AS AN INCREASING NUMBER OF PEOPLE MIGRATE TO CITIES, WILL AI'S ADOPTION EMPOWER CITIES TO PROTECT ITS INHABITANTS FROM THE RAMIFICATIONS OF CLIMATE CHANGE?
[Student IDEAS] by Seethal Reddy Kauluri - Master in Management at ESSEC Business School
This article will look into the definition of the attention economy, what it entails, how it affects society at large, and why everyone should care; particularly in this new age of exponential and wide-use of artificial intelligence. Moreover, it will guide a possible way forward for citizens to deal with the attention economy, and underline the sway that users have over the tech services that govern our daily lives.
“ AI is the new electricity.” - Andrew NG, co-founder of Coursera and Google Brain Deep Learning Project.
From bushfires in Australia, floods in India, droughts in Africa, to dry corridors in Central Asia, we have seen some of the deadliest and costliest climate disasters of history in the recent past. They won’t be the last of such disasters to occur if our efforts to fight climate change remain scant as today. According to the United Nations, the global emissions were supposed to be reduced by 7.6% per year between 2020 and 2030 in order to limit global warming below 2 degrees Celsius (3.6 degrees Fahrenheit) set at the Paris Agreement. However, global emissions increased in 2021 after a dip by 5% in 2020 due to the COVID-19 pandemic.
Cities contribute to over 70% of these global CO2 emissions due to massive industrialisation and motorised transportation systems that still use fossil fuels. They are also the most vulnerable to these disasters as they are home to more than half the world’s population today and three quarters of the population or 6.5 billion people by 2050. Asia and Africa, where the majority of the population still resides in rural areas, increasingly face the stress of rapid urbanisation, while uplifting living standards at the same time. According to Mckinsey Global Institute's report, Asia will starve due to increasing droughts, and India may lose up to 30 percent of annual daylight hours by 2050 due to rising temperatures.
As we desperately hunt for solutions to protect our cities in the wake of this emergency, this article critically compares and assesses the potential of AI as a solution as it is increasingly revered today in various applications due to its abilities to perform human-like tasks faster, with minimum errors, and at a much grander scale than was possible before.
Are smart cities enough?
To overcome the challenge of rapid infrastructural development while improving quality of living, many African and Asian countries are looking to build smart cities: to integrate advanced technologies to manage overcrowding and traffic, and to manage resources better. Smart cities also seem to be a way of attracting foreign investment as they are seen as an opportunity to expand markets, promote innovative products and services, tap into emerging hubs for skilled workforce and gain trade benefits. While there is no universally accepted definition of a smart city, in a broader sense, it has been defined by multiple sources as a city that leverages smart technology and communication, such as, data analytics, cloud computing, Internet-of-Things (IoT), or artificial intelligence to achieve operational efficiencies for the welfare of the citizens. But this definition doesn’t explicitly state that smart cities are necessarily built to be resistant to the adversities of climate change.
According to the Intergovernmental Panel on Climate Change, Climate Resilient Development is urgently needed. It defines this framework as, “reducing exposure and vulnerability to climate hazards, cutting back greenhouse gas emissions and conserving biodiversity are given the highest priorities in everyday decision-making and policies on all aspects of society including energy, industry, health, water, food, urban development, housing and transport.” Therefore, as the global smart cities market is expected to grow to the value of about $1.38 trillion by 2030, there is an increasing need to refocus this investment on building smart cities that are, in fact, resilient.
Before proposing solutions, let us first analyse some popular non-AI experimental eco-settlements and assess whether inspirations can be drawn from their styles of governance, economy, and self-sufficiency in achieving climate-resilience. One such settlement is Auroville in Viluppuram district of India, which was founded in 1968 as a community in which people from all over the world could come together and live in harmony, transcending the barriers of nationality, race, and religion. The town is resilient to climate as the inhabitants are committed to sustainable living while reducing environmental impact by adopting methods such as using renewable energy, organic farming, and implementing water conservation and waste management systems. Auroville also places a strong emphasis on educational programs that help build the knowledge and skills needed to adapt to the impacts of climate change. It is home to several research centres and innovative projects that focus on sustainability and climate resilience.
Another utopian settlement project is NEOM, which is part of Saudi Arabia’s vision 2030 that aims to offer its citizens the best of diverse climates - from sun-soaked beaches, floating industrial islands on Red sea, to ski resorts in snow capped mountains, all while running on 100% renewable energy. The region, an area larger than Israel, aims to eliminate cars and roads by replacing them with a hyper-speed rail that covers the entire Line - the shape in which the city will be built for a length of about 177 kms. Each block of this city aims to be self-sufficient by building schools, groceries and other basic amenities at a walkable distance to cut emissions. The project also claims to care for biodiversity by building a coral-reef restoration project.
However successful or utopian these experiments might sound, it is overly ambitious to adopt their styles on a large scale. Auroville is a close-knit community with a number of residents just over 3000, although the plan was initially to attract tens of thousands more residents. This sense of unity helps build social resilience, which is essential for climate resilience but cannot be replicated at a global scale due to bureaucracy and disharmony between national and local governments. This is also why NEOM, which aims to host up to 9 million residents by 2045 (only 0.09% of the world population by then), aims to have an independent governance system with laws of its own, which not only substantiates that it is easier to control climate policies in a smaller sample, but also establishes that it is impossible to achieve these climate and self-sufficiency objectives while remaining under the law and economy of its host country that still derived 68% of its budget revenues from oil in 2019. Another challenge of scaling these models is the differences in global diversity. Even though Auroville brings together diverse people from all over the world, they are still connected through a common sense of spirituality and respect for nature. These individuals share values of collectivism and insist on learning a common language, Tamil, to facilitate cooperation and smooth community functions. Finally, such models depend on local resources for self-sufficiency. While Auroville depends on its organic farms and renewable projects for food, energy and water, Neom aims to build a close-knit self-sufficient economic hub by attracting world-wide talent from 14 industrial sectors in creating a concentration of resources and economic power that makes it far easier to achieve resilience here than at a macro-level. Analysts also suggest that projects like Neom are more a delay towards climate objectives than a solution as living in Saudi Arabia necessitates the use of air conditioning, which is still challenging to shift into renewables.
Therefore, as the sustainable models implemented in the cities discussed above cannot be replicated on a larger scale, we must turn towards more pragmatic and viable solutions to improve our cities’ climate resilience.
The Promise of AI
According to a survey conducted by BCG, AI can expedite reduction of GHG emissions up to 10% per organisation and up to 5.3 billion tons of CO2e if scaled globally. This is because of its enormous power of gathering, computing, and analysing unstructured, multidimensional, and end-to-end data points. As cities in Asia and Africa suffer from inefficient city designs due to informal settlements and inadequate public transportation systems, these regions can radically transform into sustainable and resilient cities by leveraging AI’s power in 360-degree information modelling to optimise energy, reduce wastage, and improve accuracy and speed of climate-modelling for better mitigation and preparedness in the short-term and adaptation and resilience in the long-term.
In buildings, energy can be optimised by autonomous lighting, heating, and cooling based on occupancy levels, time of the day, weather conditions and occupants’ comfort. AI can be used to track carbon foot-print, water usage, and waste production in order to set consumption limits. AI can spot anomalies through data gathered from sensors and check performances of equipment outside the norm for timely maintenance and repairs. AI can predict energy requirements to ensure ample solar panels and storage are installed for uninterrupted renewable energy consumption. Comparative data analysis from other green-buildings, or with simulations based on AI-built “digital twins”, can be used to prescribe alternative architectural designs in comparable areas to further reduce emissions if required. For example, it can be prescribed to use tinted windows or green roofs to keep the buildings cooler if the major carbon emitter for a particular building is air conditioner. AI can cluster the occupants based on consumption behaviours and emissions for law-makers to either encourage favourable behaviours through reward systems or through penalties.
Similarly in transport, AI can optimise train and bus routes and frequencies by matching demand and supply based on time of day, season, weather, etc. It can be used to predict repairs and maintenance in advance to prepare for downtime and ensure that public transportation is always available and running smoothly without interruptions. It can also be used to propose on-demand services or reduce off-peak inefficiencies caused by vehicles running almost empty because of their fixed itineraries and schedules. Personal vehicles can be replaced by car-pooling that matches passengers based on pickup and destination. AI can also analyse data about walking, running, and cycling activity to make recommendations. For instance, it can suggest the best locations for basic neighbourhood amenities such as groceries, coffee shops, pharmacies, schools and emergency clinics, to ensure widespread access that does not rely on fossil fuels. It is also possible to centralise transport systems to optimise routes and traffic. A city’s transportation system can be entirely integrated on a 360-degree carbon monitoring to ensure that collectively, the city’s transport stays within the carbon budget. If at a particular time, personal vehicles on road cause more emissions, reinforcement learning can be used to encourage citizens to use more of public transport by increasing frequency of trains and metro while increasing time of red traffic signals on the road to discourage vehicle traffic or converting roads into carbon-free zones to promote cycle and pedestrian-traffic only.
AI can study vast data from weather sensors, satellites, news reports, and social media, and spot hidden patterns and events to predict risks, such as temperatures, sea-levels, droughts, rainfalls, heat-waves, hurricanes etc. Predicting these events can help prepare the necessary infrastructure for minimising the consequences, including utilities’ networks (water, gas, electricity, etc.), critical to the proper functioning of a city. These patterns and clustering can also help predict the degree of impact for each region based on population density, industrial activity, infrastructure, biodiversity, etc and compute the degree of rescue services and relief supplies required for better mitigation. An integrated biodiversity system can help monitor species, plantations and conditions of habitats to spot patterns of endangerment, deforestation and environmental factors affecting the weather patterns.
Thus, AI’s power of integration systems from macro to micro-level emission monitoring and hazard prediction can help build a harmonised source of truth and scale what smaller experiments like Auroville or NEOM have on a much larger, global scale.
The aforementioned ideas may sound like the answer to our problem, but there is resistance in the adoption of AI due to several reasons. Firstly, the data is widely scattered and dissociated as there is a lack of centralised data source. This is because the climate objectives are too broad and the firms and law-makers haven’t adequately defined the KPIs to be tracked for each sector. To establish the same could be challenging as a large infrastructural overhaul will be required to substitute independent systems with interconnected, autonomous and accountable systems and appliances which might be highly infeasible depending on the scope of project and the cost of sensors, storage, computing resources, and personnel required to implement the models. Therefore, it is the duty of the law-makers and enforcers to break down the objectives to smaller key results and introduce policies for infrastructure, construction and automobile industries to mandate accountable data collection points in their products. Secondly, there is lack of education and training on AI expertise among organisational leaders and governing bodies to realise its vast potential. Free online and offline educational programs, workshops, and certifications shall be set up from schools to organisational leaders and policy-makers to equip them with the requisite knowledge. States should also promote research centres, events and competitions in the field of green-AI to encourage new and innovative ideas. Another challenge is that setting up large-scale and centralised AI integrated systems requires huge financial investments which is challenging for countries in Asia and Africa due to the pressure of mitigating fundamental gaps in the economy and infrastructure. According to a report by OpenAI, it is estimated that the cost of training a model on a large dataset will increase to $500 million by 2030 due to an increase in the size of datasets needed to train large models. However, setting up this investment right away will ensure resilience, cost-efficiency and a significant gain in the GDP of up to 4.4% by 2030. But even if these governments were convinced to step up the financing for this preparation today, a larger issue regarding AI adoption would remain: can we trust the results that AI provides?
Training Green-AI Models
One of the reasons why AI-based recommendations are not always relied upon is due to the quality of their design and model training. Human intervention in feeding, labelling and training data while the model is built can give room for bias and skewed results. For example, if the AI-based traffic system mentioned earlier aims to manage demand and supply to optimise bus routes in a city, such a model will work inefficiently if it has been trained on historical data of a city’s bus routes where certain high-income areas never had a bus-stop because they used cars. Such a model will have to be trained on collecting data from all modes of transport to understand commuting patterns. New experimental car-pooling and bus routes will have to be set up initially to collect data on public transport preferences of the citizens. Another example is if a model has studied smartphone data of a region to understand walking and cycling patterns to plan roads and accessible neighbourhood amenities such as groceries and pharmacies. Such a model will work inadequately if the data collected was from an area of ageing population with minimal smartphone adoption. Therefore, such models will have to be trained by nudging some preferred behaviours first, and may rely on human modelling assumptions (such as that which was used in the 1970s to predict train demand for the newly created BART system in San Francisco, and for which Dan McFadden received the 2020 Nobel Prize in Economics).
An additional concern is regarding the confidentiality of the information captured by the AI systems that can compromise privacy and security. The accuracy of these models will depend on the extent to which the inhabitants are willing to share their information. The law-makers need to carefully consider the type of information to be mandatorily collected from the citizens, and ensure that it does not compromise their freedom of choice. Additionally, will highly autonomous systems be capable of considering externalities? Or new patterns? Indeed, what if green-buildings’ AI integrated systems were trained on pre-pandemic occupancy levels, but when another pandemic, like COVID, hits the region causing almost everyone to work from home? Will such AI systems accept the extra-bandwidth of energy needed to run the residential buildings? It would be necessary that such a system should also train the carbon budget at the macro-level and maintain a global balance of buildings’ emissions, by temporarily transferring the carbon budget of office buildings to that of the residential buildings. This brings us back to the general limitations of AI that when faced with adversities, especially in situations when no historic data or patterns exist, a human intervention is always required. Then the real question remains for us to ask ourselves, whether undergoing such an ambitious infrastructural project is justified given the human efforts required despite the massive capital expenditure and with the uncertainties of the risks present in adopting it.
Given the numerous applications of AI in mitigating climate risks and achieving climate-resiliency, overlooking such an immense potential will be a significant missed opportunity. However, due to the challenges present in adopting a globally-scaled autonomous infrastructure and the accuracy of its decisions, AI is only an accelerator towards achieving our climate objectives and not the solution itself. Just as it is easy to coordinate smaller-sized homogenous experimental cities, implementing AI for climate should be scaled step-by-step starting with the most vulnerable sectors by carefully monitoring the results and reimplementing models in similar and comparable conditions.
Achieving climate-objectives goes beyond adoption of AI. International organisations, governments, urbanists and corporations have a fundamental role in adopting a harmonious framework by breaking down climate objectives into smaller-measurable targets for each region, industry and sector. They play an important role in creating a clear path for investment and innovation. Governments should pave the way for adoption of new technologies that will not only make our future cities smart, but future-proof them. Policies in education, research and business will play a defining role in the future of AI and in its adoption for a green, stress-free future of habitation.
 10 costliest climate disasters of 2022. (2023). World Economic Forum.
 Cut global emissions by 7.6 percent every year for next decade to meet 1.5°C Paris target. (2019). UN Environment Programme
 Dasgupta, S., Lall, S., Wheeler, D. Cutting global carbon emissions: where do cities stand? (2022). World Bank
 Woetzel, J., Tonby, O., Krishnan, M., Yamada, Y., Sengupta, S., Pinner, D., Fakhrutdinov, R., and Watanabe, T. Climate risk and response in Asia. (2020). McKinsey Global Institute.
 What is Climate Resilient Development and how do we pursue it? Intergovernmental Panel on Climate Change report.
 Thormundsson, B. Smart cities market revenues worldwide 2019-2030. (2023)
 Koduvayur Venkitaraman, A., Joshi, N. A critical examination of a community-led ecovillage initiative: a case of Auroville, India. Clim Action 1, 15. (2022).
 Palmer, I,. Is Saudi Arabia’s New Climate City ‘Neom’ Future Or Fantasy? (2022) Forbes.
 Almasoud, A., Albasri, K., Alshammari, M., Alghamdi, R., Alanazi, A., Binladen, S., Alsikhan, S., Kingdom of Saudi Arabia Budget Report. (2018). KPMG.
 Maher, H., Meinecke, H., Gromier, D., Garcia-Novelli, M., Fortmann, R. AI Is Essential for Solving the Climate Crisis. (2022). BCG.
 AI Model Training Costs Are Expected to Rise From $100 Million to $500 Million By 2030. (2023). MetaversePost.
 Herweijer, C., Combes, B., Gillham, J. Artificial intelligence and the fate of planet Earth. (2019). PWC.
 Cowls, J., Tsamados, A., Taddeo, M. Floridi, L. The AI gambit: leveraging artificial intelligence to combat climate change—opportunities, challenges, and recommendations. (2021). AI & Soc.
 McFadden, D. The Measurement of Urban Travel Demand. (1974). Journal of Public Economics, 3, 303-328.