we interact with, plan and cohabit our cities. However, the specific impacts these technologies have on the urban night are still not fully understood. Regarding knowledge production, the night itself has been studied and examined from different angles and disciplines, from history to biology.¹Matthew Beaumont. Nightwalking: A Nocturnal History of London (Verso Books, 2016); María DeGuzmán. Buenas Noches American Culture: Latina/o Aesthetics of the Night (Indiana University Press, 2012); David E. Nye. American Illuminations: Urban Lighting, 1800–1920 (The MIT Press, 2018); Kevin J. Gaston, Marcel E. Visser and Franz Hölker. “The biological impacts of artificial light at night: The research challenge,” Phil. Trans. R. Soc. B, 370, 1667 (2015): https://doi.org/10.1098/rstb.2014.0133. It was only recently, however, that it became a key subject of study²Alejandro Mercado-Celis and Edna H. González, eds., Noche urbana y economía nocturna en América del Norte. (Universidad Nacional Autónoma de México, Centro de Investigaciones sobre América del Norte, 2020); Will Straw, “Media and the Urban Night,” Journal of Urban Research, 11 (2015), https://doi.org/10.4000/articulo.3098; Will Straw, “Night Studies: Thinking Across Disciplines”, Voices of Mexico, no. 111 (2020): 7-10. with the emergence of an interdisciplinary field called Night Studies³Michele Acuto, “We need a science of the night,” Nature, 576, 7787 (2019): 339, https://doi.org/10.1038/d41586-019-03836-2; Christopher C.M. Kyba et al., “Night Matters – Why the Interdisciplinary Field of “Night Studies” is Needed,” J, 3, no. 1 (2020): 1-6, https://doi.org/10.3390/j3010001 in the 1990s.

Concerns about the urban night go beyond the university walls. Local governments and civil society organizations also pay more attention to nighttime governance than they did three to four decades ago. Evidence of the growing concerns is the proliferation of nocturnal governance mechanisms such as night mayors,⁴Andreina Seijas and Mirik Milan Gelders, “Governing the Night-Time City: The Rise of Night Mayors as a New Form of Urban Governance after Dark,” Urban Studies, 58, no. 2, (2020): 316–334, https://doi.org/10.1177/0042098019895224 night councils, pilot projects, and specific regulatory frameworks.⁵Alessia Cibin, “Forms of Night-Time Economy Governance: A Framework Towards Clarification,” in Transforming Urban Nightlife and the Development of Smart Public Spaces, ed. Hisham Abusaada, Abeer Elshater and Dennis G. Rodwell, (IGI Global Scientific Publishing, 2021), 22-39, https://doi.org/10.4018/978-1-7998-7004-3.ch003 As cities embrace digitization and datafication,⁶Carl C. Okafor, “Datafied cities: A socio-technical study of data volume, data quality, and data utility as elements of datafication in the smart city,” (PhD diss., University of Stavanger, 2025). reflections on artificial intelligence (AI) and synthetic nights (here defined as the reconstruction and understanding of the night through computational systems) and human learning (how people understand and represent the night) become central to contemporary urban governance. If the night is barely seen, can it be understood by algorithms?

Datasets about the night are scarce and the impact this lack of data have in decision-making has been well documented,⁷See, for example, the project Data After Dark: https://dataafterdark.org/reports.html; Jess Reia, “Invisible data in night-time governance: addressing policy gaps and building a digital rights framework for cities after dark,” Data & Policy, 7, e17 (2025): https://doi.org/10.1017/dap.2024.90 maintaining the dichotomy of hyper-visible daytime vs. opaque nighttime. The efforts to “compute” the night draw from initiatives to understand and manage nighttime city life through data, algorithms, and digital infrastructure. As cities become more connected, sensors, mobile devices, and surveillance systems continuously generate large amounts of information about how people and the city operate⁸Rob Kitchin, “Conceptualising smart cities,” Urban Research & Practice, 15, no. 1 (2022): 155–159, https://doi.org/10.1080/17535069.2022.2031143 during the day or at night.

At the core of this datafication process is machine learning (ML), a subfield of AI focused on creating algorithms that can identify patterns in data (including synthetic data),⁹Benjamin N. Jacobsen, “Machine learning and the politics of synthetic data,” Big Data & Society, 10, no.1 (2023), https://doi.org/10.1177/20539517221145372 allowing computers to learn from experience and make predictions or decisions without needing explicit instructions for each specific task. The promises of ML – efficiency, prediction, and real-time response – can be alluring, becoming increasingly embedded in our everyday lives. As Mattern describes, “Our current paradigm, the city as computer, appeals because it frames the messiness of urban life as programmable and subject to rational order.”¹⁰Shannon Mattern, A city is not a computer: Other urban intelligences (Princeton University Press, 2021). By seeing cities as computers, governance might favor the perceived rationality of ML and AI to other forms of community-produced data and knowledge.

The night is often associated with unsafety, crime, and public disorder.¹¹Deborah Talbot and Martina Böse, “Racism, criminalization and the development of night-time economies: Two case studies in London and Manchester,” Ethnic and Racial Studies, 30, no.1 (2007): 95-118, https://doi.org/10.1080/01419870601006579 That makes surveillance systems an important part of the urban governance equation. Yet, computing the urban night is more nuanced than it being just a technical exercise. Social, political, economic and cultural aspects exert influence on how these systems will operate, as well as the limitations and challenges that emerge when they are deployed. A city at night can be heavily monitored through reliance on big data but still be fundamentally unknown or misunderstood, as these surveillance processes do not capture and understand the complexity and nuances of the night.

The main argument here is that AI depends on data abundance and visibility, while the urban night is defined by partial visibility, refusal, and ambiguity, which creates a tension between ML and human learning in building “synthetic nights.” Three research questions guide this reflection: (1) What is the role of human learning about the night in an increasingly automated and synthetic city? (2) Can nocturnal communities and practices be computed or automated?, and (3) What are the risks and challenges of having increasingly synthetic nights? This work is based on contributions to Night Studies scholarship and my experience in the Montreal’s MTL 24/24 Night Council.¹²See: https://www.mtl2424.ca/en/.

Night data visibility and human learning

The role of data in society changed drastically over the last 150 years. Looking beyond tired narratives of ‘data as the new oil’¹³Sy Taffel, “Data and Oil: Metaphor, materiality and metabolic rifts,” New Media & Society, 25, no.5 (2023): 980-998, https://doi.org/10.1177/14614448211017887. we can find a more complex process rooted in the commodification of data. The definitions, concepts and perceptions of what constitutes data have also shapeshifted, as well as notions of reliability, evidence and ownership. Initially treated as “stable objects whose significance was determined by a handful of professional interpreters, data are now reusable goods,”¹⁴Sabina Leonelli, “Data – from Objects to Assets,” Nature, 574, 7778, (2019): 317–20, https://doi.org/10.1038/d41586-019-03062-w. whose significance and value vary as they are aggregated and accumulated. The increasing quantification of knowledge in various forms, together with the ascension of disciplines like statistics and census, marked an important moment for the collective understanding and uses of data.

But what is considered meaningful data in the era of automation and prediction mediated by algorithms? Over the last decade, the way we talk about data changed drastically, giving it “quasi-divine status: ubiquitous, omnipotent, ineffable. In this incarnation, data has both tremendous power […] and mysterious inaccessibility. Only a cast of initiates (data scientists) can cause this data to speak, and only with the mediation of arcane spells (algorithms).”¹⁵Melanie Feinberg, Everyday Adventures with Unruly Data (The MIT Press, 2022), 2.

ML is inherently built upon human learning. Emily Bender argues that, instead of asking “what is data?,” one can ask “what makes something data?,” highlighting “the action of the data collector”. Dataset producers create datasets that are then used by model developers. The latter need to trust the dataset producers did a good job that is representative of specific tasks, topics or populations. In this process, “dataset consumers are scientifically justified in offloading the work of dataset construction because they have reason to trust that someone else has put in sufficient care in dataset construction.”¹⁶Emily Bender, “What Makes Something Data?” (November 29, 2025), https://medium.com/@emilymenonbender/what-makes-something-data-f6d9f498f312 One problem that emerges in this relationship of trust is when ML research focusses solely on the training process and overlook the accuracy of the training data, as if datasets are always trustworthy by default.

The human learning component makes itself clearer when data collection, processing, analysis and sharing are seen as the flawed stages that capture portions of reality. In messy urban contexts, knowledge production and meanings that data acquire as a tool for governance and decision-making can be equally messy. One of my arguments is that we need to expand the very notion of what is considered relevant data in nighttime governance, in a combination of synthetic precision and human unpredictability.

As Night Studies takes shape and we look at cities from a 24-hour perspective, scholars and practitioners, can explore how nightlife is experienced, understood and regulated through different methods such as ethnography, oral history, alternative spatial mapping,¹⁷See, for instance: https://www.queeringthemap.com/. focus groups, artistic practices and archival research. For Night Studies scholars, “walking, looking and mapping” can be considered core methods for understanding the social order of the night.¹⁸Manuel Garcia-Ruiz, “Night Ethnography,” International Night Studies Network, August 21, 2020, https://doi.org/10.58079/sc3k. Ethnographies, despite the limitations of the method,¹⁹Andy Hong et al., “Reconciling big data and thick data to advance the new urban science and smart city governance,” Journal of Urban Affairs, 45, no.10, (2022): 1737–1761, https://doi.org/10.1080/07352166.2021.2021085. can shed light on complex social relations, exclusion, and belonging that are not captured by big data. For instance, ethnographic work has helped to trace the shift from nightlife as a cultural redevelopment tool to a more complex governance challenge.²⁰Seijas and Gelders, “Governing the Night-time City.”

When datasets are scarce, oral history, interviews, and focus groups are other significant tools for collecting data that helps communities understand themselves and have the potential to inform nighttime governance and policymaking. Oral history projects preserve lived experiences and community narratives, providing qualitative insights,²¹Linda Shopes, “Making Sense of Oral History”, Oral History in the Digital Age (2012),  https://ohda.matrix.msu.edu/2012/08/making-sense-of-oral-history/; Robert Perks and Alistar Thomso, eds., The Oral History Reader, 3rd ed. (Routledge, 2016). and interviews reveal the complexities of nightlife ecosystems that can support bottom-up approaches to managing the night-time economy that are not centered in Western-centric models.²²Michele Acuto et al., “Night-Time Governance Trajectories: The Importance of Scale and Politics,” in Managing Cities at Night: A Practitioner Guide to the Urban Governance of the Night-Time Economy, ed. Michele Acuto et al., (Policy Press, 2021),  https://doi.org/10.1332/policypress/9781529218275.003.0005. Case studies can be used to examine overlooked nocturnal communities and practices, having been used to understand DIY and grassroots spaces often excluded from official accounts.²³Christopher Robins, “Governing the Urban Night: A Study of Night-Time Governance and DIY Nightlife Governance and DIY Nightlife in Malmö”, (Master’s thesis, Malmö University, 2025).

Lastly, archives can become key components of nighttime governance, providing way more than memory-keeping, as the silences in archives also tell stories.²⁴Michel-Roplh Trouillot, Silencing the Past: Power and the Production of History (Boston, MA: Beacon Press, 1995). Beyond institutional archives, much of the nightlife history is present in personal collections and community-led archival efforts, portraying practices and realities that frequently survive only in fleeting publications, police reports, and local media coverage (see Figure 1).²⁵Straw, “Media and the Urban Night.” This approach can also be framed as thick data, which “[…] may be collected by using and analyzing a variety of formats including photographs, videos, sound recordings, sketches, and written ethnographic field notes.”²⁶Hong et al., “Reconciling big data,” 1741. In contrast to big data, thick data derives its value from close attention to details such as objects, movements, practices, behaviors, and language, enabling a deeper and more context-sensitive understanding of the night.

In the realm of nightlife governance, what is called citizen-generated data can be a powerful mechanism to address recurring gaps in traditional datasets. Citizen-generated data refers to information created and shared by individuals or communities outside traditional institutional frameworks, often to address public issues or advocate for change.²⁷See: https://unstats.un.org/UNSDWebsite/citizen-data/. Community-driven data initiatives can illuminate marginalized experiences and underrepresented urban dynamics, particularly in nightlife districts. Queer mapping can preserve and shed light on how 2SLGBTQIA+ individuals experience the city after dark.²⁸See “Queering the Map” https://www.queeringthemap.com/ and “Mapping the Gay Guides”: https://www.mappingthegayguides.org/methodology/. Mobile reporting tools and grassroots mapping platforms can document issues such as noise complaints, accessibility barriers, and safety incidents, providing actionable insights for policymakers. By incorporating citizen-generated data as human learning and agency into night-time governance frameworks, cities can foster more inclusive and transparent planning.

Human learning about the night, combined with expansive notions of “meaningful/valid” data, allow for less misrepresentation of nocturnal activities. These can often become distorted, as less data, biased data (such as over-policing of certain regions and communities), and the overrepresentation of the problems associated with cities after dark (like crime and risk) take the urban night from barely visible to wrongfully perceived. The tension between the misrepresentation of people cohabiting the night and local governments’ priorities can also lead to refusal of becoming a datapoint. Many nocturnal individuals and groups exist in informal, irregular or illegal contexts, from sex work and undocumented labour to harm reduction initiatives and street vendors; hence the hesitation of communities on being part of datasets.²⁹Reia, “Invisible data.” The issues of data in/visibility affect systems like AI and predictive algorithms because these technologies are highly dependent on the quality of their datasets. If the night is misrepresented, the outcomes of ML and other methods cannot showcase the real needs of cities after dark.

Computing nocturnal life

There are aspects of the night that can and have been for a long time measured and computed. For instance, biological rhythms, such as circadian cycles,³⁰Catherine Rich and Travis Longcore, Ecological Consequences of Artificial Night Lighting, (Island Press, 2006). are continuously more well-modeled through data on light exposure, hormone release, and behavioural patterns. Other examples include light pollution monitoring, with projects like Earth Night using open datasets and AI to model artificial sky brightness (see Figure 2); wildlife tracking in initiatives like the mid-Atlantic nocturnal bird migration monitoring led by the Biodiversity Research Institute;³¹See: https://briwildlife.org/mabs/mabs-nocturnal-migration-monitoring/. and databases like eMammal,³²See: https://emammal.si.edu/. a data management system and archive for camera trap research projects widely used by scholars to monitor nocturnal wildlife in and out of cities. Such data-centered projects allow nocturnal life to become legible through mobility patterns, activity cycles, and spatial interactions that can be quantified and simulated. These computational approaches are especially useful in addressing environmental issues at night, as well as wildlife conservation efforts and urban planning, where understanding non-human nighttime behaviour helps reduce human impact on other animals.³³However, the acknowledgment that non-humans are also part of night-time governance in urban areas is still underexplored in Night Studies.

Still, when it comes to cities and cultural practices, the night exposes the limits of AI adoption for urban governance more clearly than the day. Much of the night knowledge and practices are embodied, relational and situational. One of the main limitations of machine learning when it comes to urban nighttime governance lies in the fact that categorization organizes reality into datasets erasing nuances, and that so many AI systems flatten complex context into patterns. These are necessary – although limiting – processes to facilitate computing nocturnal life as part of a broader agenda in which urban governance, especially in large and megacities, increasingly depends on data and evidence for improved policymaking.³⁴Jess Reia, “The Boundaries of Evidence-based Policymaking for Night-time Governance,” Sociológica México, 41, no.113, (2026): 275-306, https://doi.org/10.24275/lslw4286.

The corporate smart cities agenda that emerged in the early 2010s contributed to wider deployment of tools like sensors, facial recognition cameras, AI-based automated license plate readers (ALPRs), and predictive algorithms. Cities like Rio de Janeiro and Barcelona were pioneers in having data-centric control centers.³⁵Christopher Gaffney and Cerianne Robertson, “Smarter than smart: Rio de Janeiro’s flawed emergence as a smart city,” Journal of Urban Technology, 25, no.3, (2018): 47-64, https://doi.org/10.1080/10630732.2015.1102423. Rio, a coastal city, has been struggling with extreme weather events like floods and landslides that drastically affect low-income areas, highlighting the need for a 24-hour monitoring tool. One response was a community-based early flood warning system, in partnership with its flagship operation center,³⁶See: https://www.c40.org/case-studies/inclusive-water-resilience-accelerator-fund-creating-a-community-based-early-flood-warning-system-to-protect-low-income-frontline-communities/. that rely on data provided by residents, big data and ML to improve the city’s capability to prevent disasters. In this context, climate change data and AI modeling are becoming key parts of the governance agenda of cities worldwide. Increasingly, it is also getting attention from night studies scholars,³⁷Andreina Seijas, Shashi Karunanethy and Dean Magee, “Adaptive urban economies: evidence of intra-day temporal behavioural adaptation to extreme heat in Australian cities,” npj Urban Sustainability, 6, (2026). https://doi.org/10.1038/s42949-025-00297-7. nocturnal communities and practitioners concerned with the city after dark.

Aside from urgent responses to environmental issues, cities are increasingly trying to use AI and ML to govern the urban night. These technologies are applied in areas such as crowd management, transport planning, predictive policing, noise monitoring, and the regulation of nightlife economies. In places like Amsterdam, London, and New York City, night-time governance structures have incorporated data-driven tools to monitor mobility patterns, identify high-density nightlife zones, and allocate municipal resources more efficiently.³⁸Seijas and Gelders, “Governing the Night-time City.” Beyond data collected by the government, other kinds of urban data also become relevant in this case: CCTV feeds, ride-share activity, food delivery apps, mobile phone data, and social media interactions (through Application Programming Interface), allowing authorities to better understand how the city operates during night hours. However, this understanding can come at the cost of the violation of fundamental rights, like privacy.

There are risks when the urban night becomes synthetic, raising concerns about the social and political implications of the AI-led night governance.³⁹Reia, “Invisible data.” Privacy is at the forefront of these implications. Beyond the issues of under- and misrepresentation of night culture in datasets and ML, some of the other problems may include extensive surveillance (through systems like facial recognition technologies and automated license plate readers) of informal workers and harm reduction organizations, which end up exacerbating existing issues.⁴⁰Joy Buolamwini and Timnit Gebru, “Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification,” Proceedings of the 1st Conference on Fairness, Accountability and Transparency, in Proceedings of Machine Learning Research, 81 (2018): 77-91 https://proceedings.mlr.press/v81/buolamwini18a.html. Other groups are particularly vulnerable to extensive nocturnal surveillance, such as sex workers, queer communities and the unhoused.⁴¹Reia, “Invisible data.” Thus, a central risk is that computation can make the night too visible, therefore undermining practices and communities that make it valuable in the first place.

Another cost is environmental. Emerging conversations at the intersection of night studies and environmental justice about the impact of AI infrastructure, such as data centers, on local communities shed light on various ramifications of AI that are relevant for nighttime governance. The high consumption of water and electricity⁴²Tamara Kneese and Maia Woluchem, “Data Centers Aren’t the Future of American Prosperity. Policy brief”, Data & Society Research Institute (2025), https://datasociety.net/wp-content/uploads/2026/06/Myths-of-AI-Data-Centers-Arent-the-Future-of-American-Prosperity.pdf. and noise pollution that affects sleep⁴³See: https://www.eesi.org/articles/view/communities-are-raising-noise-pollution-concernsabout-data-centers. put AI in a broader context that prompts questions of where and whether we need ML for nighttime governance.

Moreover, economic development priorities may not align with the local communities, especially when defining areas of interest based on biased data. Overall, techno-solutionism,⁴⁴Evgeny Morozov, To save everything, click here: The folly of technological solutionism (PublicAffairs, 2013). the belief that all complex social, political, and environmental problems can be solved through technology, is a pervasive issue. These systems might also create problems related to surveillance of nocturnal labour considered illegal or irregular, monitoring of dissidence in authoritarian and democratic contexts, and arbitrary law enforcement that disproportionately impact Black people.⁴⁵Pablo Nunes et al., “Mapeando a vigilância biométrica: levantamento nacional sobre o uso do reconhecimento facial na segurança pública”, (CESeC, 2025), https://cesecseguranca.com.br/textodownload/mapeando-a-vigilancia-biometrica-levantamento-nacional-sobre-o-uso-do-reconhecimento-facial-na-seguranca-publica/.

The focus on technological solutions might also deepen the perception of the night as a risk to be managed, instead of a complex ecosystem that needs resources and care to flourish. To avoid compressing the night into reductionist and simplistic narratives while also protecting human rights (e.g. privacy) and civil rights (e.g. the right to protest), ML-driven nighttime governance requires human-centered approaches.⁴⁶Reia, “Invisible data.” In many cases, there is a sociocultural dimension to be considered: refusal to be visible and computable. Nocturnal life can represent what is hidden or beyond direct observation, challenging the idea that everything can (or should) be visible. Often, certain marginal practices such as DIY creators, artists, harm reduction and sex workers are invisible to data-centric systems by necessity or choice.

Rather than relying solely on technocratic systems, scholars increasingly advocate hybrid models where data science and ML support rather than replace local expertise, oral histories, and community-based forms of knowledge production. As many cities lack data strategies, especially strategies that consider the night as a fundamental part of urban governance, the advancement of big data into the city after dark should be done with caution, in a dialogue with communities and advocates for ethical use of technology.⁴⁷Renee Sieber, et al., “What is civic participation in artificial intelligence?”, Environment and Planning B: Urban Analytics and City Science, 52, no.6 (2024): 1388-1406. https://doi.org/10.1177/23998083241296200.

Reclaiming the Synthetic Night

While computational tools, AI systems, and ML solutions can illuminate patterns of nocturnal existence, they cannot fully encompass the complexity of nocturnal life. The night resists computation not because data is scarce and fragmented; it does so because it often exceeds the logic of datafication – which is the systematic process of transforming aspects of the world into quantified and machine-readable data points.

The nuanced data produced through various qualitative methods and community-based data practices provide insights and interpretations that may not be properly captured by data science and big analytics. Effective nighttime governance requires combining computational insights with human insights, ensuring that the computed version of the night acknowledges and supports lived experiences rather than obscuring or distorting them.

It is important to differentiate human learning and knowledge about the night from the appropriation of humanist values by powerful companies selling technology solutions to local governments. These companies often promise “that these humane, egalitarian, inclusive and sustainable futures have to be rooted through technological change”,⁴⁸Myria Georgiou and Scott Rodgers, “The Mediapolis Q&A: Myria Georgiou’s Being Human in Digital Cities,” Mediapolis: A Journal of Cities and Culture, 9, no. 2 (2024). without taking into consideration the many limitations of AI. The urban night exposes the limits of AI more clearly than urban daytime and questions the very notion of human-centered city by teaching us that humans cohabit the night with other species, and that action taken to govern the night should have a better assessment of impacts on humans and non-humans.

Human learning about the night, instead of being seen as an opposition to machine learning, can be framed as the process of knowledge creation that guides adoption of AI, when necessary, as a supporting tool – not a governing logic. An expansive notion of valid data for policymaking and governance can be combined with lived experiences, community engagement and citizen data to contain techno-solutionist discourses about the future of 24-hour cities.

Protecting the night means protecting ambiguity, difference, opacity and rights. The future of our cities may depend on reclaiming the synthetic night, understanding not all should be neither visible nor computable.

Jess Reia

Jess Reia is an Assistant Professor of Data Science & Public Policy and a 2025-2027 Andrew Carnegie Fellow at the University of Virginia, where they co-lead the Digital Technology for Democracy Lab. They are also a Visiting Scholar at East China Normal University and a Research Associate at the Data & Society Research Institute. Reia works primarily on technology policy, human rights, and urban governance. A policymaker by training, they have collaborated with governments for over a decade and conducted research published in four languages.