Artificial Intelligence for Urban Planning—A New Planning Process to Envisage the City of the Future

Abstract

Assuming that Artificial Intelligence (AI) is changing the approach to urban planning issues, this study investigates whether changes will start to occur at a theoretical level or if technological innovations will mostly be endured rather than used with full knowledge. The authors observed that technological innovation often occurs without a unifying theoretical framework to provide knowledge and a basis for its adoption. The first use of technology in urban management dates to the late 1950s, and it has recently regained attention within the scientific literature; however, a significant deficiency still exists regarding the definition of a theoretical framework for its use. Focusing on the use of AI in urban and regional planning, this study aims to address this gap by outlining theoretical observations that can guide the integration of AI into new approaches for the management of urban transformations. The enormous impact that the rapid and pervasive spread of AI is having on all human activities necessitates the definition of new educational and disciplinary processes, especially in fields like urban planning, which rely on the high potential of such technology for envisioning future scenarios. It is therefore essential to assume that AI will also modify the management of urban and territorial transformations. This study aims to suggest a framework for scholarly debate on the need to define new historical–disciplinary dimensions by appropriately using AI in the phases of urban planning, avoiding the risk of passively accepting AI’s potential by delegating the development of urban planning tools to artificial reasoning. Building on these premises, this study first provides a thorough and critical literature review regarding the use of AI in urban planning and then proposes a methodological framework. The final section discusses the possibilities and limitations of this approach, thereby contributing to the scientific debate on defining a theoretical framework for the adoption of AI within urban and regional planning processes.

1. Introduction

This study arises from the need to propose an innovative theoretical–disciplinary framework in the field of spatial urban planning, based on the possibilities offered by artificial reasoning. The existing literature includes many examples of the application of AI to urban and territorial planning, which often assume—sometimes uncritically and perhaps dangerously—the high potential of AI. What theoretical and educational frameworks are required to guide research and practices in urban planning, thereby fostering the conscious and effective use of AI in the governance of complex urban systems? How can future urban operators prepare to leverage the potential of AI in a thoughtful and informed manner? Is it possible to respond to new crises using AI, given the current absence of established theoretical frameworks?

Seeking to answer these questions, this study focuses on a theoretical reflection rather than presenting case studies or applications—which, at this stage, would risk obscuring the main goals of the research.

In this sense, a systemic approach appears to be the most suitable for evaluating the potential of AI, considering the various urban subsystems and the system governance cycle borrowed from the cybernetics discipline. Through the adoption of a new perspective in the urban planning process, it is possible to understand how AI can be used in interpreting and formalizing the interactions among the parts of the urban system that determine its evolution in space and time [1]. This possibility was largely infeasible with the traditional phases and processes of urban planning. Recent advancements in hybrid AI architectures have informed the need for a multi-component approach; in particular, a hybrid architecture that is capable of analyzing diverse data streams and interpreting dynamic social behaviors would be highly suitable for addressing the inherent complexity of urban systems.

The relationship between humans and technology is undergoing an acceleration that has never been observed before. The discovery of the possibility to extend the physical capabilities of human beings through the creation of technical tools represents an epochal turning point due to the introduction, diffusion, and utilization of Artificial Intelligence (AI). This change, which can be defined as an “entropic watershed” [2], is happening in cities (where a growing number of people are concentrated) and will inevitably affect the lifestyles, behaviors, and interactions of many individuals. Within the scientific literature on Artificial Intelligence, two opposite—or, rather, complementary—positions can be delineated, mostly referring to the ethical aspects of AI. On the one hand, Kate Crawford [3] underlined the negative aspects of AI; thus, her view can be defined as the “pars destruens” of the question. On the other hand, Luciano Floridi [4] goes beyond the ethical aspects of AI and highlighted the positive effects of its use; thus, his vision can be defined as the “pars costruens” of the same question [5].

These two visions are very interesting, especially because they provide evidence that AI cannot be considered as a simple “new technology” but, instead, as a complex material network composed of natural resources, fuels, human labor, infrastructure, logistics, histories, and classifications; in this sense, AI systems are also the fruit of large social and political structures. Thus, their impacts do not depend on the technology itself but on how a given society, shaped by its own values, appropriates it [2].

Another relevant aspect concerns the fact that it must be possible to explain how AI works to affirm that it is ethically correct (otherwise known as the principle of explicability) [4].

The ethical implications of technology—and, thus, AI—remain contentious among scholars and experts; as such, despite a shared recognition of the need for an ethical code, a consensus is unlikely. In this context, this study focuses on the roles of AI in urban planning, particularly referring to the management of territorial transformations and the requirement to arrange the necessary tools, which is the main task of planning.

The smart city model, in all its various forms, has been centered on technologies to date. Even though different definitions still exist, depending on the various perspectives from which a city can be analyzed, there is a widespread consensus on the centrality of information and communication technology in a smart city, supported by continuous interactions among the people living in it. A more recent evolution of the concept refers to the use of the term “sustainability” which, when associated with a smart city, can result in substantial differences. Many scholars have emphasized that a smart city is not inherently sustainable, highlighting differences and sometimes contradictions between smartness and sustainability [6,7,8,9,10].

This discrepancy arises because these concepts lack a solid theoretical foundation, with their usage often depending on the specific context in which they were developed, as is the case with technologies and AI in general. Additionally, the need for theoretical references typically emerges only after these concepts have been applied or widely used in everyday language.

The significance of this study lies in the abovementioned observation. The interplay between cities and technology is examined, advocating for a holistic–systemic approach to urban and territorial transformations. This underlines the importance of developing a new urban planning theoretical framework to effectively utilize the significant—yet still unclear—potential of AI, with reference to its ability to rapidly analyze and formalize systemic connections and interactions. The main goal of this study is to evaluate the possibility of reconsidering prominent urban planning methodologies, such as the systemic approach, in this era of “technological singularity.”

2. Literature Review

2.1. First Part: Current Trends and Applications

It is a common opinion that technologies help to make cities more performant, efficient, and smart. The relationship between cities and technologies is not new, and the significant focus of scientific research on this relationship—even though not yet defined—is probably due to the speed of their evolution. Although the smart city concept arose during the 2000s, the study of this relationship has recently gained attention again, almost forgetting its primordial roots. The scientific debate that started in this period has been largely dedicated to the theme of a smart city, finally arriving at the definition of a smart city as a city that “makes optimal use of all the interconnected information available today to better understand and control its operations and optimize the use of limited resources” [11,12]. Some representative phases can be outlined in research focused on smart cities. In the first phase, studies mainly searched for a shared definition of the concept. In the second phase, studies focused on the applicability of the concept to different sectors. In the third phase, the concept was inextricably linked to sustainability, which marks a “point of no return” and led to the establishment that a smart city must inevitably be sustainable. At present, it cannot be said that clarity has been achieved; nevertheless, some convergences can be identified.

Gracias et al. [12] developed a structured literature review examining scientific studies produced in the period 2012–2022. They observed the lack of a unique definition, listing about thirty different ways to define a “smart city,” and finally proposed a new definition with the aim of identifying the goals that a smart city should pursue: “Smart cities use digital technologies, communication technologies, and data analytics, to create an efficient and effective service environment that improves urban quality of life and promotes sustainability” (p. 1723). Some years later, Shao and Min [13] presented an in-depth review on the theme, considering more than four thousand scientific papers on sustainable smart cities with the aim of developing a multidimensional framework to help improve understanding regarding the concept’s applicability. They found that four main pillars (environment, society, governance, and economy) are integrated into the framework to improve the understanding of the concept’s capability. As other authors have underlined, a positive correlation between the adoption of technology and sustainable outcomes does not necessarily exist [14], and the need for an in-depth explanation of the state of smart city development, trends, and approaches to smart city building in the context of rapidly evolving technological fields has emerged for planners and policymakers who are unfamiliar with the smart city concept [15,16]. Methods for assessing the sustainability of SCs vary [17,18,19], depending on the definition of a smart city and the area in which performance indicators are introduced [20,21].

Referring to this first part of the literature review, four main considerations can be highlighted: (a) research on smart city develops after the use and spread of the concept in technological sectors; (b) the need for a clear definition of “smart city” is still a common constraint within the scientific debate; (c) beyond such a definition, the concepts and applications associated with smart cities have shifted to align with sustainability principles; and (d) urban studies are still focused on the theme of smartness in cities. The lack of a theoretical framework to border this topic, as well as to define a common approach, remains the principal weakness of this field of scientific research, especially in the urban planning sector. Urban planning theorization has been slow to mature, despite some scholars’ [22,23] attempts to propose a shift in the vision from smartness to intelligence of a city, underlining that a city is a complex system. The exploration of the city as a complex system, as extensively discussed in the literature [24,25,26,27,28], was transformed when Michael Batty (2013) reflected on the use of Big Data, observing a significant shift in the information available regarding events, locations, and timing within cities [29]. This shift can be considered as a turning point in the literature, which led to the development of studies on the digital twin as an environment for city management and planning [30] and, more recently, on the use of AI in the smart city context.

The relationship between smart cities and AI was first investigated to understand how AI could contribute to various fields of urban activities [31]. Emphasis has recently been placed on the interconnections among smart cities, big data, and artificial intelligence, while also considering sustainability and livability goals in urban development. Scholars studying this topic have warned against the uncritical adoption of technology and advocated for its deeper integration into society, emphasizing that careful calibration and contextualization are essential for creating sustainable cities that are both resilient and intelligent [32,33,34].

The number of review papers published in the last ten years within the scientific literature can be considered as an indicator of the search for delineating the state of the art in research exploring the meaning, impacts, and possible evolution of the relationships between AI and sustainability in cities. Despite this interest, it is not yet possible to affirm that a clear and solid theoretical reference exists.

This study aims to support the efforts of other scholars in defining a theoretical framework for understanding how urban planning processes are evolving due to the potentialities and applications of AI; it also aims to predict the future development of research on this topic. Peng et al. [35] have discussed the current and future problems and solutions that urban planners face when using AI, providing an overview of the application of AI in the regional planning process through a scoping review.

The authors aimed to outline the potential direction of urban planning using AI, spanning from planning support to plan-making. The main contribution of [35] is the definition of urban planning AI as “the process of plan-making that is conducted, either partially or entirely, by an AI agent during the urban planning process” (p. 2268).

The authors also defined a typology of urban planning AI, based on different levels of participation from planners and the autonomy levels of AI agents in the plan-making process. They indicated four phases, in which the role of the AI agent increases and the role of planners decreases, as urban planning AI evolves from AI-assisted planning to AI-autonomized planning (

Table 1. The role of AI in the phases of planning.

Phases	Types	Planner–AI Relationship	Roles *
Phase 1	Assisted planning	The AI agent gathers planning data and resources, identifies urban planning issues and patterns, and assists the planner-led planning process.	Planner   +++ AI   ---
Phase 2	AI-augmented planning	The AI agent simulates and assesses various planning schemes suggested by planners to help to identify the optimal ones.	Planner   ++ AI   --
Phase 3	AI-automated planning	The AI agent semi-automates urban planning processes and suggests the optimal option under the planners’ supervision.	Planner   + AI   -
Phase 4	AI-autonomized planning	The AI agent autonomously creates plans by self-learning from past experiences with minimal input from urban planners.	Planner   -- AI   ++

* The roles of planners and AI in the different phases of the process of planning are indicated in the last column, referring to a higher (+) or lower (-) level of involvement.

).

The study by Peng et al. [35] presents a sound analysis of the prefiguration of the evolution of urban planning from the perspective of using AI to make plans. However, as noted in [36], we must consider that AI will likely inform the plan-making process in the same way that other technologies did before, and that it will be very difficult to totally replace the role of planners—except for their routine activities—as “plan design is very different from the search for good models of routine behavior and this is reflected in the fact that even in routine processes such as in the development of driverless cars, it is the non-routine, the unexpected that cannot be anticipated by an AI” [36] (p. 5).

This statement allows us to clarify the purpose of this work, which aims to understand the roles that AI can play in redefining the approach to studying urban phenomena, considering the complexity of cities as systems.

2.2. Second Part: Methodological Gaps and Challenges in Integrating AI for Urban Planning

Since the 1960s, the theory of general systems (and, later, the theory of complexity) has offered a new perspective for conceptualizing urban systems as Complex Adaptive Systems (CAS), in which non-linear interactions, emergent behaviors, and self-organization govern urban dynamics. This theoretical framework enables exploration of how urban environments evolve and adapt, offering insights into strategies that promote sustainable urban growth while addressing the uncertainties inherent to complex systems. Batty and Marshall [37] showed that cities, as complex adaptive systems, need planning approaches that embrace uncertainty and promote resilience through flexible strategies. Their research indicated that taking principles of complexity into consideration can enhance urban resilience, enabling systems to adapt to changes and absorb shocks from climate change or economic fluctuations. Despite the difficulties scholars have highlighted regarding the application of complex theory to urban planning [38], the systemic approach—which has been further examined using urban complexity and fractal functions [24]—remains the most effective framework for understanding urban systems and physical realities [39], underscoring the importance of interactions among components in understanding their behavior [40].

This perspective views territory as a complex system that evolves autonomously over space and time, with technology acting as a dynamically influential catalyst.

The potential of AI applications in urban planning was envisioned about fifteen years ago; however, their practical implementation has only recently become feasible due to advancements in information and communication technology and the availability of complex datasets. Many studies have referred to systematic reviews and often propose individualizing the main field of application of AI, especially when considered as a tool to manage large amounts of quantitative data. The recent literature has focused on identifying the key applications of AI in urban contexts. For instance, Othengrafen et al. [41] indicated four principal fields of AI application in urban planning (Mobility and Transport Optimization; Energy and Infrastructure; Public Management, Public Health, and Safety; Real Estate, Urban Planning, and Land Use Policies), indicating how AI can be useful for urban governance. In [42], the authors identified various AI paradigms based on their problem-solving capabilities:

• Logic-based tools, used for knowledge representation and problem-solving;
• Knowledge-based tools, based on ontologies and huge databases of notions, information, and rules;
• Probabilistic methods, which allow agents to act with incomplete information and data;
• Machine learning, which allow agents and systems to learn from historical data and to use the gained knowledge to interpret new data;
• Embodied intelligence, an engineering toolbox that has the ability to affect the physical environment;
• Search and optimization, methods that allow for intelligent searches with many possible solutions.

Based on these approaches, the authors defined different typologies of AI (analytical AI; functional AI; textual AI; visual AI; interactive AI), explaining their characteristics and applicability [42].

The relationships between city planning and technology have consistently shaped urban dynamics, with socio-functional changes evolving more rapidly than the physical components of a city. This disparity, which is crucial when studying cities as complex systems, is now amplified by the comprehensive impact of Artificial Intelligence (AI) on urban development. Scholarly reflection on this topic, however, appears weak or entirely absent while, in our opinion, considering both the systemic nature and the complexity of a city is fundamental.

Systemic vision serves as a foundation for the evolution of town planning as a science and for research focused on monitoring urban transformations (Figure 1).

The notion of complexity is linked to the concept of aleatory [43], which suggests that random effects in a system can lead to chaos. The theory of chaos is anchored in the principle of uncertainty [44], which describes the unpredictability of chaotic systems that are highly sensitive to even minor changes, resulting in imprecision regarding their evolution. Complex chaotic systems, therefore, cannot be known with certainty nor can they be subject to long-term predictions, making the testing of theory a very difficult task [45].

Over the past fifty years, urban and regional planning research has embraced these concepts, viewing cities as dynamic and complex systems [46,47,48,49]. The complexity of modern cities, reflecting collective expression and spatial organization, is so great that finding indisputable solutions to urban system challenges is impractical. Regarding the potential for AI to replace planners, it is widely believed that this will not occur in the near future; instead, AI will most likely play an increasingly central role in supporting analyses and decisions.

When considering a city as a dynamically complex system, it is possible to identify several models that may be particularly useful for the disciplinary redefinition of urban planning. Among these, the Hybrid-KAN (Kolmogorov–Arnold Network) approach can be applied to urban planning, especially in contexts where there is a need to model, explain, and predict complex urban phenomena by combining predictive accuracy with model interpretability. These two features make Hybrid-KANs particularly promising for data-driven urban planning and advanced urban modeling. Specifically, the Hybrid-KAN approach can be used for predictive analyses of urban phenomena, forecasting behaviors of actors within the socio-anthropic system, evaluating scenarios and plan alternatives, assessing the consistency between settlement patterns and sustainability goals, and measuring urban entropy [45]. However, a comprehensive exploration of this topic falls outside the scope of this paper, which instead aims to lay the theoretical groundwork for such future studies, as supported by the recent literature [46,47,48,49].

3. The Theoretical and Methodological Framework

Given the limitations identified in the existing literature—namely, the fragmented application of AI and the absence of a comprehensive theoretical framework—this study aims to support the scientific community by defining a theoretical framework for understanding how urban planning processes evolve in response to the potential of AI applications. We posit that a theoretical foundation is essential to move beyond the simple application of AI tools and towards a deeper integration that considers a city as a complex system. The study of a city as a complex system, as extensively discussed in the literature, provides the most effective framework for understanding urban dynamics and physical realities. This systemic approach, which views a territory as a complex system that evolves autonomously over space and time, with technology acting as a dynamically influential catalyst, is fundamental to our work. By adopting this perspective, we can better explore the roles that AI can play in redefining the approach to studying urban phenomena, moving away from a technology-centric view towards an integrated, systemic one. The following sections therefore present a methodological approach that is designed to reflect this theoretical foundation.

Urban planning aims to define the best possible conditions for living in a city, both for its residents and users. It is a “discipline of foresight” La Rocca [50], in the sense that it envisions the future arrangement of human settlements, seeking to understand how the system will evolve, posing rules for transformation and, in recent years, emphasizing sustainable choices. This is a complex process in which the planner plays a central role; however, this role is not sufficiently supported by adequate tools, primarily urban plans and laws, which evolve very slowly compared to the present needs of urban life.

Since the 1960s, the approach to urban studies has moved from a static vision to a dynamic and systemic vision of cities as complex systems [51,52] that evolve continually in time. Concurrently, the role of the planner has changed from a static function, in which they primarily focus on the definition of a plan as a fixed picture of the city in the future, towards a guiding role, in which they drive the city system through different states that can be assumed during its evolution, choosing the best trajectory possible using the resources at their disposal [53,54]. Within the literature in that period, the reference to cybernetics as a science focused on the study and control of complex systems [55,56,57], which also spread to the urban studies sector [58,59,60,61,62,63], affirming the vision of increasing urban system complexity. In this period, starting with the work of McLoughlin [64], the concept of urban planning as a cyclic process became a cornerstone reference. The urban planning process [64] is conducted in different phases that can be summarized as (1) performing context analysis; (2) setting general and specific objectives; (3) identifying actions to achieve these objectives; (4) evaluating actions based on costs and benefits; and (5) implementing feasible actions. The sequence of phases is a continuously monitored circuit, allowing planners to control the system’s evolution towards the desired state of transformation that they want to obtain.

Interpreting this theoretical framework, it is possible to define the process of urban planning as a cyclic process called Urban Transformation Governance (UTG), which develops in three main phases (Figure 2): (1) knowledge, (2) decision, and (3) action.

The knowledge phase (1) aims to define a cognitive framework of the system conditions in the starting stage of the cycle as completely as possible. The decision phase (2) focuses on the definition of strategies and objectives to achieve. It is the central moment of the process, in which choices must be made based on the available resources, in order to reach the objectives of transformation. The action phase (3) corresponds to the effective transition of the system from the starting stage to the final stage—that is, the best possible transformation state for the system—considering the starting conditions and the available resources. As a complex and dynamic system, there is no absolute final condition but, instead, multiple conditions that can correspond to equilibrium states of the system. Both the monitoring of transversal action and the circularity of the process are meaningful in this context.

Given the complex nature of urban systems, they change constantly [65] and are non-stationary [66,67]. This means that it is not possible to find an optimized and fixed solution, as the optimal solution changes with the problem. The adaptive ability of complex systems and the study of this behavior represent the key interpretation in the evolution of urban planning approaches [68].

The study of adaptivity in systems was developed within the context of cybernetics [55]. A key concept from cybernetics is the control loop [69], where a controller receives input from the system it regulates and exerts outputs to influence it. As the controlled system has its own dynamics, the controller must perceive changes, make decisions, and take actions to maintain the variables within a desired state.

Referring to the systemic approach, elements and the relations between them must be investigated. However, system complexity impedes the predictability of its transformation trajectories, and an interpretative model is needed to reduce the degree of uncertainty.

Considering complex system properties, it is possible to identify three main urban subsystems that can be analyzed to understand the initial condition of the whole urban system. The first subsystem encompasses a city’s material components, including streets, squares, buildings, infrastructure, settlement patterns, urban layout, and all other components of the urban system that compose its physical part and that mainly refer to adapted spaces (i.e., “where” components). The second subsystem is composed of immaterial components, such as the urban activities that make the urban system alive and dynamic. Urban activities can be defined as human actions inside adapted spaces within a city (i.e., “what” components). The third subsystem refers to social components, including the residents, stakeholders, and city users or actors who carry out activities inside adapted spaces within a city (i.e., “who” components) (Figure 3).

This reasoning deserves to be explored in depth and has numerous developments in the literature. It also served as the basic theoretical reference for the objectives of this study, in terms of investigating the possible roles of AI in this process loop and where and how AI could intervene in achieving a state of dynamic equilibrium. In the next section, we explore this issue in detail, examining the three main phases of the transformation process and considering the roles that AI can play in each phase (Figure 4).

3.1. AI in the Knowledge Phase

In this study, the knowledge phase is the phase in which the characteristics of an urban system are identified and defined. The final product of this phase consists of a fact-finding framework allowing for the interpretation and synthesis of information; it also involves the individuation of variables, allowing for measurement and representation of the initial state of the urban system. This phase is crucial, as it forms the groundwork for the key decisions and actions that drive the system’s evolution.

During this phase, it is important to collect and process data that can help to configure the initial conditions of the urban system. Advances in AI offer new opportunities for urban planners, especially to accelerate this step of the planning process. Even though they have not yet been clearly ascertained, the roles of AI in this phase serve to support the expertise of urban planners, certainly not to replace it [70]. There are many examples describing the use of AI to collect, manage, and process large amounts of data, and it is clear that a core question is how to formulate queries to obtain reliable solutions. The use of ChatGPT models (generative AI) has spread in urban planning, both to manage data and to delineate land use future configurations [70]. For the knowledge phase, as data and information constitute the basis for the decision and action phases, ChatGPT models are becoming an almost indispensable tool. AI applications, like every technology, must be well understood for their conscious and productive use. To generate solutions, AI must be taught to perceive and understand the instructions of human planners and it is clear that, in this phase, the key role of AI is to support human planners in the elaboration of the “state of the art” of the urban system. This condition can be defined as a “collaborative phase,” in which the planner plays a predominant role in controlling the correctness of elaborations. Figure 5 shows the result of an exercise during technical planning with engineering students. The students were studying the territorial characteristics of the city of Cercola, close to Naples (Italy), with the aim of defining a final master plan. In the knowledge phase, the individuation of historical patterns was a principal step of the urban analysis. The students identified patterns using GIS during the first step of the exercise; later, the students were asked to use ChatGPT for the same purpose. The results are shown in Figure 4, in which image (A) shows the results obtained using GIS, while image (B) shows the results obtained using ChatGPT to illustrate the distribution of historical patterns along the main streets.

3.2. AI in the Decision Phase

The decision phase is crucial in urban transformation, in which future assets are envisioned based on acquired knowledge. The roles of AI in decision-making have been well-documented in the literature, and this topic has recently gained attention in the context of urban planning processes. Generative AI enhances planners’ roles by enabling them to understand and simulate the potential outcomes of their decisions, helping to achieve transformative objectives. Due to its potential, AI can generate possible scenarios simulating the impacts of choices, allowing human planners to reduce uncertainties and develop solutions to mitigate risks for urban system evolution. The relationship between planners and AI continues to be collaborative, with AI agents working in an autonomous manner to elaborate optimal schemes and support planners in their final decisions. In this phase, the use of digital twins (DTs) has been largely experimented with to develop simulation models and manage urban transformation scenarios. Many cities in Europe (Cambridge, Gothenburg, Herrenberg, Helsinki, Stockholm) have used DTs to simulate the response of the urban system to specific scenarios as well as to show, through the virtual representation of reality, the state of the environment and associated risks, improving the involvement of stakeholders and citizens in a short timeframe. Both the definition and use of DTs are still objects of debate, even though a convergent point can be identified in the conviction that a DT is much more than a 3D city [71], thus enhancing the smart city model. Some scholars [72,73] have suggested that a digital twin (DT) can be defined as a realistic digital representation of the system of anthropic–infrastructural elements and, less often, of the physical naturalistic system that presents some invariable characteristics. The main advantage of a DT is that it can simulate the behavior of a complex system and enable physical objects to be controlled remotely in real-time. Thus, a digital twin expands the possibilities to monitor and predict the condition of complex physical systems, while its predictive ability contributes to assessing alternative scenarios and possible solutions to the problems and/or errors that can occur in cyber–physical systems. In Europe, the most famous success case is probably the Helsinki DT [74]; however, many other cities have followed its example and have developed digital twin platforms to improve the decisional phase of the planning process.

3.3. AI in the Action Phase

The action phase focuses on achieving the optimal future state of the urban system, enhancing the overall functioning of the city. This phase requires identifying specific actions, stakeholders, and necessary resources to meet the established goals. As a complex, evolving system, a city will not reach a static state. In this phase, the role of AI is expected to be significantly stronger and more autonomous than in previous stages. In this sense, it may be useful to develop AI agents dedicated to interfacing with the urban community and the various stakeholders involved in the process, in order to collect and formalize opinions, insights, and suggestions from the social subsystem. In this stage, it is essential to effectively synthesize all the perspectives and inputs from the actors within the urban arena. However, human planners remain essential, as they possess a comprehensive understanding of the system’s evolution and can ensure that AI-generated plans align with the expectations of all stakeholders, including citizens and users. The term “strong AI” has been used in various disciplines [75,76]; it has been partially used as a synonym for high-level machine intelligence, human-level AI, or general AI and, more recently, to indicate a Cognitive AI as being able to develop common-sense reasoning abilities [77]. However, this level of AI remains largely theoretical. Within the scientific literature, the use of AI in the decisional phase is limited to the involvement of stakeholders and citizens and models to generate automated, optimized land-use plans. Wang et al. [78] have demonstrated the possibility of teaching AI to obtain an automatic plan. However, they also concluded that although the model they developed (LUCGAN) was able to generate land-use configurations using contextual data, it was not able to integrate the requirements requested by the human experts. This limitation stems from AI’s current inability to understand the nuances of expert considerations, which are necessary for effective land-use planning. They concluded that although urban planning can be assisted by AI in the present stage, the use of AI is much more remarkable in configuring the assets of urban systems rather than in the decisional and active phases, in which “human choices” are essential (at the moment) to guarantee equity and sustainability.

4. Discussion

The use of AI in urban planning seems to be a growing experimental research path and may also be considered as an evolution of the smart city approach [79]. Even though urban planning and AI developed separately, much attention is currently being given to the possibilities that their interconnection could offer in the relevant literature. Nevertheless, the question is still open and without a defined perspective, at a theoretical level as well as in practice. Urban planning, as a discipline that prefigures the future evolution of cities, is responsible for the quality of life of citizens and users. As the urban population continues to increase, the roles of urban planning and planners are crucial to ensure sustainable development, social equality, resource use awareness, service efficiency, and so on. The successful integration of cutting-edge technologies into the planning process has demonstrated that it is already possible to optimize the performance of the global organization of urban systems. The smart city paradigm has revealed that the intelligence of cities lies in the convergence of different elements at stake, indicated as the six dimensions of the well-known smart city model (Smart Government, Smart Economy, Smart Environment, Smart Living, Smart Mobility, and Smart People); above all, this paradigm underlines that the complementary roles of technology must be taken into consideration.

The complexity of cities, assumed as dynamic systems, must be understood as different aspects that cannot yet be replaced by technologies. However, potentialities must be recognized, and the roles that technologies—particularly AI—can play in defining increasingly precise cognitive frameworks is indubitable.

The Snap4City platform (https://www.snap4city.org, accessed on 10 March 2025), currently in use in many urban areas, can be considered as a possible transition from the smart city model to the Artificial Intelligence city model in the sense that it could be seen as an experiment outlining new theoretical paths for governing urban and territorial transformations.

Moving from a technocentric perspective, this study attempted to highlight the need to define a more solid theoretical framework to both appreciate generative potential of AI and appropriately adopt its capability to prefigure innovative solutions to cities’ challenges, creating sustainable, resilient, equitable urban scenarios.

As a particularly relevant aspect in UTG, artificial intelligence can assist in various phases by identifying specific activities when triggered through well-designed prompts.

The proper sequencing of these prompts is crucial for effective human–AI collaboration, and they must be tailored to the sector of application. In other words, during the sequence of interactions, it is necessary to consider the responses provided by the consulted agent at that moment and the process implemented by the synthetic reasoner. For example, for certain actions, using and subsequently verifying Large Language Models (LLMs) that can perform targeted web searches may be more appropriate. In this case, LLMs that are capable of citing their sources could be particularly useful, especially in terms of controlling and verifying hallucinations. Moreover, particularly in the decision-making phase, it is advisable to refer to LLMs that can use Thought Chains and present the reasoning steps that led to a specific result.

5. Practical Implications of AI Use in UTG Process

Integrating multiple LLMs across different phases of territorial transformation governance seems more promising than relying on a single model for planning activities. Considering recent developments, it also seems possible to integrate several model environments which are specifically customized to carry out certain activities in the urban planning process. While initial attempts have been limited, following Peng et al. [35], this study argues for distributing LLM support based on the specific phase of territorial transformation governance. A preliminary categorization of related LLMs that can be adopted in the three phases of UTG (knowledge, decision-making, and action) is outlined below.

In the knowledge phase, AI-powered LLMs can perform actions aimed at:

• Supporting the classification of satellite images of the territory in question, with the ability to automatically recognize spectral signatures.
• Analyzing urban economy data and producing summary reports.
• Analyzing areas and functions established in the territory.
• Processing and extracting insights from demographic big data obtained through population censuses.
• Supporting the creation of dynamic GIS environments through automatic code generation for original plugins.
• Engaging the public in three specific phases:
  • Collecting collective sentiment through chatbots coordinating specific online forums.
  • Using models to formalize the emotional state of the population regarding various urban transformation proposals.
  • Building perception-based foresight scenarios through augmented and mixed reality to digitally simulate territorial transformation.

In the decision-making phase, AI can assist in:

• Automatically defining transformability maps by collecting and integrating territorial constraint data.
• Formalizing available resources and defining possible development scenarios.
• Creating and coordinating collective participation arenas through the use of virtual agents.
• Assisting in the drafting of regulations by training models on updated municipal regulations.
• Automatically generating environmental impact matrices and prefiguring assessments.

In the action phase, AI can facilitate:

• Supporting the selection of the best development scenario based on municipal administration guidelines for possible transformations.
• Assisting in drafting technical implementation standards.
• Writing urban and building regulations.
• Developing a virtual agent to communicate the plan’s guidelines.

A preliminary indication of different models that can be integrated to support the listed activities is provided below. For knowledge-related activities, AI-powered LLMs could perform the following functions:

• Supporting satellite image classification with automatic spectral signature recognition (e.g., Google Gemini, IBM Watson).
• Analyzing urban economy data and producing summary reports (e.g., OpenAI GPT-4, Anthropic Claude).
• Analyzing areas and established functions within the territory.
• Processing and extracting insights from demographic big data obtained from population censuses.
• Supporting the creation of dynamic GIS environments through automatic code generation for original plugins (e.g., Code Llama by Meta).
• Engaging the public through three specific phases:
  • Collecting collective sentiment through chatbots coordinating specific online forums (e.g., ChatGPT with sentiment analysis APIs).
  • Using models to formalize the population’s emotional state regarding urban transformation proposals (e.g., Hume, Sesame).
  • Building perception-based foresight scenarios through augmented and mixed reality to digitally simulate territorial transformation (e.g., Unity AI and Blender).

AI support in the decision-making phase can include the following:

• Automatically defining transformability maps by integrating territorial constraint data (e.g., Esri ArcGIS AI).
• Formalizing available resources and defining potential development scenarios (e.g., DeepMind AlphaTensor).
• Creating and coordinating collective participation arenas through virtual agents (e.g., Anthropic Claude 3).
• Assisting in regulation drafting by training models on updated municipal regulations (e.g., LLaMA by Meta).
• Automatically generating environmental impact matrices and prefiguring assessments (e.g., Bard with environmental analysis tools).

In the action phase, AI can facilitate the following activities:

• Supporting the selection of the best development scenario based on municipal administration guidelines for possible transformations (e.g., prompt sequencing and UrbanSim AI).
• Assisting in drafting technical implementation standards, including training LLMs on examples of existing municipal regulations.
• Automatically drafting plan reports (e.g., Notebook LM, Cortex).
• Writing urban and building regulations (e.g., AI21 Labs).
• Developing a virtual agent to communicate the plan’s guidelines (e.g., Rasa, IBM Watson Assistant).

The integration of LLMs into urban planning processes could benefit from the development of hybrid work environments that combine generative artificial intelligence, advanced GIS systems, and urban simulation tools. The evolution of AI platforms to facilitate their interaction with dynamic, real-time territorial databases could represent a significant advancement in city planning and management. Furthermore, improved explainability (Explainable AI) would allow urban planners to better understand the reasoning behind AI-generated decisions, fostering greater transparency and acceptance of proposed solutions.

This aspect is crucial for effectively integrating AI into urban governance processes.

Finally, the adoption of AI-based tools necessarily requires a regulatory framework defining their limits and responsibilities. The involvement of public institutions, universities, and technology companies will be essential in establishing international guidelines and operational standards.

The study of the support provided by AI in urban planning must take multiple approaches and necessarily consider the different characteristics and potential of each LLM. This remains a key area for further exploration.

6. Final Considerations

The debate on the roles of AI in urban planning has already produced interesting insights [77]. However, this study emphasizes the need to change the perspective regarding the significant changes that AI is causing in every aspect of human activity. When studying urban phenomena to develop actions in territorial transformation governance, it is necessary to adopt AI to interpret the transformations of city systems. This is now possible thanks to the ability of AI to formalize the structure of systemic relationships, thus enabling the prefiguration of future states of the urban system.

Recent advances in AI offer urban planners new opportunities by enabling deeper insights into complex urban systems, like land use and transportation. AI algorithms and data analysis techniques allow for more informed decision-making and address the limitations of traditional planning methods. For example, AI models facilitate the rapid processing of large geospatial and social datasets, identifying patterns and trends that may have been previously undetectable. However, the integration of AI in urban planning also presents challenges and concerns. As this study aimed to show, AI tools could serve as planning assistants, enabling human planners to adjust machine-generated plans to accommodate specific requirements. Scholars [78,79,80] agree on the prefiguration of collaborative planning between AI and human expertise, fostering a multidisciplinary approach.

The contribution of this study to the ongoing debate lies in its attempt to elaborate a reasoned reflection on how such integration can take place. Furthermore, as repeatedly emphasized, the aim of this study is to propose the creation of an innovative and solid theoretical framework that can serve as a foundation for the development of updated educational pathways and land management processes that consciously adopt AI support, avoiding uncritical acceptance. As a concise definition, this contribution proposes a perspective of reasoning opposite to that which gave rise to another interesting field of AI in Planning Support Systems, which is related to PlanGPT and/or ChatGPT-like models [80].

7. Limitations of This Research

The main limitation of this research lies in its lack of reference to case studies, best practices, or urban plans developed using AI. However, this contribution aims to draw the attention of researchers and academics to the need to develop new theoretical approaches that appropriately incorporate AI in the shaping of innovative processes for urban and regional planning. AI is transforming urban planning by offering new ways to understand and manage complex urban environments. However, research is needed to ensure the sustainable and equitable application of AI in the context of urban development. There are still many open questions, and it is considered highly important to define the impacts of AI not only as a tool but also in terms of the influence that it could have on human choices and behaviors [78,79,80,81].

This study reflected on the need to assume theoretical references for the application of technologies before they spread, which inevitably leads to unfocused technological evolution rather than their conscious use.

Even though this need has been generally expressed in the literature, there is a lack of uniformity in references and methodologies that this study does not resolve. The interpretative model of city complexity can also be debated, as many other subsystems can be indicated as components of a complex urban city. Nevertheless, in our opinion, this study delineates a possible research path that can be followed with appropriate awareness.

8. Limitations and Future Research Perspectives

The primary limitation of this study is its theoretical nature, which is a conscious choice that aligns with its core objective. Specifically, this research does not include empirical validation through case studies, best practices, or urban plans developed with AI. While this approach allows for a deep theoretical exploration, it necessarily limits the direct application of our proposed framework to real-world scenarios.

However, this limitation is a necessary precursor to our main goal: to address the pressing need for a robust theoretical foundation for AI in urban planning. At present, there is a lack of uniformity in the references and methodologies used in this nascent field. This study, therefore, aims to draw the attention of researchers and academics to the need for new theoretical approaches that can appropriately incorporate AI, shaping innovative and conscious processes in urban and regional planning.

Building upon the framework presented in this paper, future research can proceed in several key directions, including the following:

• Empirical validation: The most immediate step is to apply the proposed methodological framework to a specific case study. This would involve testing its efficacy, identifying practical challenges, and refining the model based on real-world data and outcomes.
• Investigating the Human–AI Nexus: As this study highlights, a crucial open question is to define the impacts of AI not only as a tool for planning, but also in terms of its influence on human choices and behaviors. Future research should delve into the ethical and social implications of AI-driven planning, ensuring sustainable and equitable applications that prioritize human-centric outcomes.
• Developing a unified theoretical framework: While this study delineates a possible research path, it does not resolve the lack of a uniform theoretical reference in the literature. Future work should focus on building consensus around key concepts and methodologies, consolidating the field, and providing a shared foundation for all subsequent research.

The effort in this study lies in delineating a viable and necessary research path. By establishing a sound theoretical and methodological framework, it could be possible to move beyond simply reacting to technological evolution, thus enabling the use of AI in a more conscious and deliberate way. Further research should move beyond this theoretical foundation to apply and test the proposed framework in real-world urban planning scenarios. This will require the implementation of advanced AI architectures which are capable of handling complex urban data. Hybrid AI models, such as the Hyb-KAN approach [81], could offer a promising direction, as they have shown the ability to process diverse, multi-domain data and capture dynamic dependencies. It would be interesting to explore how such architectures can be adapted to analyze urban data streams (e.g., social media activity, traffic flow, infrastructure sensor data) to forecast urban emergencies, model resident behavior, and enhance AI modules within urban planning support systems. This empirical application could both validate the theoretical claims of this study and provide a possible roadmap for the conscious adoption of AI in urban governance.

Finally, it is necessary to consider the extremely rapid evolution of LLMs, considering the enormous amount of funding in this field. Such advances are expected to enable further insights and the characterization of the proposed process.