Abstract
This publication represents the second of a cycle of analyses exploring the geopolitical consequences of Artificial intelligence (AI). AI continues to advance affecting our lives, the environment and shaping the geopolitical landscape in profound ways. This article will assess the impacts of AI on the environment, focusing on the regulations applied as of today to allow a sustainable development of new AI technologies and some of the fields in which AI is used to improve environmental protection.
Authors:
Gabriele Silini, Junior Researcher - Politics, Matilde Pierattini, Senior Researcher - Environment
Introduction
In the latest decades, Artificial Intelligence (AI) has always been praised for its transformative potential, which has the potential to reduce the impact we are having on the environment (Nordgren, 2023). AI has become part of our everyday lives and it is used to tackle several issues in fields ranging from agriculture to urbanisation.
AI plays an ambivalent role when it comes to its impact on the environment. On the one hand, it can indeed aid in mitigating the effects of the climate crisis through the development of low-emission infrastructure, intelligent electrical system designs, and climate change prediction modelling. On the other hand, it must also be taken into consideration that artificial intelligence emits a considerable amount of carbon itself, which is destined to dramatically grow as AI becomes more complex (Dhar, 2020).
However, AI has already been used to enhance environmental protection practices. Indeed, UNEP is promoting the Digital Transformation sub-programme. The aim is to exploit data and digital technologies to finally reach global sustainability. UNEP has launched many collaborations and action plans for monitoring air and water quality through geospatial statistics, previous data collection and analysis, national database and other tools. Moreover, AI is the primary tool for the development of smart cities based on digital urbanisation. In many cases, AI showed to have a positive impact on the environment.
AI and its use of energy
The amount of energy required to train and execute AI models, for example, increases dramatically with the complexity of the solution the AI provides. The rise in energy use exacerbates climate change, as this has a direct impact on greenhouse gas emissions. On this matter, studies claim that since 2012 the power needed to produce and train AI models has doubled every 3.4 months.In addition, the information and communications technology (ICT) sector is expected to reach 14% of worldwide emissions by 2040. These emissions are estimated to originate from ICT infrastructure, mainly data centres and communication networks. These findings highlight how crucial it is to reduce AI's carbon footprint and contribution to environmental degradation (Nordgren, 2023).
Many AI methods, in particular deep learning, typically demand large amounts of data. For example, it is necessary to collect and then store and process the data to allow an AI to work and these actions have an effect on the environment as they require energy (Bolte et al., 2022).
At the same time, AI is becoming more and more popular, and it is frequently advertised as an alternative to solve, or at least mitigate, environmental issues. Although the negative impact on the environment is usually briefly mentioned, including effects where unitary efficiency gains might result in a rise in global greenhouse gas emissions, no quantification of all the environmental costs associated with AI is usually offered in order to close the gap between AI for Green and Green AI. It is therefore more important than ever to be able to evaluate the actual implications, accounting for both favourable and unfavourable outcomes (Bolte et al., 2022).
AI and the lack of a unified approach to address its environmental impact
As mentioned, the training phase of AI is particularly alarming. Strubell et al. (Strubell, Ganesh and Mccallum, 2019) from the University of Massachusetts have studied the energy considerations for Deep Learning in Natural Language Processing (NLP), noticing that the amount of greenhouses gases (GHG) emissions used for NLP is equal to 300 flights between San Francisco and New York. Although AI is currently one of the main drivers of innovation, it still lacks a unified approach to assess its environmental impact. Quantifying the effects of digital technologies and, in particular, artificial intelligence presents a number of methodological decisions that demonstrate a lack of a clear conceptualisation of the system. For example, evaluating the effects of an AI algorithm or service is not the same as evaluating the worldwide consequences of the AI domain.
The growing topic of measuring AI's effects is currently hampered by a lack of standard methods, and it frequently concentrates solely on the implementation phase of the devices that are part of an AI service. Another weak point when it comes to AI is the lack of regulations, in particular environmental ones, resulting in the absence of an univocal way to define the effects of AI on the environment. Moreover, it is also worth mentioning that AI technology is causing a growing environmental issue, as it generates e-waste. Materials like lead, mercury, and cadmium are just a few of the dangerous substances found in e-waste that can contaminate soil and water.
As already analysed, the training phase of AI generates a considerable amount of CO2 emissions; however, this is only one side of the problem. The carbon footprint of the infrastructure supporting big tech's AI deployment requires additional consideration. Tech workers, for instance, urged their employers to take into account their role in climate change during the worldwide climate strike in September 2019. The main aim during this climate strike was to draw attention to big tech's partnership with fossil fuel companies and its role in the repression of climate refugees and frontline communities. As part of the Tech Workers Coalition, workers from Amazon, Google, Microsoft, Facebook, Twitter, and other companies marched to demand that their employers commit to reducing emissions to zero by 2030, avoiding renewing their contracts with fossil fuel companies (Dhar, 2020).
Although national AI initiatives are now slowly starting to take into account the environmental impact of AI, it is still not considered as a common practice. Given their energy and water requirements, countries are beginning to inquire more and more about the environmental effects of running large-scale AI models. However, there is still no standardised indicator to provide a guide to allow sustainability decisions for both public and private sector investments in AI (OECD, 2022).
The lack of accepted standards, standardised indicators and metrics, uniform reporting criteria, and common terminology make it difficult to measure the environmental effects of AI computing and applications. The lexicon employed to depict the ecological consequences of artificial intelligence is predominantly diverse: expressions such as “green AI,” “computational sustainability,” or “sustainable AI” are frequently employed interchangeably, and diverse entities delineate environmental implications in disparate circumstances. International standard-setting organisations, such as the UN and the EU, as well as NGOs could create a comprehensive framework that would allow for the benchmarking, comparability, and compatibility of national computing initiatives and their effects on the environment, including artificial intelligence.
The first international regulation on AI: the EU AI Act
The European Union is the first actor working on an AI regulation with the EU AI Act. The main aim of the act is to have safe, transparent, non-discriminatory and environmentally friendly AI systems in the EU (European Parliament, 2023b). In June 2023 the EU disclosed the first priority rules to negotiate on AI which also considered the impact on the environment. However, the Amendment proposals for a regulation only take into consideration the potential of AI in mitigating the effects on the environment, without considering the potential threat that AI can be if there is not a unified system to calculate the CO2 emissions of AI:
«Artificial intelligence is a fast-evolving family of technologies that can and already contributes to a wide array of economic, environmental and societal benefits across the entire spectrum of industries and social activities if developed in accordance with relevant general principles in line with the Charter and the values on which the Union is founded. By improving prediction, optimising operations and resource allocation, and personalising digital solutions available for individuals and organisations, the use of artificial intelligence can provide key competitive advantages to companies and support socially and environmentally beneficial outcomes, for example in healthcare, farming, food safety, education and training, media, sports, culture, infrastructure management, energy, transport and logistics, crisis management, public services, security, justice, resource and energy efficiency, environmental monitoring, the conservation and restoration of biodiversity and ecosystems and climate change mitigation and adaptation (European Parliament, 2023a).»
It is worth mentioning that after the first report of the EU AI Act was published in June 2023, Germany, France and Italy have shared their scepticism on the matter. In particular, the European Parliament proposed to apply a code of conduct initially only to major AI servers, which are mainly located in the US. However, in this way, as the three States argued, there will be a drawback in the long run for European providers, as EU citizens would trust more large providers, as they are the ones respecting the EU Act, resulting in less customers and hindering the possibility for EU providers to expand in the EU market (Euronews, 2023).
Despite the EU AI Act being under discussion, it seems clear that the Act intends to tackle the impact of AI systems on people's health, safety and fundamental rights. As such, AI impact on the environment has only been mentioned related to human health and it does not, at the time of writing, represent a topic per se within the priorities of the EU AI Act.
AI as a tool for environmental protection
The definition Mr Jensen, coordinator of UNEP, proposed for the AI is “a system or machine that performs tasks that typically require human intelligence, and can iteratively improve themselves over time, based on the information they collect”. The areas where AI can be an effective tool to protect our environment are four: agriculture, conservation of resources, waste management and pollution monitoring and treatment. Nowadays, Machine Learning systems (ML) are used to solve environmental challenges. (Nti, Cobbina , Attafuah, Opoku, & Gyan, 2022)
A first example of AI deployed to solve environmental issues is the case of the World Environment Situation Room (WESR). WESR is the UN Environment Programme (UNEP)’s online data library dedicated to the world environment. The idea behind the realisation of WESR was to create One Global Partnership among physical situation rooms for monitoring the environment. These rooms have been organised to cover every region all over the world. Obviously, this project could not rely just on these rooms. One Global Partnership is supported by the Global Resource Information Database Network (GRID) and other types of international organisations. (United Nation Environmental Programme, 2021) The data available on WESR is directed to policy makers, international organisations, the scientific community, business companies and world citizens.
The project was created to satisfy the ambition of the international community, which desired to foster cooperation on the global environmental protection issue. Since the 1970s, the international community realised that not every country can tackle the enormous climate change issue exclusively focusing on its own environmental challenges. Thus, nowadays the WESR is concretely based on the “Acting as One” through tools such as geospatial statistics, Sustainable Development Goals (SDGs) statistics, Multilateral Environmental Agreements (MEAs), global monitoring systems, assessments, citizen science and private data, foresight analysis. (United Nations Environmental Programme, 2021)
WESR is an innovative and practical instrument to comprehend, in nearly real-time, how our global community is dealing with the climate change crisis. The creation of such a library, available to anyone, is expanding the access to data about specific topics to any world citizen who is able to surf the internet. To become familiar with the climate change problems, WESR has been structured in four main research options. Every internet user can easily access data on environmental changes by choosing a specific topic, a specific SDG, a specific region and country or choosing a story map.
An example of information obtainable with WESR is the existence of TAHMO. The Trans-African Hydro-Meteorological Observatory (TAHMO) is a network of 20.000 hydro-meteorological stations in the Sub-Saharan African region. TAHMO is fundamental for understanding how the Sub-Saharan region could face the obstacles of forecasting food production in a sustainable way in the future when atmospheric conditions will be even more unpredictable. In a region where national governments struggle to have proper data on water and air quality, TAHMO could be a huge source for the agricultural sector. However, the weather stations do not completely cover all the African continent, especially the central part. This circumstance leads to a lack of a proper improvement in monitoring African climate. (TAHMO, s.d.) Nevertheless, using AI to improve environmental protection is still a work in progress. As with every scientific innovation, TAHMO can always change in the future.
AI for monitoring Methane emissions
In 2021, UNEP launched the International Methane Emissions Observatory (IMEO), whose purpose is to provide data on Methane emissions mainly to researchers, organisations, governments and companies. Before creating IMEO, many organisations such as the Global Methane Initiative (GMI) and the International Energy Agency (IEA) have started to focus themselves on the Methane emissions issue. To impede users from providing approximate data on their Methane emissions, IMEO has set transparency standards (UN environment programme, s.d.) hoping that this action leads to an effective database. Methane is 80 times more potent than carbon dioxide and has accounted for 30% of global warming since pre-industrial times. (European Commission, 2021) This is why the observatory was created: to collect reliable data through the Oil & Gas Methane Partnership, the Science measurements studies, the satellite data and national database.
AI for smart cities
As renewable energy is becoming more and more important to tackle the global climate crisis, it must be considered the impact of AI in the energy sector. AI can be extremely useful to manage renewable energy systems. As a tool aimed to monitor and forecast new phenomena by analysing previous data, AI can be essential to maximise solar energy production. As the quantity of solar energy is based on the inclination of solar panels towards the sun and how much heat its rays release, AI application to solar energy systems can considerably change solar panels productivity. On one hand, a machine learning system connected to a solar panel can collect data to analyse, monitor and forecast the future productivity through its integrated sensors. On the other hand, AI can optimise the energy storage by changing its use when the solar panels cannot directly produce solar energy from the sun. (Mohammad & Mahjabeen, 2023)
Adding AI to solar energy systems could finally boost the development of smart cities. As the European Commission states, “a smart city is a place where traditional networks and services are made more efficient with the use of digital solutions for the benefit of its inhabitants and business.” (European Commission, s.d.). One of the concepts behind the development of smart cities is the sustainable development of urban spaces. Among all the new approaches the creation of a smart city can introduce, the decarbonization of buildings is considered to be one of the best ways to effectively reduce carbon emissions. Green buildings are buildings with lower environmental impact and lower maintenance costs due to a resource-efficient method of construction. (Green Building, s.d.) Moreover, in a smart city AI sensors could be used for the waste management improving the recycling system, thus reducing the environmental impact. AI in smart cities could be used also for governance, to improve citizens’ security, to boost the economy by efficiently reallocating resources and improving infrastructure.
Dismantling the traditional agriculture to tackle future challenges
AI could also boost a recent upscaling phenomenon: the increase of vertical farming. In the last few years, scientists affirmed that soil and water are more frequently diminishing. Vertical farming has spreaded as a new idea of agriculture that helps to tackle the future environmental challenges. AI offers digital assistance to the farmer by monitoring elements such as the water use and temperature, and carrying out actions such as the automatic irrigation. An example is Clovy, the vertical farming created by the startup named Hexagro. The Italian startup proposed a revolutionary way to save soil, water and space to cope with the climate change consequences. (Garancini, 2023)
Producing our own agricultural products significantly reduces emissions. Independently farming, consequently reducing the use of vehicles to reach markets or supermarkets, decreases the shopping of industrial products whose supply chain is one of the biggest responsible of the global GHG emissions.
Conclusion
In conclusion, AI has indeed the potential to mitigate the impacts on the environment; however, the lack of regulations and an internationally accepted system to calculate the impacts of AI solutions on the environment are hindering this opportunity. In doing so, AI is dramatically growing as a pivotal player for innovation without fulfilling the promise to mitigate the effects of climate change.
The article also focused on the first discussions of the EU AI Act, stressing how important it is to create binding regulations, and hopefully a common model in the EU to assess AI impact on humanity. Although the discussions of a EU AI Act have just started in June 2023, the AI impact on the environment has not yet been tackled. Nonetheless, it is undeniable that AI impact on the environment is present and strong and, thus, the EU has the responsibility to include the topic within the priorities of the upcoming EU AI Act.
It must be indeed noticed that AI has proven to be a powerful tool when it comes to tackling environmental issues, such as the future shortage of cultivable space or potable water. As scientists continuously state, there is no time left, renewable energies must flourish and fossil fuels must be banned. Although the use of AI can be considered extremely polluting and counter-productive, AI could be a valid tool to dismantle the traditional methods of producing agricultural goods, avoiding the use of the already scarce essential raw materials. If widespread, AI could highly reduce human’s carbon emissions, reducing human's impact on the environment.
The present article analysed some of the many implications of AI on humanity; in the next publication of this cycle, the geopolitical implications of AI on Economic Competition and Finance will be analysed.
Classification of sources and information:
1 | Confirmed | Confirmed by other independent sources; logical in itself; coherent with other information on the topic |
2 | Presumably true | Not confirmed; logical in itself; coherent with other information on the topic |
3 | Maybe true | Not confirmed; reasonably logical in itself; coherent with some other information on the topic |
4 | Uncertain | Not confirmed; possible but not logical in itself; no other information on the topic |
5 | Improbable | Not confirmed; not logical in itself; contradicts with other information on the topic |
6 | Not able to be evaluated | No basis to evaluate the validity of the information |
Trustworthiness of the source | ||
A | Trustworthy | No doubt about authenticity, reliability or competence; has a history of total trustworthiness |
B | Normally trustworthy | Small doubts about authenticity, reliability or competence, nevertheless has a history of valid information in a majority of cases |
C | Sufficiently trustworthy | Doubts about authenticity, reliability or competence; however, has supplied valid information in the past |
D | Normally not trustworthy | Significant doubt about authenticity, reliability or competence, however has supplied valid information in the past |
E | Not trustworthy | Lack of authenticity, reliability or competence; history of invalid information |
F | Not able to be evaluated | No basis to evaluate the validity of the information |
References
Bolte, L. et al. (2022) ‘Unraveling the Hidden Environmental Impacts of AI Solutions for Environment Life Cycle Assessment of AI Solutions’, Sustainability 2022, Vol. 14, Page 5172, 14(9), p. 5172. doi: 10.3390/SU14095172. 1B
Building, G. (s.d.). What is green building? Tratto da https://greenbuildingcanada.ca/green-building-guide/what-is-green-building/. 2A
Commision, E. (s.d.). SMart cities. Tratto da https://commission.europa.eu/eu-regional-and-urban-development/topics/cities-and-urban-development/city-initiatives/smart-cities_en. 1A
Commission, E. (2021, 10 31). International Methan Emissions Observatory launched to boost action on powerfyl climate-warming gas. Tratto da https://ec.europa.eu/commission/presscorner/detail/es/ip_21_5636. 1A
Dhar, P. (2020) ‘The carbon impact of artificial intelligence’, Nature Machine Intelligence, 2(8), pp. 423–425. doi: 10.1038/S42256-020-0219-9. 2B
Euronews (2023) Disagreements from key EU member states, including France, on how artificial intelligence (AI) should be regulated threatened to derail EU legislation. Available at: EU AI Act: Germany, France and Italy reach agreement on the future of AI regulation in Europe | Euronews. 2A
European Parliament (2023a) Artificial Intelligence Act. Available at: https://www.europarl.europa.eu... (Accessed: 15 November 2023). 1A
European Parliament (2023b) EU AI Act: first regulation on artificial intelligence. Available at: https://www.europarl.europa.eu... (Accessed: 15 November 2023). 1A
Garancini, C. (2023, 09 25). Clovy, l’ultima generazione di orti verticali intelligenti per coltivare in città (e non solo). Tratto da Lifegate: https://www.lifegate.it/clovy-hexagro-agricoltura-verticale-urbana-crowdfunding. 2B
Mohammad, A., & Mahjabeen, F. (2023). Revolutionizing Solar Energy: The Impact of Artificial Intelligence on Photovoltaic Systems. International Journal of Multidisciplinary Science and Arts, 2(1). doi:https://doi.org/10.47709/ijmdsa.vxix.xxxx. 2A
Nti, E. K., Cobbina , S., Attafuah, E., Opoku, E., & Gyan, M. (2022). Environmental sustainability technologies in biodiversity, energy, transportation and water management using artificial intelligence: A systematic review. Sustainbale Futures, 4. 2A
Nordgren, A. (2023) ‘Artificial intelligence and climate change: ethical issues’, Journal of Information, Communication and Ethics in Society, 21(1), pp. 1–15. doi: 10.1108/JICES-11-2021-0106/FULL/HTML. 2A
OECD (2022), "Measuring the environmental impacts of artificial intelligence compute and applications: The AI footprint", OECD Digital Economy Papers, No. 341, OECD Publishing, Paris, https://doi.org/10.1787/7babf571-en. 1A
Programme, U. N. e. (s.d.). ABout IMEO. Tratto da https://www.unep.org/explore-topics/energy/what-we-do/methane/about-imeo. 1A
Programme, U. N. (2021). World Environment Situation Room – Work Plan 2020-2021. Tratto da https://wedocs.unep.org/20.500.11822/35275 1A
TAHMO. (s.d.). TAHMO. Tratto da https://tahmo.org/. 1A