Many researchers and policymakers believe Artificial Intelligence, or AI, will provide the most revolutionary solutions to climate change. Applications for optimizing energy demand and supply, speeding the discovery and development new materials, and helping in the prediction and mitigation of adverse climate effects are some of the potential applications.AI development and climate changes are however deeply embedded in geopolitical contexts. This makes it difficult to find solutions.
The current trajectory and geographic concentration in AI development and deployment as well as the institutional capacity of these countries, will ensure that only a few countries will reap the benefits of AI technology. Indeed, the top 10 in Oxford Insights’ Government AI Readiness Index (2020) lists only two countries outside of North America and Europe. The report notes, “The lowest-scoring regions on average are Sub-Saharan Africa, Latin America and the Caribbean, and South and Central Asia. This reflects a persistent inequality in government AI readiness.”
This inequity will be reflected in climate change policy, which itself is marked by inequalities in responsibility, capacity and ability to monitor and respond. As analysts have pointed out, the majority of emissions have been attributed to developed countries in the past. However, developing countries are burdened with the responsibility of complying. These observers call for a distinction between “survival emissions” of vulnerable communities, especially in developing countries, and “luxury emissions” of the developed ones.
The mainstreaming and adoption of AI and other emerging technologies will require significant emissions. At the same time, AI capacity in terms of R&D, investment, data, and infrastructure is currently skewed, focused within a handful of countries, primarily in the developed West. This report examines how global inequalities in AI and climate changes interact and concludes by making recommendations. It builds on expert views shared during ORF’s digital roundtable, “Solving” Climate Change: AI for a Sustainable and Inclusive FutureIn early 2021.
Mise en scène: A History of Emissions
The international community’s measurement of emissions and how it defines the problem and responsibility for them is highly political. Different stakeholders have sought out to influence debates on emission reduction by using metrics or models that best suit their narratives.
The principle of Common but Differentiated Responsibilities (CBDR) was formalised in the 1992 Rio Declaration, and institutionalised in the 1997 Kyoto Protocol—a result of efforts by the G77 bloc, led by China and India.The 2007 Bali Conference saw the signing of voluntary mitigation measures by developing countries (nationally appropriate mitigation actions, or NAMAs). These were eligible for financing from developed nations.CBDR is based upon the accumulation of GHG emissions over time as well as the different financial and technological capabilities to combat climate change. The Kyoto Protocol, while acknowledging the argument that historical emissions were made by developing nations, did not explicitly define CBDR in such terms due to opposition from wealthy countries.
In the early 2000s, the rapid economic growth of a subset of the G77—Brazil, South Africa, India, China (BASIC)— led to increasing pressure on these emerging economies to contribute to mitigation efforts. BASIC broke with the G77 and announced they would reduce their emissions intensity involuntarily.The 2009 Copenhagen conference established a three-tiered system of CBDR. It separated least developed countries (LDCs), and small island developing states (SIDS), from developing countries. Developed countries also agreed to contribute USD 30 billion between 2010-2012 and USD 100 billion by 2020 for mitigation efforts.
In 2015, the Paris Agreement settled on a system of differentiated pledges, called “intended nationally determined contributions” (INDCs).Bis now, 192 countries submitted their INDCs. They are expected to update them every five years.While the Paris Agreement retains the CBDR principle in spirit, it allows for more flexibility for all parties. It moves out of the Annex I – Annex 2 dichotomy of the Kyoto Protocol, and leaves NDCs to a country’s own assessment of its “national circumstances”. The agreement also, notably, mentions climate justice and “the imperatives of a just transition of the workforce and the creation of decent work and quality jobs in accordance with nationally defined development priorities.”
That said, the idea of “just transitions” is still underdeveloped, and “the framing is emerging from a Global North perspective” that does not reflect how vulnerable communities in the ‘Global South’ would disproportionately bear the risks for this shift.Emissions trading is an excellent example of this skew.[a] a mainstay of climate change mitigation since the Kyoto Protocol. Political scientist Lorenzo Fioramonti writes,
The lure of economic reasoning and its claim of neutrality can be very appealing. In fact, the reliance on cost–benefit analysis is a fundamentally macabre exercise, which overly simplifies the multidimensional character of social problems and makes us blind to the persistence of power structures that oppose the resolution of longstanding global problems.
Critics say that emissions markets are a way for rich countries to avoid ethical obligations. They also perpetuate exploitative “colonial” patterns that benefit developed countries at the cost of developing ones.Will climate change action create an inequitable situation?
AI and Climate in an Unequal World
A small number of countries have the most AI resources in terms of investment, research and talent. These same countries also reap the economic and social rewards. In other words, “Those best-positioned to profit from the proliferation of artificial intelligence (AI) systems are those with the most economic power.”Inequality is a multifaceted problem in AI. It includes the data that feeds algorithms, their coders, the existence of well-funded research institutes, and the government’s capacity to support and give direction to the development and implementation of AI.
As of December 2020 32 countries had developed an AI strategy and 22 more are currently in the process. According to Oxford Insights’ AI Readiness Index, Sub-Saharan Africa (SSA), Latin America, the Caribbean and South and Central Asia (with some exceptions) are the lowest-scoring regions: “If inequality in government AI readiness translates into inequality in AI implementation, this could entrench economic inequality and leave billions of citizens across the Global South with worse quality public services.”Low and middle income countries (LMICs), face challenges due to the unreliable availability of basic infrastructure, such as electricity or high-speed internet. Similar inequalities mark R&D, patents, startups, funding, skilling and hiring in AI, with the United States and Europe accounting for the lion’s share of investment, academic output, and hiring (see Figures 1a,1b, and 1c).
From the Top Left:Number of AI Patents; Global distribution private investments in AI, AI hiring.
Source:Global AI Vibrancy Tool. Stanford University. Regional labels are the author’s own.
How would the concentration of AI development and capacity—technical and governance—in the Global North affect emissions, and by extension, emission politics and narratives? In their 2019 essay, Nandini Chami of IT for Change and Anita Gurumurthy write:
The AI-led global order is deeply rooted in what activists and scholars call neocolonisation. Economic power is today a function of how AI technology is used in networked systems based around incessant data processors. The internet allowed data to flow on a global level, creating and multiplying economic connections and predatory capitalism was revived.
The first-mover advantage for AI lies with the incumbents of digital revolution. These are the ones who have shaped global value chain structures. This advantage is not just in data, capital, and digital infrastructure. It also includes their ability to determine the terms under which other actors can engage in governance and ethical discussions. (This idea will be explored in the last section.)
Elephant in the Dark: Granularity In Emissions
ICT was responsible for between 0.8 to 2.3 percent of global GDP in 2020. gigatons Global GHG emissions: CO2eq. Researchers put ICT’s contribution at 1.8 and 2.9 percent of global emissions according to low and mid estimates, and up to 6.3 percent per the “worst-case” estimates.The adoption of AI across different sectors has increased dramatically, while at the same time, AI development has accelerated and adoption has increased.[b]These models can be used in conjunction with larger AI models.
The compute demand of large AI models, according to a 2018 study by OpenAI, has been doubling every 3.4 months—meaning that since 2012, compute has grown by 300,000 times.Some studies also attempted to quantify the carbon emissions from neural network training in different regions based on location of the server, type and duration of training.Another study on Natural Language Processing models (NLP) found that training one model produced five times as much CO than a single model.2 Emissions as a car throughout its entire life.
Table 1: Carbon Footprint for Major NLP Models
As a proxy for AI-related compute need, we can use data centre energy usage. A 2015 research paper on ICT-linked electricity consumption estimated energy use of data centres to hit 539 TWh in 2018, and 2967 TWh in 2030, even with improvements in efficiency.In 2017, the paper was updated to reflect that data centers could account for 3.2 per cent of global carbon emissions by 2025. A 2020 study measuring energy use against compute demand, from 2010 to 2018, noted a 6-percent increase in energy use. Global data centre energy use, it further found, accounted for 1 percent of global electricity consumption, which—for comparison—is more than the total electricity consumption of a country like Thailand.Current projections suggest that the APAC market for data centres will grow by 12.2 per cent between 2020-24. However, Southeast Asia will only grow at 12.9 per cent. Europe, Africa, and the Middle East follow at 11.1 percent and North America at 6.4%, respectively.
Figure 2: Regional growth in Data Centre Markets
One problem with many of these data center studies is that the geographic groups they use do not provide granular insights. Figure 3 demonstrates this using regional data centre statistics from four major cloud service providers—Amazon Web Services, Google Cloud, IBM Cloud, and Microsoft Azure—followed by a breakdown by country in Table 2.
Figure 3: Data centres by region
Table 2: Data Centers by Country
Regional data can be useful for providing high-level insights such as the lack of centres in South America and Sub-Saharan Africa, but it is not complete. For example, the United States alone is home to the vast majority of data centres run by the big four and an average of 39.5 per cent of availability zones. Worldwide.Despite the fact that other regions are expecting double-digit growth over the next decade, the data centre infrastructure is unbalanced and will likely remain so for the foreseeable future.
Looking for an Equitable Model of Sustainable AI?
Governments, regional institutions, as well as technology companies, have already begun to create narratives on AI, and other emissions. This could create the same inequalities as have been associated with climate agreements.
For instance, technology giants have responded to climate concerns by announcing “net zero” policies and initiatives. Microsoft has pledged that it will be carbon-negative by 2030 and eliminate all carbon emissions from the company since 1975.Alphabet has, for its part announced sustainability bonds in the amount of USD 5.75 billion to fund environmentally and socially-responsible projects.Facebook is taking initiatives to create sustainable supply chains in the meantime.Amazon has pledged to be net-zero in 2040.
Such pledges are a signalling tool, and they indicate that internet giants have acknowledged their enormous carbon footprint. They often rely on the inequitable carbon offset system, which has been criticized for allowing companies to buy their way out from making fundamental changes in their operations.[c] There is also little transparency regarding the lifecycle emissions of their operations, including not just the facilities under their direct ownership, but within their broader global supply chains—this makes these “net zero” claims nearly impossible to measure.
Another example is the emphasis on compute efficiency. Google’s Switch Transformer was launched in January 2021. This is a smaller, more efficient version of the more cumbersome Transformer. The idea behind increasing compute efficiency involves increasing the number and performance of neural networks while maintaining constant costs. Yet, the emphasis on “efficiency” as the silver bullet to offset emissions distracts from the fact that there is still little transparency on the impact of such measures on actual life cycle emissions, leaving independent researchers who may want to verify these claims in the dark. The Jevons Paradox also states that efficiency-oriented solutions can have the second- or third-order effect of increasing consumption and reducing costs. This relationship has been observed historically in ICT-enabled efficiency improvement, where efficiency gains in energy usage required for ICT decreased production costs, which led increase in overall energy consumption.
This phenomenon is captured in two relatively new terms—“ethicswashing” and “greenwashing”. Entities want to set up an ethical checkbox to address the concerns of their customers and shareholders who are increasingly concerned about climate change. However, they do not have to make any significant changes to their global operations.
Assessment of local impact. As we have already mentioned, policy action is only possible with granular data. Researchers have already begun to develop models for assessing the impact of emissions.But they need more robust data to be able to make actionable suggestions. Other researchers also recommend integrating an Environmental, Social, and Governance framework (ESG) into AI governance.
We are working towards the establishment of common standards across all geographies for AI emissions governance. Some countries have already adopted carbon-neutral requirements to their data centres. For instance, several CSPs and data centre operators with a presence in Europe—including Google, IBM, AWS, Intel, and Microsoft—have signed a Climate Neutral Data Centre Pact, part of the EU’s roadmap to becoming carbon-neutral by 2050.The Pact aims to achieve targets in energy efficiency and transition to clean energy. The European Commission will monitor these voluntary commitments. The danger of non-uniform standards is the creation of a new form of “carbon havens”, where global enterprises might move operations to developing countries with comparatively lax regulations on emissions linked to AI and allied technologies.
In the context of climate effects of AI, developing countries should consider the CBDR principle. Developing countries need to get ahead of the game by actively participating in the process of defining parameters that will affect the climate impact of AI. Small and developing countries are playing catch-up in AI and competing against large and powerful economies. While AI’s economic growth imperative is understandable, it is important to engage in emerging climate and AI debates. Otherwise, these narratives, and soon, governance processes will be shaped by the terms and contexts set by a small number of powerful actors.
“Solving” Climate Change: AI for a Sustainable and Inclusive Future
22 February 2021
ORF organized virtual consultations in February 2021 with a focus of stakeholders from the Global South countries. Participants included representatives from industry, academia, civil society organizations, and relevant government officials. The discussions were focused on AI within the context of sustainable development goals, specifically SDGs 10, 13 and 13, on reduced inequality and climate action, respectively.
The purpose of the consultations was to find answers to the following questions.
Participants identified four key issues and ideas. First, striking a balance between state- and community-centred approaches. While the state is still the center of global governance efforts and the main focus of the state’s efforts to ensure sustainable AI, relying on only state-centred approaches to sustainable AI could lead to the marginalisation of stakeholders whose interests are not represented at the national level. This could be due to a lack of visibility or resources to make their voices heard, or even persecution. Second, the inability to establish interfaces between AI governance processes and climate change. Sustainability needs to become a core principle under ethical AI, and requires the active buy-in of industry, government and multilateral/multistakeholder bodies. Third, developing countries should make sustainable AI a priority policy priority. The COVID-19 pandemic will further intensify the focus on economic recovery for developing countries, but sustainable recovery—including through sustainable AI—should remain in focus. The need to recognize differences in capacities is also important. A common framework to sustain AI should recognize differences in capacities while also balancing the geopolitical framing which characterizes global governance of emerging technologies.
|Abhishek Gopta||Founder, Montreal Institute of AI Ethics & Machine Learning Engineer, Microsoft|
|Anirudh Kanisetti||Associate Fellow, Takshashila Institution|
|Arthur Vieira||Plataforma CIPÓ|
|Attlee Gamundani||Young ICTD Fellow at United Nations University Institute in Macau|
|Christopher Cordova||Director, AI for Climate with Co-founder|
|Danit Gal||Associate Fellow, Leverhulme Centre for the Future of Intelligence|
|Emanuela Girardi||Founder, Pop AI & High-Level Expert Group on AI of the Italian Government, Italy|
|Eniola Mafe||Lead, 2030 Vision Secretariat World Economic Forum|
|Gabrielle Alves||Junior Researcher, Plataforma CIPÓ|
|Janet Salem||Economic Affairs Officer, UN Economic and Social Commission for Asia and the Pacific|
|Marie-Therese Png||Ph.D. Candidate, Oxford Internet Institute|
|Olga Cavalli||Co-founder/Academic Director, ARGENSIG – SSIG|
|Priya Donti||Climate Change AI Co-founder|
|Serge Stinckwich||Head of Research, UN University Institute of Macau|
Trisha RayAssociate Fellow at Centre for Security, Strategy and Technology, Observer Research Foundation.
[a] “The countries or companies that reduce emissions below their cap have something to sell, an unused right to emit, measured in tonnes of CO2 equivalent. Countries and companies that don’t meet their target can buy these one-tonne units to make up the shortfall. This is called emissions trading, or cap and trade.”
See: “What are Market and Non-Market Mechanisms?”, UNFCCC
[b] ‘Compute demand’ refers to the demand for computational power to carry out computing tasks, such as storage, processing and analytics.
See: United Nations Carbon Offset Platform
Cedric Villani, “For a Meaningful Artificial Intelligence: Towards a French and European Strategy”, AI for Humanity (2018).
“Artificial Intelligence Strategy of the German Federal Government”, The Federal Government of Germany (2020),
“Austria AI Strategy Report”, European Commission (2019),
 “Government AI Readiness Index (2020)”, Oxford Insights https://www.oxfordinsights.com/government-ai-readiness-index-2020
 Anil Agarwal and Sunita Narain, “Global Warming in an Unequal World: A Case of Environmental Colonialism”, New Delhi, India: Centre for Science and Environment (1991).
 ORF Consultations: “Solving” Climate Change: AI for a Sustainable and Inclusive Future”, February 22, 2021 (see Annex).
“The Cancun Agreements: Outcome of the Work of the Ad Hoc Working Group on Long-term Cooperative Action Under the Convention,” 15 March 2011, Doc. FCCC/CP/2010/7/Add.l,http://unfccc.int/resource/docs/2010/cop16/eng/07a01.pdf
Sikina Jinnah, “Makers, Takers, Shakers, Shapers: Emerging Economies and Normative Engagement in Climate Governance”, Global Governance, Vol. 23, No. 2 (Apr.-June 2017), pp. 285-306,https://www.jstor.org/stable/44861124
 Sikina Jinnah, “Makers, Takers, Shakers, Shapers: Emerging Economies and Normative Engagement in Climate Governance”.
Take, for example: The Kyoto Protocol on Climate Change and its Implications: Hearing 105-457 before Senate Committee on Foreign Relations, 105th Congress, 1998
 Anirruddh Mohan, “From Rio to Paris: India in global climate politics”, ORF Occasional Paper 130, Observer Research FoundationDecember 2017,
 Anirruddh Mohan, “From Rio to Paris: India in global climate politics”.
 “Paris Agreement”, UNFCCC (2015), https://unfccc.int/sites/default/files/english_paris_agreement.pdf
 Kate Hampton, Mridula Pandey and Shirish Sinha, “A just path to a decarbonised future”, Observer Research Foundation, November 12, 2021 https://www.orfonline.org/expert-speak/a-just-path-to-a-decarbonised-future/
 Lorenzo Fioramonti, How Numbers Rule The World: The Use and Misuse of Statistics in Global Politics (Zed Books, London: 2014): p. 86.
 Angela Carpenter, and Marcus Wagner, “Environmental justice in the oil refinery industry: A panel analysis across United States counties”, Ecological Economics 159 (2019): 101-109.
Heidi Bachram, “Climate fraud and carbon colonialism: the new trade in greenhouse gases”, Capitalism is nature socialism 15, no. 4 (2004): 5-20.
 “AI Policy and National Strategies”, Artificial Intelligence Index Report 2021, Stanford University
 “Government AI Readiness Index (2020)”, Oxford Insights.
 “Government AI Readiness Index (2020)”, Oxford Insights.
 Data from “Global AI Vibrancy Tool”, Stanford University.
 “Global AI Vibrancy Tool”, Stanford University
 Anita Gurumurthy and Nandini Chami, “The Wicked Problem of AI Governance”, Artificial Intelligence in India Vol. 2 (2019), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3872588
 Charlotte Freitag, Mike Berners-Lee, Kelly Widdicks et al, “The climate impact of ICT: A review of estimates, trends and regulations “, arXiv (2020),https://arxiv.org/ftp/arxiv/papers/2102/2102.02622.pdf
 Alexandre Lacoste, Alexandra Luccioni, Victor Schmidt et al, “Quantifying the Carbon Emissions of Machine Learning”, arXiv1910.09700v2 [cs.CY], November 4, 2019, https://arxiv.org/pdf/1910.09700.pdf
 Emma Strubell, Ananya Ganesh and Andrew McCallum, “Energy and Policy Considerations for Deep Learning in NLP”, arXiv:1906.02243v1 [cs.CL],
 Charlotte Freitag, Mike Berners-Lee, Kelly Widdicks et al, “The climate impact of ICT: A review of estimates, trends and regulations “, arXiv (2020),
 Anders S.G. Andrae, “Total Consumer Power Consumption Forecast”, Nordic Digital Business Summit (October 2017).
 Eric Masanet, Arman Shehabi, Nuoa Lei, Sarah Smith, and Jonathan Koomey, “Recalibrating global data centre energy-use estimates.”, Science 367, no. 6481 (2020): 984-986.
Because there is no standard for measuring ICT’s energy and emissions, estimates can vary.
See also: Lotfi Belkhir, Ahmed Elmeligi, “Assessing ICT global emissions footprint: Trends to 2040 & Recommendations”, Journal of Cleaner Production Volume 177 (2018): Pages 448-463
 “Energy Consumption: International”, U.S. Energy Information Administration,
 “Data centre Market Global Comparison”, Cushman & Wakefield (2021),
 “Data centre Market Global Comparison”, Cushman & Wakefield.
 “Regions and Availability Zones”, Amazon Web Services https://aws.amazon.com/about-aws/global-infrastructure/regions_az/
“Discover our Data Center Locations”, Google Data Centers https://www.google.com/about/datacenters/locations/
“IBM Cloud global data centers”, IBM Cloud https://www.ibm.com/cloud/data-centers/#datacentermap
“Global Infrastructure”, Microsoft Azure, https://infrastructuremap.microsoft.com/
  “Regions and Availability Zones”, Amazon Web Services.
“Discover our Data Center Locations”, Google Data Centers
“IBM Cloud global data centers”.
“Global Infrastructure”, Microsoft Azure.
There are also variations in electricity consumption per capita. However, there is also variation in energy mix and use of data centres between locations within the same region. This is hidden in grouped data. For example, cloud servers located in North America emit anywhere between 20g CO2eq/kWh, Quebec, Canada, to 736.6CO2eq/kWh, Iowa, US.
 Brad Smith, “Microsoft will be carbon negative by 2030”, Microsoft BlogJanuary 16, 2020 https://blogs.microsoft.com/blog/2020/01/16/microsoft-will-be-carbon-negative-by-2030/
 Ruth Porat, “Alphabet issues sustainability bonds to support environmental and social initiatives”, Google Blog August 3, 2020 https://blog.google/alphabet/alphabet-issues-sustainability-bonds-support-environmental-and-social-initiatives/
 “How Facebook partners with academia to help drive innovation in energy-efficient technology”, Facebook Research, February 5, 2020https://research.fb.com/blog/2021/02/how-facebook-partners-with-academia-to-help-drive-innovation-in-energy-efficient-technology/
 William Fedus, Barret Zoph, Noam Shazeer, “Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity”, arXiv:2101.03961 [cs.LG], https://arxiv.org/abs/2101.03961
 Miriam Börjesson Rivera, Cecilia Håkansson, Asa Svenfelt and GöranFinnveden, “Including second order effects in environmental assessments of ICT”, Environmental Modelling & Software, 56 (2014): pp.105-115.
Ray Galvin, “The ICT/electronics question: Structural change and the rebound effect”, Ecological Economics, 120 (2015): pp. 23-31, https://files.ifi.uzh.ch/hilty/t/Literature_by_RQs/RQ%20301/2015_Galvin_The_ICT%EF%80%A2electronics_question.pdf
 Natasha Bernal, “Google, Microsoft and the strange world of corporate greenwashing”, Wired, January 31, 2020 https://www.wired.co.uk/article/corporate-greenwashing
See also: “Ethics Washing”, AI Ethics Living Dictionary, https://montrealethics.ai/dictionary/
 See, for example: Dennis Bouley, “Estimating a Data Center’s Electrical Carbon Footprint”, White Paper 66, Schneider Electric, https://www.insight.com/content/dam/insight/en_US/pdfs/apc/apc-estimating-data-centers-carbon-footprint.pdf
David Patterson, Joseph Gonzalez, Quoc Le et al, “Carbon Emissions and Large Neural Network Training”, arXiv:2104.10350v3 [cs.LG], April 23, 2021 https://arxiv.org/pdf/2104.10350
 Abhishek Gupta, Camylle Lanteigne and Sara Kingsley, “SECure: A Social and Environmental Certificate for AI Systems”, arXiv:2006.06217v2 [cs.CY], July 19, 2020