Contributions and Risks of Artificial Intelligence (AI) in Building Smarter Cities: Insights from a Systematic Review of the Literature

: Artificial intelligence (AI) is one of the most disruptive technologies of our time. Interest in the use of AI for urban innovation continues to grow. Particularly the rise of smart cities notion—urban locations that are enabled by community, technology, and policy to deliver productivity, innovation, livability, wellbeing, sustainability, accessibility, good governance, and good planning—has increased the demand for AI-enabled innovations. There is, nevertheless, no scholarly work that provides a comprehensive review on the topic. This paper generates insights into how AI can contribute to the development of smarter cities. Systematic review of the literature is selected as the methodologic approach. Results are categorized under the main smart city development dimensions—i.e., economy, society, environment, and governance. The findings of the systematic review—containing 93 articles—disclose that: (a) AI in the context of smart cities is an emerging field of research and practice; (b) Central focus of the literature is on AI technologies, algorithms, and their current and prospective applications; (c) AI applications in the context of smart cities mainly concentrate on business efficiency, data analytics, education, energy, environmental sustainability, health, land use, security, transport, and urban management areas; (d) There is limited scholarly research investigating the risks of wider AI utilization, and; (e) Upcoming disruptions of AI on cities and societies have not been adequately examined. Current and potential contributions of AI to the development of smarter cities are outlined in this paper to inform scholars of prospective areas for further research.


Introduction
There exists a strong scientific consensus that anthropogenic climate change is the biggest crisis of our time [1,2]. In a rapidly urbanizing world, climate change and the misuse and mismanagement of land and resources are triggering natural disasters and increasing their intensity [3,4]. Subsequently, cities are becoming frequently subjected to the direct or indirect impacts of natural disasters-for example, the 2019 Amazon Rainforest Fires [5], and the 2020 Australian Bushfires [6]. There have been numerous top-down (e.g., the Paris Agreement, Intergovernmental Panel on Climate Change, UN's Sustainable Development Goals, UN Climate Change Conferences), and bottom-up (e.g., school strikes, extinction rebellion protests, climate emergency declarations) attempts to raise awareness and develop policy actions to address the climate emergency [7,8].
These efforts provided some hope, despite the political and policy quagmires in many countries. Nevertheless, there has been no significant climate action undertaken to address the crisis. Instead, in recent years, with the advancement of the current digital revolution, a large While we have treated each AI application in isolation, it is common to have them bundled and integrated. Researchers from Carnegie Mellon University collaborated with the City of Pittsburgh to develop a Scalable Urban Traffic Control (SURTRAC), which was able to simultaneously monitor and control the flow of traffic [49]. The system has been deployed in the East Liberty neighborhood since 2012 and covers nine intersections. On average, 29,940 vehicles pass through this area daily. SURTRAC is a schedule-driven system designed to manage multiple competing traffic flows shifts. SUTRAC is a multi-agent decentralized system, where an agent system runs each intersection. Each agent system controls traffic signals for their intersection and monitors traffic flow by dynamically coordinating with other agents in real-time. The deployment of SURTRAC resulted in a 34% increase in vehicle speed, and a reduction of 25% in travel time, 40% in waiting time, 31% in traffic stops, and 21% in emissions.
While AI systems have significant potential, their deployments are never straightforward. In Detroit, a $9 million initiative 'Neighborhood Real-Time Intelligence Program' implemented facial recognition software and video surveillance cameras at 500 different Detroit intersections. This initiative built on the previous 'Project Green Light' Initiative, which installed 500 cameras outside of businesses capable of recording and reporting real time video footage to the police. The software boasts an ability to match faces with 50 million driver's license photographs in the Michigan police database. Nevertheless, recent research has shown that current facial recognition software more often misidentifies black faces than white faces [50]. Whilst intended to increase public safety, there is widespread public criticism towards this technology, as residents feel their privacy is compromised and knowledge of the racial biases continue to increase.
Fake news in its purest form are completely made up information, nonetheless, they are often hard to identify as they resemble credible journalism and attract maximum attention as they spread like wildfire through various social media channels [51]. A man was caught after he carried an assault rifle and fired shots at a pizza parlor in Northwest Washington Upon arrest, the man informed the police that he was investigating a conspiracy theory which claimed that the pizza parlor-Comet Ping Pong-was the headquarters of a pedophilia ring [52]. This incident caused panic among people in the neighborhood, resulting in lockdown of several businesses.
In Arizona, the hotbed for testing of AVs by major technology providers, we have seen incidents of residents throwing rocks at these vehicles [53]. There AVs are seen as threat to jobs, livelihood, and are a source of frustration. As noted by Selby & Desouza [54], "If theory and practice advance over the next few years without paying attention to fragility, then cities will continue to be vulnerable to manageable threats. As the trends of urbanization continue, it is even more imperative to attend to fractures of social compacts. Cities will continue to grow, and their complexity will only increase. This complexity will continue to mask fragility in the city and could result in the breakdown in one of our society's most valuable artifacts, developed cities, representing a potential loss of life and economic prosperity". AI technologies, for all their good, do make cities more fragile [55] as they put pressures on local governments to maintain existing, and strengthen, social compacts in the face of job losses, automation, shifts in public finances, and so on.
In May 2016, a Tesla Model S car collided with a tractor-trailer in Williston, Florida. The accident occurred on the highway when the car, on autopilot mode, collided with the truck while crossing an uncontrolled intersection. The Tesla driver sustained fatal injuries, raising several questions about the autopilot functionality. However, the National Highway Traffic Safety Administration's (NHTSA) final investigation report concluded that the accident was caused by the driver's inattention [56]. Since the accident, Tesla has implemented several features to keep drivers engaged while their car is in autopilot mode.
Chrysler recalled about 1.4 million cars and trucks, because these vehicles could be hacked remotely over the Internet [57]. These vehicles used UConnect features to connect with Sprint networks for navigation. Hackers could access these cars' navigation systems to control air conditioners, cut off brakes, and shut down engines, and so on. Software installed in vehicles need to be constantly upgraded to protect against hacks and security breaches [58].
Economically, local governments may lose revenue streams because of AVs due to a decrease in speeding tickets, towing fees, and driving under influence charges [59]. Cities in Arizona such as Phoenix and Mesa collected about $10.8 million and $4.2 million from drivers for traffic violations [60]. On an average, cities in California generated $40 million in towing violations annually [15].
The conceptual and application background of AI as presented above underlines the importance of further investigations into how we can best integrate AI systems in addressing critical urban issues. Particularly the challenges we face-e.g., climate emergency-calls for smarter systems in place. This is also highly critical to increase the smartness of our cities. A recent study that evaluated the smartness levels of 180 Australian local government areas argues the importance better integration of urban technologies, including AI, into local service delivery and governance [14].

Materials and Method
We undertook a systematic review of the literature to answer the following research question: How can AI contribute to the development of smarter cities? We adopted a three-phase methodologic approach (Figure 2) following the study steps of Yigitcanlar et al. [61]. The first phase is planning, which involves developing the research aim, research question, a list of keywords and the criteria for inclusion and exclusion of articles. The research aim was framed to generate insights into forming a greater understanding on how AI can contribute to the development of smarter cities. The inclusion criteria were intended to be peer-reviewed journal articles, that were available online, in English, and had relevance with respect to the research aim. A university's library search engine, which gives access to 393 different databases including: Directory of Open Access Journals, Science Direct, Scopus, TRID, Web of Science, and Wiley Online Library, was used to complete an online search. The search was carried out towards the end of December 2019 using the query string of (("artificial intelligence" OR "AI") AND ("smart" OR "smarter" OR "smartness") AND ("city" OR "cities" OR "urban"))) to search the titles and abstracts of available articles. The publication date was left open. From this search it was determined that the one of the earliest studies on AI and the city was from Schalkoff [23]. The abstracts were then read and if the article was considered to be relevant to the research aim the full-text was reviewed to decide whether it was suitable to include in final analysis.
The second phase involved carrying out the review of relevant articles. The initial search resulted in a total of 799 records. These records were then screened and reduced to 245 by applying the inclusion criteria-i.e. journal articles that were peer-reviewed, and available online. The articles were then 'eye-balled' to ensure they were consistent with the keyword search, the abstracts assessed against the research aim, and duplicates removed. The total number of articles was reduced to 208. The full-text of the selected articles were read to determine the relevance with respect to the aim of the study and the results were narrowed down to 119 articles. After another round of full-text screening, the number of articles was reduced to 93. Finally, these 93 articles were reviewed, categorized, and analyzed. The criteria for formation of the themes is presented in Table  1. For the categorization the main smart city development dimensions-i.e., economy, society, environment, and governance-were selected. Figure 3 below presents these dimensions in the context of smart cities.  The third, and final phase is reporting and dissemination. This phase involved critically documenting and presenting the results from the 93 articles analyzed. A discussion of the perceived benefits, and concerns associated AI implementation were outlined.

General Observations
The first step in the analysis of the selected 93 articles was to classify them by date of publication. This process revealed interest in AI technology had grown exponentially in recent times. Almost half of the reviewed articles were published in 2019 (n=46; 49%), one-fifth in 2018 (n=18; 20%), and nearly one-tenth in 2017 (n=8; 9%). Slightly less than a quarter of all articles were published before 2016 (n=21; 23%). The earliest article included in the literature review was published in 1999-despite articles on AI and the city dated back to 1990, the earliest one that meet our selection criteria was [63]. These figures reflect results of other review studies and are representative of the increased interest in how AI systems interact and impact the development of smart cities [64].
Leading authors are affiliated with institutions in Europe (n=28), East Asia (n=24) and North America (n=21). Nonetheless, the Oceania context, particularly in Australia, is often referenced among the best applications of AI in practice [65,66]. Regarding the academic journals the articles were published in, IEEE Access appears most often (n=8), followed by Sensors (n=5), and then three articles in both Artificial Intelligence in Medicine (n=3) and Procedia Computer Science (n=3).
Articles were categorized under four groups. These groups were based on the main smart city domains of: (a) Economy; (b) Society; (c) Environment; and (d) Government. More than one-sixth of them (n=17) were in the economy domain, slightly more than a quarter (n=24) in the society domain, over one-third (n= 33) in the environment domain, and slightly less than quarter of them (n=19) were in the governance domain.
With reference to the 'AI Knowledge Map' (Figure 1), the majority of articles discusses AI applications or techniques that fit within the machine learning paradigm (n=71), followed by probabilistic methods (n=25), knowledge-based (n=21), search and optimization (n=21), logic-based (n=18), and embodied intelligence (n=15). Regarding AI applications the most common mentioned in the selected literature are neural networks (n=55), evolutionary algorithms (n=21), distributed artificial intelligence (n=15), computer vision (n=13), and decision networks (n=9). The most common AI paradigms and applications identified in the literature are outlined in Table 2. A complete list of all articles and the identified AI paradigms, applications, techniques, and supporting technology are shown in Appendix A. Please note, there is some crossover with the number of AI paradigms and applications identified in the literature as many articles discuss multiple, or hybrid AI systems.

AI in the Economy Dimension of Smart Cities
Papers categorized under the economy dimensions of smart cities are those that provide insights into 'how AI can contribute to and enhance the productivity and innovation of smart cities' [33]. Research in this area focused predominately on the technological innovation, economy and management areas, and the contribution of AI can be summarized as: (a) Enhancing productivity and innovation by automating data management and analysis; (b) Reducing costs and increasing resources through pattern recognition; (c) Supporting decision-making by analyzing large volumes of data from multiple sources, and; (d) Drawing conclusions based on logic, reason and intuition. Appendix A lists the analysis highlights of the reviewed literature.
AI is a useful tool to quickly and accurately manage and analyze large volumes of data to support business decisions [103,112]. This is particularly relevant in combination with IoT-a system enabled by the internet, which allows communication between a large network of devices without the need for human intervention [77,106,110]. Together with technologies such as blockchain, cloud storage and fog computing-which help facilitate the recording, distribution, storage, and decentralized processing of data [128,149]-, AI could improve productivity by automating the data management process and removing the need for intermediaries, and hence increasing profitability [126]. Furthermore, AI can improve the stability and effectiveness of IoT contributing to improved network communication; this in return would help to improve knowledge sharing, and foster innovation and entrepreneurship [106,127,128].
AI can be used to recognize patterns in datasets, helping to optimize the data management process, improve the overall productivity of the data management system [111,149], identify cyberattacks [107], coding errors, and other inefficiencies [103]. Deep learning has already had success recognizing patterns from a wide range of data sources including images, audio, video, and other sensors [104]. The application of AI has the potential to remove the need for humans to complete many repetitive business tasks-particularly those relying on observation-potentially reducing costs and freeing up resources for more productive or innovative fields [112,130].
In analyzing the large amounts of data collected by the sensors, devices and other sources in a smart city, AI has the potential to accelerate the decision-making process by automating complex statistical analyses [111,130,150]. This is particularly relevant with regards to the application of deep learning technology to the process of data fusion-i.e. the processes of taking data from a variety of sources, combining it, and improving its quality and usefulness by producing sophisticated statistical models [126]. AI has the capability to automate this process and conduct statistical analyses that are much larger and more complex than could be completed with human intervention. This information can then be used to reduce economic uncertainty and assist business decisions [77,103,107], and/or create marketplaces that are more responsive to user needs and desires [130].
The ability of AI to complete complex statistical analysis can also be used to automate decisionmaking [128,130,150]. The ability to learn, can ensure that AI systems are responsive to uncertainty, particularly when working with and around humans [94,119]. This can reduce the possibilities of accidents, errors, and improve the operational efficiency of business and industries [119]. Smart control systems can monitor traffic, collect and analyze data and in combination with connected-AVs there is potential to make real-time decisions which enhance the efficiency of transport operations-including freight and supply chain logistics [104,130]. AI systems can be developed with a human-like abilities such as creativity, design, intuition, inventiveness, trust, ethics and values to perceive, understand and make informed reason-based decisions that would benefit companies [76,94].

AI in the Society Dimension of Smart Cities
Papers in this category provide insights into 'how AI can contribute to and improve the livability and wellbeing of citizens in smart cities' [33]. Research in this area focused predominately on the health and education sectors, and the contributions of AI can be summarized as: (a) Improving health monitoring; (b) Enhancing health diagnosis outcomes, and; (c) Providing autonomous tutoring systems that are highly individualized and adaptive to needs and external changes. Appendix A lists the analysis highlights of the reviewed literature.
AI systems, in combination with sensors, cameras and other data collection devices, have been developed to monitor the health and wellbeing of individuals [70,87]. Machine learning techniques can be used to improve the cost and efficiency of fall detection devices [99], and detect changes in sleep, mood, heartbeat, respiration, and other vital signs [118,129]. Wearable devices, or 'smart textiles, enabled by AI, can detect changes in the human body and report findings to health care providers [129]. In intensive care settings, monitoring devices equipped with AI can be used to adjust the settings on bedside devices reducing total healthcare costs and improving patient outcomes [67]. In rural settings, the ability to monitor patients remotely could contribute to reduced inequities and improve access to healthcare [89]. This is a particularly critical issue in countries like Australia, where in some remote areas the nearest health service provider or hospital could be located 1,000 km or further [151]. Additionally, smart tracking of health symptoms could improve communication between patient and health care professional [142].
In addition to health monitoring, AI systems can greatly improve health diagnosis by providing an effective repository of medical knowledge and the ability to access, analyze, and apply complex medical data more efficiently [68,139]. Assisting health care workers with tasks involving the collection and recording of data and knowledge could increase the amount of resources available for patient care [68,139], improve the quality of life for patients [120], and expand the professional learning capabilities of professionals [122]. Similarly, the improved analytics and reasoning capabilities of AI would provide a decision-support mechanism with the potential to reduce chances of misdiagnosis [137], facilitate greater communication and collaboration between health care professionals [70] and assist with the development of more personalized medical treatments [137,91].
With regards to the education sector, intelligent tutoring systems that mimic the one-on-one interaction between tutor and student can provide highly individualized teaching programs for students [140]. Furthermore, these systems can develop multiple paths to answer any given question and provide highly detailed feedback [140]. Advances in AI technology can increase the effectiveness of these, and similar systems, by automatically collecting information from the web, ensuring the most up-to-date content, and using machine learning to increase the adaptability to individualized learner requirements [81,146]. There is potential to create systems that are far more effective than one-on-one tutoring [143], with improved communication between student and teacher, and superior assessment methods [85]. This is especially promising with regards to identifying and adapting syllabus to the individual strengths and needs of students with learning disabilities or other special learning requirements [71].
Increased adaptability is important as rapid technological changes are likely to result in an unstable job market [135,143]. AI could potentially bring new skill requirements across multiple sectors and the education sector needs to be at the cutting edge of these changes to ensure students are prepared for future job markets [152]. For example, even in the education sector itself, AI will replace many of the time-consuming tasks, changing teacher roles to one based on student support and the management of AI systems [88,135,153]. Managing the education needs of residents is therefore important to ensure they are able to take advantages of the potential benefits of AI including improved working conditions, better work-life balance, and improved quality of life [135].
Lastly, AI has been used for the modelling of the spread of the recent COVID-19 epidemic. The predictions seem to be showing reliable results as the modelling predicts COVID-19 infections with an accuracy of 96%, and deaths with an accuracy of 99%, up to one week into the future. This information would help governments implement effective contingency plans, and prevent the virus's spread and turn into a global pandemic [154]. Similarly, Lin et al. [155] utilize blockchain with AI to efficiently manage water use under the changing climate conditions, and contribute to climate change adaptation efforts.

AI in the Environment Dimension of Smart Cities
These papers provide insights into 'how AI could contribute to sustainable urban development and improved accessibility in smart cities' [33]. Research in this area focused predominately on the transport, energy, land use and environment sectors/areas, and the contribution of AI can be summarized as: (a) Monitoring changes in the environment; (b) Using smart energy systems to optimize energy consumption and production; (c) Planning, development, and use of households to reduce energy consumption, and; (d) Operationalizing smart transport systems. Appendix A lists the analysis highlights of the reviewed literature.
When faced with complex environmental issues and large quantities of data, AI systems have the potential to make knowledge-based decisions that balance the environmental outcomes of the city against the social and economic wellbeing of its residents [63,136]. AI systems can be used to monitor changes in the environment including, noise, temperature, humidity, emissions [90], water pollutants [133], fish stock, and other environmental indicators [136]. AI systems can respond to these changes, and quickly implement solutions for dealing with any issues [156]. Furthermore, improved data quality from AI systems can contribute to more robust and accurate environmental modelling systems [69,124].
AI has also been identified as a means of creating more energy efficient cities [156]. Smart grid systems, integrated with AI technology, can be used to control power systems and optimize energy consumption [79,78]. Including the planning and management of electric vehicle charging [145], public lighting [75], and data [121]. AI can also assist with the distribution of renewable electricity generated from multiple, often non-traditional sources-including body heat [125]-, the identification of inefficiencies, and future forecasting [134,157]. By optimizing the management of resources, monitoring energy consumption, and better planning for future requirements, cities will be able to use resources more efficiently and better achieve renewable energy goals [80,86].
Smarter homes can be developed with AI systems that monitor changes in the environment, adapt to user requirements, and improve energy efficiency [73,92,116]. AI systems could be used to predict future household energy requirements which can help identify inefficiencies, faults, and control future energy use [81]. In addition, AI can contribute to reduced energy consumption in the construction process [102], and improved environmental outcomes in the design [146] and planning process [26].
With regards to sustainable transportation, the goal of AI in the context of smart cities is to calculate the most efficient means of moving people and goods between places, reducing the number of vehicle kilometers traveled (VKT). This in turn leads to a reduction in energy consumption which in turn leads to lower air and noise pollution, congestion, and other externalities such as the requirements for transportation and parking infrastructure [98,156]. AI can be used for transport optimization by analyzing real-time measurements-such as traffic signal control-to adjust routes [97,74], balance user demands [96], and make parking more efficient [113,104]. Particularly in AVs these changes can result in substantial reductions in travel time and energy savings [108]. From a transport planning perspective AI can also be used to differentiate spatial structures in aerial images [113] and collect masses of data for the development of more accurate, and responsive models which can be used to develop a more environmentally efficient transportation system [117].
Despite no studies in the reviewed literature directly focus on how AI can tackle climate change, we are aware of some relevant research. For instance, in their paper entitled 'tackling climate change with machine learning', Rolnick et al. [158] offers areas where machine learning can be deployed. These areas include better climate predictions and modeling, energy production, CO2 removal, education, solar geoengineering, and finance. Within these areas, the possibilities include more energy-efficient buildings, creating new low-carbon materials, better monitoring of deforestation, and greener transportation. They state that "although AI is not a silver bullet, it brings new insights into the problem". Another study, by O'Gorman & Dwyer [159], demonstrates the use of machine learning to parameterize moist convection and climate change, and extreme event modelling. Likewise, Dayal et al. [160] model Queensland (Australia) droughts based on AI and neural network algorithms for decision-makers and local inhabitants to take precautions.

AI in the Governance Dimension of Smart Cities
How AI can contribute to establishing good governance and planning in smart cities is the focus of this set of papers [33]. Research in this area focused predominately on the security, governance and decision-making areas, and the contributions of AI can be summarized as: (a) Enhancing the operability of surveillance systems; (b) Improving cyber security; (c) Aiding the disaster management planning and operations, and; (d) Assisting citizens with new technology to support citizen scientist and contribute to the urban decision-making process. Appendix A lists the analysis highlights of the reviewed literature.
Advanced AI surveillance technologies, enabled by motion detection, predictive analytics, drones and other autonomous devices, can be used to monitor urban areas, recognize threats, such as crime [72,93,100,148], fraud [109], accidents, and fire [101,123,160]. On a broader scale AI can be used to monitor communication networks and recognize potential terrorist threats, trafficking, crime syndicates, and other illegal behavior [100]. Once targets are identified, intelligent surveillance systems can evaluate and track targets [93], and collect forensic evidence-such as video recording [138]. AI can also be used to better predict future crime incidents and ensure the optimal allocation of crime law enforcement [144].
Cyber threats also pose significant risks to smart cities both in terms data privacy and the protection of connected infrastructure [83,161]. AI can be used to identify irregular behavior, determine what is a threat and implement mitigation measures at speeds beyond that of human ability [83,100]. This together with encryption technologies such as blockchain [100], and a focus on data security at all levels of design [131], can alleviate individual concerns regarding data security and contribute to increased transparency and trust regarding online systems [83,141]. This would allow increased avenues for citizen engagement in policy decisions and citizen scientist engagement with policy development via crowdsourcing [163], along with other online services such as electronic voting [141] and smart contracts [115].
Given the ability for AI to analyze large amounts of data, scenarios to deal with potential threats could be constructed simultaneous to the detection of threats [161]. This would give decision-makers and other authorities more time to respond to threats such as natural disasters [132], house fires [123], or other incidents. Furthermore, AI can be used to assess the extent of damage caused by these events helping authorities better respond to, and mitigate any damage caused [101,132].
Finally, AI systems can be used to both assist and analyze acceptance of new systemsparticularly those associated with new technologies [109,163]. Online 'chatbots' can help residents navigate new websites and online platforms [109], with training and tips customized based on individualized needs and interests [164]. Furthermore, AI use reasoning and intuition to assist decision-makers understand the reasons behind citizen acceptance, or non-acceptance, of new technology [165]. Where new approaches are required AI can develop innovative solutions [95], and address future challenges [84].

Discussion
This review study investigated the impact of the two very powerful and highly popular phenomena of our time-i.e., AI and smart cities. On the one hand, the smart city notion is seen as a potential blueprint for the development of future cities to provide improved productivity, innovation, livability, wellbeing, sustainability, accessibility, and governance, and planning [166,167]. Nevertheless, we still do not have the technical capabilities to develop these technologically advanced futuristic cities [168]. On the other hand, AI provides a hope for overcoming the limits of human capabilities, in the computational sense [169]. Hence, in theory, a happy marriage of AI and the smart cities concepts would bring us closer to produce smarter cities [22].
In this paper, we attempted to generate insights into forming a better understanding on how AI can contribute to the development of smarter cities by undertaking a systematic review of the literature. Appendix A lists the analysis highlights of the reviewed literature. The results of the review disclosed the following main points, and some of the critical issues are discussed further:  AI has an evident potential to provide a positive change in our cities, societies and businesses by promoting a more efficient, effective and sustainable transition/transformation;  AI, with its technology, algorithms and learning capabilities, can be a useful vehicle in automating the problem solving and decision-making processes; that in return could reform urban landscapes, and support the development of smarter cities;  AI in the context of smart cities is an emerging field of research and practice; hence, further research is needed to consolidate the knowledge in the field;  Central focus of the literature is on AI technologies, algorithms, and their current and prospective applications;  AI applications in the context of smart cities mainly concentrate on business efficiency, data analytics, education, energy, environmental sustainability, health, land use, security, transport, and urban management areas, and;  Upcoming disruptions of AI on cities and societies have not been adequately investigated in the literature; thus, further investigations are needed on that issue. The results of the review revealed that AI-inspired computational systems are bound to make a profound impact on our cities. The impact will not only be on the physical setup of our cities but also in how our cities operate and achieve system-level objectives (e.g., livability, resilience, and so on). In order to ensure that cities advance in keeping with the values and aspirations of its key stakeholders (i.e., residents, businesses, and so on), it will be vital for us to ensure that AI systems are designed take on a value-sensitive design approach [170]. AI systems will need to account for the multiple aspects of diversity that are omnipresent in our cities. In addition, these systems will need to possess a degree of transparency, adaptability (to respond to varying environmental conditions), and accountability (for levels of performance). Doing so is non-trivial but is paramount to achieving responsible innovation in the context of AI and cities.
While advances in computational science and technologies will continue to progress at an astounding rate, the level of impact they will have on our future cities comes down to the level of trust individuals and organizations place in these systems. As we continue to live through times where levels of trust in government is at an all-time low [171], planners and public managers needs to consider how does social license [172] impact their ability to deploy emerging technologies. Engaging stakeholders into the design processes when it comes to AI-systems will be critical. Stakeholders should be allowed to shape the elements of these systems and their expected deployment trajectories. Engaging stakeholders will also enable a city to increase the overall knowledge of the community when it comes to the innovative potential of these technologies. To date, we have limited frameworks on how to engage many diverse stakeholders, who have varying knowledge of the intricacies of AI systems, into design processes for urban innovation [173].
We need to enable multiple stakeholders to contribute their technology solutions. Cities need to build platforms that promote the co-creation and sharing of technology solutions [174,175]. While cities have embraced the notion of open data [176] and have created periodic programs to source innovation from external stakeholders (e.g., Hackathons) [177], much more is needed when it comes to designing platforms for co-creation in the context of AI technologies. Stakeholders can be engaged in the auditing of algorithms that underlie AI applications. In addition, they can provide feedback on the performance of these systems, and identify critical choke points. As an example, consider the following innovation by a resident in Berlin, who was able to create traffic jam alerts on Google Maps by slowly moving 99 phones with location services turned on around the city [178]. Residents, such as this individual, enable us to see the limits of AI technologies and their failure points. Such perspectives are critically important as we infuse and design next-generation smart urban technologies. AI systems will impact cities at multiple levels, from the individual, to the local community (the residents), the neighborhood to the organizational (the city), and even the ecosystem (the city is connected to other cities) level. Impacts at the local level, will have effects up and down the hierarchy. Consider the case of algorithms used to promote sharing economy platforms (e.g., Uber, Lyft, AirBnB, and so on). These algorithms not only provide opportunities to individuals to earn rents on their assets and fees for their services, they also impact zoning rules, they impact use of established public transport networks, they in turn impact the creation of new service opportunities for existing businesses, and even shape the nature of public finances of a city. More effort is needed to understand the cascading effects of AI innovations across the various levels of a city's functions. In addition, the interdependencies between functions and the implications for overall objectives (i.e., ensuring that local optimization does not compromise global performance) is also critical.
From a design perspective, research on how to design AI technologies in a more agile [179] and frugal [180,181] manner is of critical importance. The public sector has a notorious record of accomplishment when it comes to managing, and delivering on, information systems projects [182]. Cities around the world have had to contend with failed deployments of information system projects that has wasted significant public resources. The study by Desouza et al. [183] provides examples of both success and failure factors of technology-driven smart city attempts-including AI.
Given the significance of technology investments in our cities, we need to see vast improvements in projects management to deliver on their intended value. In this regard, two considerations are critical. First, is that we build technologies that are agile, i.e., they have the ability to adopt, adjust, and have the capacity for transformation under changing environmental conditions. Second, is that we build technologies in a manner that is in keeping with frugal engineering. Doing so, will require us to move away from mega-scale smart cities projects and reconsider the issue of scale. Today, a dominant design paradigm is to build AI technologies that can scale and promote re-use of components. This thinking is outdated; today it is possible to build technologies that work for specific contexts, in an agile and frugal manner, to promote personalization to a specific context and purpose [184].
The security of our next-generation urban technologies is of paramount importance. AI technologies, like most technologies, should be secured and this normally takes the form of traditional information security. Technologies that traverse our urban environments are already targets of hackers and have vulnerabilities. For instance, Greenberg [185] highlights a deep flaw in cars that lets hackers shut down safety features. But, even beyond what one thinks of when it comes to traditional security, today, AI-driven systems can cause harm even if they are not hacked. For example, as stressed by BBC [186], China coronavirus misinformation spreads online about its origin and scale; this is to say, AI inspired platforms that are used for different urban functions can also be manipulated to spread fake news. Cities need to be aware of this as they use these platforms to share information on urban functions, such as the use of social media to engage with citizens. For instance, as reported by Martinez [187], "Rumors of child abductors spread through WhatsApp in a small town in Mexico…. The rumors were fake, but a mob burned two men to death before anyone checked." Likewise, as we have discussed earlier, cities have had their information systems been held for ransom. The incidents will only increase as cities infuse more technologies into their environments. Hence, there is a need for research to examine the security and risk implications of AI-enabled system deployments.
The success of AI deployment to make our cities smarter will depend on the knowledge and care with which such technologies are deployed responsibility and in keeping with our public values. If done well, AI can help us tackle some our most complex urban challenges. However, it can also make our cities more fragile [54]. As stated by Stephen Hawking on the BBC, "The development of full artificial intelligence could spell the end of the human race…. It would take off on its own, and re-design itself at an ever-increasing rate. Humans, who are limited by slow biological evolution, couldn't compete, and would be superseded." On that very point, Gherhes & Obrad [188] report the findings of their study on technical and humanities student perspectives on the development and sustainability of AI. The study discloses that out of 928 participants 58.3% considers that AI will have a positive influence on the society. On the other hand, the percentage of those confessing to being confused or concerned is also quite significant (41.7%). The probability that AI might destroy humankind and might replace people in certain activities and jobs are among the greatest fears [188].
As the AI applications are becoming more common, there grows the skepticism on the misuse of the technology. For example, most recently the Clearview AI facial recognition system has generated major concerns on the privacy issues. Australian police have been using this unaccountable facial recognition service that combines machine learning and wide-ranging data gathering practices to identify members of the public from online photographs. Stated by Golenfein [189] "Beyond the ethical arguments around facial recognition, Clearview AI reveals Australian law enforcement agencies have such limited technical and organizational accountability that we should be questioning their competency even to evaluate, let alone use, this kind of technology". Similarly, the New South Wales state government of Australia is using AI to spot drivers with mobile phones (often mixed phones with other rectangular items), and the Australian government welfare agency Centrelink is using AI (often incorrectly) debt notices to welfare recipients [190].
Furthermore, scholars warn us from the possible risks of advanced AI. For instance, these risks range from unsafe recommendations for treating illnesses [191] to fatal autonomous car accidents [192] and from racist chatbots [193] to social manipulation [194]. While various dystopian futures have been advanced, including those in which humans eventually become obsolete, with the subsequent extinction of the human race, [195] put forward the following scenarios to think about the ways to protect us from the risks of advanced AI: (a) An AI system tasked with preventing HIV decides to eradicate the problem by killing everybody who carries the disease, or one tasked with curing cancer decides to kill everybody who has any genetic predisposition for it; (b) An autonomous AI military drone decides the only way to guarantee an enemy target is destroyed is to wipe out an entire community, and; (c) An environmentally protective AI decides the only way to slow or reverse climate change is to remove technologies and humans that induce it.
Lastly, abovementioned challenges also relate to the specific characteristics of AI technologies that include opacity ('black box-effect'), complexity, unpredictability and partially autonomous behavior, may make it hard to verify compliance with and may hamper the effective enforcement of rules of existing laws meant to protect fundamental rights [196]. In order to address this issue, the [197] white paper entitled 'Artificial intelligence: a European approach to excellence and trust' underlined the following seven key requirements for a successful AI utilization: (a) Human agency and oversight; (b) Technical robustness and safety; (c) Privacy and data governance; (d) Transparency; (d) Diversity, non-discrimination, and fairness; (e) Societal and environmental wellbeing, and; (f) Accountability. On that very point, Salmon et al. [195] propose the immediate application of the following three sets of controls for AI development and testing: (a) The controls required to ensure AI system designers and developers create safe AI systems; (b) The controls that need to be built into the AIs themselves, such as 'common sense', morals, operating procedures, decision-rules, and so on, and; (c) The controls that need to be added to the broader systems in which AI will operate, such as regulation, codes of practice, standard operating procedures, monitoring systems, and infrastructure. As Elon Musk stated, "we need to regulate AI to combat 'existential threat' before it's too late" [198]. Fortunately, we are not short of ideas and plans to tackle these issues, and now is the time to implement them before it is too late [199].

Conclusions
The study reported in this paper offers a novel contribution to the literature by mapping out the scientific landscape of the understudied 'AI and the smart city' area. This study helps not only in identifying the current and potential contributions of AI to the development of smarter cities-to aid urban policymakers, planners and researchers-, but also in determining the gaps in the literature to bridge them in prospective studies. The study also gives a heads up for urban policymakers, planners and scholars for them to prepare for the disruptions that AI will cause in our cites, societies and businesses [200].
The broad findings of our systematic literature review findings reveal that: (a) AI has an evident potential-but only if utilized responsibly [201]-to provide a positive change in our cities, societies and businesses by promoting a more efficient, effective and sustainable transition/transformation [202,203], and; (b) Particularly, AI, with its technology, algorithms and learning capabilities, can be a useful vehicle in automating the problem solving and decision-making processes; that in return could reform urban landscapes, and support the development of smarter cities [62].
The specific findings of our systematic literature review disclose that: (a) AI in the context of smart cities is an emerging field of research and practice; (b) Central focus of the literature is on AI technologies, algorithms, and their current and prospective applications; (c) AI applications in the context of smart cities mainly concentrate on business efficiency, data analytics, education, energy, environmental sustainability, health, land use, security, transport, and urban management areas; (d) There is limited scholarly research investigating the risks of wider AI utilization, and; (e) Upcoming disruptions of AI on cities and societies have not been adequately investigated in the literature.
AI provides a new hope for addressing some of the urbanization problems we failed to solve due to the complexities involved. Nevertheless, AI is not a silver bullet. While we are currently far away from such advance application of AI, there are numerous contributions of the rapidly developing technology for our cities and societies. Some of these contributions are presented in the paper and some warnings have been made for the good use of the technology. While there is a promise of the emerging advanced technologies, such as AI, our rapid urbanization, industrialization, and globalization practices are perhaps making even technology struggle with coming up solution. The recent anthropogenic climate change triggered environmental catastrophes and disasters-such as 2020 Australian Bushfires-and urbanization and globalization triggered epidemics-such as COVID-19-require more than technology for them not to be repeated again.
The paper opened with a viewpoint on technocentric solutions being widely seen as remedies for global issues-including climate change and urbanization problems. Indeed, AI and other technologies will definitely equip us with better data analytics and prediction models in more efficient and effective ways. To date, there are two different approaches to AI: rules-based (coded algorithms of if-then statements that are basically meant to solve simple problems) and learningbased (diagnoses problems by interacting with the problem), where both AI approaches have valid use cases when it comes to studying the environment and solving climate change. In other words, when it comes to helping solve climate change, a learning-based AI could essentially do more than just crunch CO2 emission numbers, where a learning-based AI could actually record those numbers, study causes and solutions, and then recommend the best solution [204]-'in theory'. We say 'in theory', because "fully functioning AI systems do not yet exist, and it has been estimated that they will be with us anywhere between 2029 and the end of the century" [195].
While we do not disagree with the positive contributions of technological prescriptions-such as AI and other urban technologies- [205], we close the paper with the following quote by Andrew Ng, co-founder and lead of Google Brain. "Much has been written about AI's potential to reflect both the best and the worst of humanity. For example, we have seen AI providing conversation and comfort to the lonely; we have also seen AI engaging in racial discrimination. Yet the biggest harm that AI is likely to do to individuals in the short term is job displacement, as the amount of work we can automate with AI is vastly bigger than before. As leaders, it is incumbent on all of us to make sure we are building a world in which every individual has an opportunity to thrive." To combine IoT with AI in smart machines to simulate intelligent behavior and assist autonomous decision making.
Describes the use of AI to obtain data from IoT to understand acceptance of new technologies.
Governance n/a n/a n/a IoT ICT Chau [69] A review on integration of artificial intelligence into water quality modelling

Marine Pollution Bulletin
To reviewing the current state-ofthe-art AI and its application in water quality modelling. To advocate using AI to solve urban problems. Reflects, critically, on the optimistic viewpoint of AI in relation to its potential to respond to urban problems (e.g. congestion, population growth, energy efficiency, environmental degradation and safety).
Environment n/a n/a n/a n/a To discuss the importance and consequences of smart city development.
Identifies the opportunity for AI and smart cities to achieve better security measures Governance n/a n/a n/a n/a To explore the use of AI in education Identifies the need for education to be adaptive in the face of rapid technology advances, and changes to employment.
Society n/a n/a n/a n/a Pieters [141] Explanation and trust Ethics and Information Technology To investigate the relationship between explanation and trust in the context of AI Describes the importance on online security for trust in AI systems.
Governance LB KB ES n/a n/a Ponce & Gutiérrez [124] An indoor predicting climate conditions approach using internet-of-things and artificial hydrocarbon networks

Measurement
To predict the temperature of remote locations using field sensors and information from network. To improve the precision of small area population forecasting. Describes the use of AI in population forecasting Environment n/a n/a n/a n/a