AI and the transformation of accountability and discretion in urban governance

“On the other hand, urban governance faces complex, “wicked ” problems ( Rittel & Webber, 1973 ), necessitating new governance models that utilize flexi-ble and cross-agency institutional arrangements ( Stead & Meijers, 2009 ).”

“AI-driven recommendations, either requested by human agents or integrated into on-site applications, can generate prompt, scenario-specific suggestions, reducing variability and personal biases ( Selten et al., 2023 ).”

“In the context of AI, deploying AI in bureaucratic systems may further create opportunities to augment the discretion of bureaucrats while upholding stringent accountability standards due to AI’s enabling effects on the data-processing capacities of government employees.”

“Their choices are shaped by their understanding of policy goals and political agenda ( May & Winter, 2009 ), professional knowledge ( Evans, 2011 ), as well as their own understanding of the relationships among citizen clients, coworkers, and the system ( Maynard-Moody & Musheno, 2000 ).”

“Algorithmic applications in high-stakes domains must be assessed not only for fairness and transparency but also for their broader societal implications ( Mittelstadt et al., 2016 ).”

“Although GenAI is often discussed in terms of its technical capabilities, such as generating text, images, or predictive insights ( Ouyang et al., 2022 ; Cao et al., 2023 ), the transformative feature is its interactive capacity.”

“…mechanisms such as administrative procedures, which formalize processes and standards to be followed by bureaucrats ( McCubbins et al., 1987 ) as well as democratic oversight, which allows citizens to be engaged in monitoring and evaluating government actions ( McCubbins & Schwartz, 1984 ).”

“As AI continues to reshape bureaucratic visibility, new forms of work relationships and positions will emerge, while carrying both the legacy of hierarchical control and the possibilities of data-enabled collaboration ( Giest & Klievink, 2024 ; Kellogg et al., 2020 ).”

“It emphasizes an agent’s obligations to explain and justify their actions to another entity, which could be a person, an organization, or the public in general, holding the power to question and evaluate the agent’s actions ( Bovens, 1998 , 2007 , 2010 ; Mulgan, 2000 ; Mulgan, 2003 ).”

“It emphasizes an agent’s obligations to explain and justify their actions to another entity, which could be a person, an organization, or the public in general, holding the power to question and evaluate the agent’s actions ( Bovens, 1998 , 2007 , 2010 ; Mulgan, 2000 ; Mulgan, 2003 ).”

“The case of two Swedish municipalities, previously mentioned, offers a clear example of how AI-enabled systems can reinforce professional accountability ( Ranerup & Svensson, 2023 ).”

“By capitalizing on this trust, local governments can translate technical systems into accessible terms and address concerns about data privacy, bias, or automation ( Lee, 2018 ; Araujo et al., 2020 ), while also incorporating citizen feedback into the deployment and evaluation of AI tools.”

“In this context, AI can assist street-level bureaucrats by standardizing critical processes, improving procedural fairness, and aligning goals while still allowing for professional discretion where necessary.”

“It emphasizes an agent’s obligations to explain and justify their actions to another entity, which could be a person, an organization, or the public in general, holding the power to question and evaluate the agent’s actions ( Bovens, 1998 , 2007 , 2010 ; Mulgan, 2000 ; Mulgan, 2003 ).”

“On one side of the debate, scholars like Snellen (1998) , Zuurmond (1998) , Barth and Arnold (1999) , and Bovens and Zouridis (2002) argued that the expanded use of early ICT applications may curtail discretion by automating tasks and reducing the need for human judgment.”

“When designed collaboratively, such tools can also support continuous learning and professional development ( Cukurova et al., 2024 ).”

“That can reduce the efficiency of government operations and the quality of services ( Hupe & Hill, 2007 ).”

“The repositioning proposition connects with the theory of Pääkkönen et al. (2020) , which compares today’s algorithmic systems with Michel Crozier’s (1964) conceptualization of bureaucratic organizations as complex socio-technical setups where inflexibility is a defin-ing feature.”

“In contrast, the “enablement ” perspective posits that ICT tools enhance the ability of street-level bureaucrats by improving data-processing capacities and facilitating more informed decisions.”

“…by enabling city-level data integration, agency-tailored interfaces and systems for management, and interactive digital twinning interventions ( Appio et al., 2019 ; Hawken et al., 2020 ; Kitchin et al., 2015 ; O’Reilly, 2011 ; Rabari & Storper, 2015 ; Schrotter & Hürzeler, 2020 ; Yang & Rong,…”

“…addresses these theoretical gaps by re-examining the discretion–accountability relationship as a dynamic process rather than a zero-sum exchange, under conditions of algorithmic governance, setting the stage for subsequent discussions on achieving a more adaptive and transparent governance model.”

“In addition, the complexity of tasks in public administration also makes it difficult for digital systems to capture the full scope of frontline social works ( Buffat, 2015 ; Nagtegaal, 2021 ).”

“Second, they should organize initiatives, such as workshops, mini-lectures, and lunch talks, that enhance AI literacy, data interpretation, and AI risk awareness ( Allen et al., 2024 ).”

“Section 5 concludes the paper and outlines directions for future research.”

“…enabling city-level data integration, agency-tailored interfaces and systems for management, and interactive digital twinning interventions ( Appio et al., 2019 ; Hawken et al., 2020 ; Kitchin et al., 2015 ; O’Reilly, 2011 ; Rabari & Storper, 2015 ; Schrotter & Hürzeler, 2020 ; Yang & Rong, 2024 ).”

“Moreover, considering variations across frontline staff and senior management, this paper aims to reveal how different actors adapt and reallocate their discretionary powers in response to AI-enabled transformations.”

“…maintain algorithm registers and conduct impact assessment on AI deployment • Create internal ethics boards and escalation channels for override or appeal • Conduct periodic reviews for model drift and ethical impact • Provide training in ethical reasoning, bias awareness, and deliberative judgment”

“In this context, the concentration of innovation power among a few leading technology corporations further complicates governance and intensifies risks to regulation and oversight ( Hacker et al., 2023 ; Widder & Nafus, 2023 ; Narechania & Sitara-man, 2023 ).”

“Although GenAI is often discussed in terms of its technical capabilities, such as generating text, images, or predictive insights ( Ouyang et al., 2022 ; Cao et al., 2023 ), the transformative feature is its interactive capacity.”

“…enabling city-level data integration, agency-tailored interfaces and systems for management, and interactive digital twinning interventions ( Appio et al., 2019 ; Hawken et al., 2020 ; Kitchin et al., 2015 ; O’Reilly, 2011 ; Rabari & Storper, 2015 ; Schrotter & Hürzeler, 2020 ; Yang & Rong, 2024 ).”

“Second, managerial pressures for efficiency can conflict with practitioners’ commitment to client-centered service ( Evans, 2011 ; Wang et al., 2022 ).”

“Professional accountability refers to adherence to professional standards and norms, ensuring that bureaucrats perform their duties com-petently and ethically, and in alignment with public values ( Hupe & Hill, 2007 ; Moore, 2013 ).”

“…enabling city-level data integration, agency-tailored interfaces and systems for management, and interactive digital twinning interventions ( Appio et al., 2019 ; Hawken et al., 2020 ; Kitchin et al., 2015 ; O’Reilly, 2011 ; Rabari & Storper, 2015 ; Schrotter & Hürzeler, 2020 ; Yang & Rong, 2024 ).”

“…surveys and interviews can capture how public officials perceive their discretionary autonomy, fairness, and accountability, providing insight into the subjective and interpretive dimensions of discretion (for example, Aoki, 2020 ; de Boer & Raaphorst, 2023 Nagtegaal, 2021 ; Wang et al., 2022 ).”

“It reduces discretion in routine activities by automating processes, yet it can enhance decision-making in complex cases by offering high-quality data and predictive insights ( Young et al., 2019 ; Bullock, 2019 ; Bullock et al., 2020 ).”

“On the other hand, urban governance faces complex, “wicked ” problems ( Rittel & Webber, 1973 ), necessitating new governance models that utilize flexi-ble and cross-agency institutional arrangements ( Stead & Meijers, 2009 ).”

“Brayne and Christin (2021)) show how legal professionals resisted algorithmic oversight through subtle practices like data obfuscation and foot-dragging, seeking to preserve autonomy in the face of managerial control.”

“Each principle captures a key dimension of governing with AI: promoting equity in capability building, enabling adaptive administrative structures, ensuring robust data governance, embedding proactive human oversight, and engaging citizens in continuous review.”

“Overall, literature has provided evidence for both enablement and curtailment theories when addressing ICT’s effects on discretion, reaching a consensus that ICT’s impacts on discretion are highly dependent on the nature and contexts of the tasks ( Buffat, 2015 ; Busch & Henriksen, 2018 ).”

“It is equally important that these structural changes are implemented in close dialogue with union representatives, ensuring that the necessary flexibility does not undermine job security or the foundational fairness of public employment ( Engstrom et al., 2020 ).”

“For the latter, because interpreting algorithmic outcomes remains technically difficult, scholars have”

“…enabling city-level data integration, agency-tailored interfaces and systems for management, and interactive digital twinning interventions ( Appio et al., 2019 ; Hawken et al., 2020 ; Kitchin et al., 2015 ; O’Reilly, 2011 ; Rabari & Storper, 2015 ; Schrotter & Hürzeler, 2020 ; Yang & Rong, 2024 ).”

“These bureaucrats are acknowledged as professionals who use their judgment to interpret and apply policies in ways that directly impact public service delivery ( Hupe, 2013 ; Thomann et al., 2018 ).”

“OECD (2024) reports that 67 % of OECD countries already use AI to improve service delivery, predominantly through tools such as chatbots, automated document-processing systems, and predictive analytics for resource management, while only 30 % apply AI for policy-making.”

“On the one hand, aligning with Weber’s view that technology as an enabler for bureaucratization ( 1946 ), emerging data-and AI-driven innovations might reinforce traditional bureaucratic structures and practices ( Meijer et al., 2021 ; Newman et al., 2022 ).”

“…surveys and interviews can capture how public officials perceive their discretionary autonomy, fairness, and accountability, providing insight into the subjective and interpretive dimensions of discretion (for example, Aoki, 2020 ; de Boer & Raaphorst, 2023 Nagtegaal, 2021 ; Wang et al., 2022 ).”

“High levels of discretion challenge the liberal framework of separation of powers of law-making and enforcement ( Pires, 2011 ), leading to a perception of inadequate accountability.”

“…maintain algorithm registers and conduct impact assessment on AI deployment • Create internal ethics boards and escalation channels for override or appeal • Conduct periodic reviews for model drift and ethical impact • Provide training in ethical reasoning, bias awareness, and deliberative judgment”

“…surveys and interviews can capture how public officials perceive their discretionary autonomy, fairness, and accountability, providing insight into the subjective and interpretive dimensions of discretion (for example, Aoki, 2020 ; de Boer & Raaphorst, 2023 Nagtegaal, 2021 ; Wang et al., 2022 ).”

“This supply chain spans private vendors, public-sector IT units, and open-source contributors ( Cen et al., 2023 ; Cobbe et al., 2023 ).”

“For the latter, because interpreting algorithmic outcomes remains technically difficult, scholars have”

“When bureaucrats embrace digital tools in daily operations, they still deem discretionary power an indispensable component of governance ( Busch et al., 2018 ; Busch & Eikebrokk, 2019 ).”

“This paper has examined how the rise of AI, especially interactive and generative systems, reshapes the relationship between bureaucratic discretion and accountability in urban governance.”

“…maintain algorithm registers and conduct impact assessment on AI deployment • Create internal ethics boards and escalation channels for override or appeal • Conduct periodic reviews for model drift and ethical impact • Provide training in ethical reasoning, bias awareness, and deliberative judgment”

“Section 5 concludes the paper and outlines directions for future research.”

“…collaboration and impedes the ability to critically assess and integrate AI-generated insights with professional judgment, as observed in Section 3.2 , where AI implementation demanded new inter-pretive skills from caseworkers and supervisors (see Flügge et al., 2021 ; Ranerup & Svensson, 2023 ).”

“Government chat-bots illustrate how conversational interfaces help citizens navigate public services and comprehend administrative processes, while also enabling agencies to detect informational gaps and improve communication ( Chen et al., 2023 ; Larsen & Følstad, 2024 ; Pislaru et al., 2024 ).”

“Section 5 concludes the paper and outlines directions for future research.”

“At the city-level, when governments treat AI adoption as a technical procurement rather than a policymaking endeavor, they risk ceding value-laden decisions to algorithm designers beyond public scrutiny ( Mulligan & Bamberger, 2019 ).”

“First, when automation is perceived as opaque or rigid, professional public workers may resist any technological tool perceived as undermining professional expertise or limiting case-sensitive flexibility ( Kellogg et al., 2020 ; Pruss, 2023 ; Stevenson & Doleac, 2024 ).”

“…maintain algorithm registers and conduct impact assessment on AI deployment • Create internal ethics boards and escalation channels for override or appeal • Conduct periodic reviews for model drift and ethical impact • Provide training in ethical reasoning, bias awareness, and deliberative judgment”

“…standardized documentation tools, such as model cards and datasheets, and public algorithm registers (as created in Helsinki and Amsterdam, see Floridi, 2020 ), all of which exemplify institutional responses that aim to improve AI interpretability, data transparency, and expose potential…”

“Spain’s SALER early-warning system also produces auditable, explainable outputs aligned with inspectors’ work questions, which senior officials can use to flag anomalies and justify oversight actions, thereby enhancing upward visibility ( Criado et al., 2020 ).”

“The case of two Swedish municipalities, previously mentioned, offers a clear example of how AI-enabled systems can reinforce professional accountability ( Ranerup & Svensson, 2023 ).”

“Digital technology has facilitated this shift towards networked organizational structures ( Greenwood & Lawrence, 2005 ), creating odds with the Weberian bureaucratic model which traditionally emphasizes rigid rules and standardized processes.”

“For example, Singapore’s Smart Nation Sensor Platform harnesses real-time data from thousands of interconnected sensors to optimize public transportation, monitor environmental quality, and enhance public safety ( MDDI, 2024 ).”

“…organizations to reach under-represented groups • Provide accessible, multilingual interfaces and report back on how citizen input is used A shift toward tech-based capacity building and inclusion is crucial for reinforcing managerial control and public oversight ( Engstrom & Haim, 2023 ).”

“As discretion becomes embedded in digital routines, frontline staff may focus on procedural compliance rather than contextual fairness or empathy ( Ranerup & Svensson, 2023 ).”

“The lack of interpretability can undermine trust, prompting resistance or selective disregard of AI-generated analysis or information ( Pruss, 2023 ).”

“…instance, caseworkers noted that predictive scores encouraged attention to “easier ” clients, thereby un-dermining equity goals by incentivizing work efficiency over needs ( Flügge et al., 2021 ); this finding also echoes with evidence from Swiss public employment agencies ( Körtner & Bonoli, 2021 ).”

“…or rigorous methodology ( Goodman & Trehu, 2023 ); public algorithm registers are typically voluntary, often omit high-stakes use cases (like policing or fraud detection), and rarely incorporate concrete channels for citizen input or contestation ( Cath & Jansen, 2021 ; Goelzer, 2023 ).”

“…maintain algorithm registers and conduct impact assessment on AI deployment • Create internal ethics boards and escalation channels for override or appeal • Conduct periodic reviews for model drift and ethical impact • Provide training in ethical reasoning, bias awareness, and deliberative judgment”

“Similarly, Law and Varanasi (2025) show how workers in diverse sectors enact a new managerial responsibility to verify or override GenAI outputs.”

“When bureaucrats embrace digital tools in daily operations, they still deem discretionary power an indispensable component of governance ( Busch et al., 2018 ; Busch & Eikebrokk, 2019 ).”

“…enabling city-level data integration, agency-tailored interfaces and systems for management, and interactive digital twinning interventions ( Appio et al., 2019 ; Hawken et al., 2020 ; Kitchin et al., 2015 ; O’Reilly, 2011 ; Rabari & Storper, 2015 ; Schrotter & Hürzeler, 2020 ; Yang & Rong, 2024 ).”

“It emphasizes an agent’s obligations to explain and justify their actions to another entity, which could be a person, an organization, or the public in general, holding the power to question and evaluate the agent’s actions ( Bovens, 1998 , 2007 , 2010 ; Mulgan, 2000 ; Mulgan, 2003 ).”

“Frag-mented data infrastructures and poor data quality can result in signifi-cant adverse outcomes, including discrimination, political polarization, and the erosion of public trust in AI-driven processes ( Angwin et al., 2016 ; Simons, 2023 ).”

“While such adaptability has the potential to enhance urban resilience, it concurrently intro-duces a fundamental tension with the demands of policy stability and democratic planning (Goldsmith & Yang, 2025).”

“…raising critical governance concerns regarding responsibility for errors, unintended consequences, and innovation outcomes generated without direct human control, consistent with critiques of algorithmic opacity and the “black box” nature of AI systems (Pasquale, 2015; Goldsmith & Yang, 2026).”

“Goldsmith and Yang [1] similarly argue that trust in automated governance cannot be restored through technical transparency alone, because the deeper challenge is relational distance between institutions and citizens.”

“…is further intensified by the “digita-lization of accountability,” which transforms traditional forums into persistent, data-driven surveillance mechanisms, further shifting the focus from internal professional virtues toward external procedural compliance (Busuioc, 2021; Goldsmith et al., 2026 ).”

“Discretion now flows across agencies and contractors through automated pipelines, and existing accountability frameworks were not designed to follow it [43, 4].”

“To better understand the above, it is worth considering the work of Goldsmith & Yang [7], who approach this question from the angle of bureaucratic discretion and argue that AI does not simply restrict or expand the discretion of public officials, but redistributes it between institutional levels,…”

“…of specialized expertise because system configuration, data governance, procurement workflows, authorization hierarchies, and financial controls require knowledge that political principals, administrators, and oversight institutions may possess to different degrees (Goldsmith & Yang, 2026).”

“The absence of explainable logic underlying the algorithm can lead to a counter-revolution from the citizens of a targeted place when applied in a context that requires some cultural sensitivity, such as the tax sector or fraud detection [38, 108].”

“Researchers believe that the government regulation must not merely prescribe but also encourage learning, capacity building, and managerial discretion by the within the clear accountability framework (Kessy, 2023; Goldsmith & Yang, 2025).”