Jump to content

Course:CONS200/2026WT2/AI and climate governance: an overview of opportunities and potential risks

From UBC Wiki

Introduction

Climate governance refers to the systems and institutions that collectively take part in efforts to mitigate and adapt to climate change through governments, organisations, and communities at the international, national, and local levels[1]. Artificial intelligence (AI) encompasses machine learning, large language models, and data-driven prediction systems, and has begun to emerge as a potentially transformative tool within climate governance, offering new capacities for processing complex environmental data and supporting policy decisions[2].

The potential applications of AI in climate governance are wide-ranging. These include improving the forecasting of climate phenomena [3][2] analyzing critical infrastructure for adaptation planning [3][2], and encouraging human-centred climate policy design by synthesizing perspectives from a broader range of stakeholders [4] [3]. Furthermore, AI’s ability for regional models and countries’ pursuit of “AI Sovereignty” has the potential to increase equitable benefit-sharing [5].

However, these opportunities come with significant risks that cannot be overlooked. AI systems demand substantial energy, water, and emit greenhouse gasses during both training and operation, raising questions about their net benefit to climate mitigation efforts [2][3] [6][7]. Concerns around algorithmic bias, data inequities, and the concentration of AI's benefits in wealthier nations further complicate its deployment in climate governance [3][8][9][5]. Without explicit attention to equity, AI risks reinforcing existing global inequalities and deepening climate injustice for the communities most vulnerable to its impacts [5]. This page examines both the opportunities and the risks of AI in climate governance, before turning to recommendations for a more equitable and accountable path forward.

Opportunities

Data Collection and Analysis

Artificial Intelligence is a tool built on machine learning and as such can be used to better collect, sort, and analyze large amounts of data in ways humans cannot [4]. Being able to learn from existing information and information given to it lets it combine information in ways that simpler search engines cannot [2].  Large and broad datasets that were once slow to examine can now be processed at a much faster rate and efficiency, allowing for more concise and organized data [3]. This allows for easier processing and management which offloads the work required by analysts, scientists, decision makers and more, in turn, allowing for faster and more holistic decisions to be made [4].

In the context of climate change, AI has been used for climate risk assessment, climate forecasting and future vulnerability models, policy analysis, human behaviour pattern detection during disaster, and more [3]. Due to AI’s ability to both process large amounts of data efficiently and decode multi-variable problems with non-linear relationships, it has become a key tool in aiding climate data analysts [3]. The biggest use of AI has been for climate risk assessment of natural disasters and their impacts [3]. This has been done through the facilitation of real-time data collection, improved analysis of data, helping increase communications between parties, and aiding decision making processes [3]. Climate forecasting models are further used to predict weather, its impacts on droughts and floods, and analyse the potential vulnerabilities in infrastructure and communities [3].

Aside from scientists and data analysts, AI also has the potential to aid data collection and analysis for communities and impacted individuals directly. AI has the ability to provide farmers with information about crop-resilient agriculture, such as providing crop management advice through chat services, pest identification and eradication strategies, and AI-powered weather apps and satellite imagery for agricultural planning [2]. Furthermore, AI chatbots and LLM powered services can provide cohesive data around biodiversity conservation support, local clean energy adoption strategies, and help write climate policy reports and project financing proposals [2]. Through these tools, data needed for climate mitigation strategies have become more available and accessible to a large range of people, especially in developing countries, slowly bridging the information gap and barriers to access vital information [2].

Human Centred Policy Design

Public opinion offers vital information for decision making, yet this data has been hard to conceptualize; AI offers a way to examine it and determine large trends in a way that wasn’t possible before [2]. By leveraging social media and other public internet posts, it offers a fast and effective way to condense this data into a manageable size while still accounting for a large array of information [4]. This speed allows for data to be collected and processed only moments after being posted online, giving access to real time, large scale analysis that couldn’t be done by hand [2].  With this, decisions can be made based on a wider polling of opinions than ever before, simultaneously allowing for decision makers to see public reaction to policies or programs at the moment they are announced. Through this AI takes a “supporting” role in decision making, helping to shape action around human opinion and needs [4].

Decision makers now have better access to what the public wants, and this can dramatically help to shape the way climate decisions are handled from local to global scales [3]. With information gathered from public posts, AI can integrate previous opinions to similar decisions to simulate public reaction to decisions before they are made, thereby helping to create more equitable and fair climate action [3]. It can also help to reshape existing, unpopular or inefficient decisions in a more thorough way than without it [4]. Integration of public opinion through AI is now allowing for more thoughtful and inclusive action to be made, benefiting both the public and environment. It can also take a more unbiased approach to potentially political or polarized topics, ensuring a more neutral approach to topics, helping to reduce argument or heavily opinionated action [4].

Regional Models and Equitable Benefit Sharing

Current climate governance systems, while generally effective on a global level, [10] are often too generalized to be extremely effective on regional and ecosystem-based levels. AI has the potential to change this, as it offers potential to delve into large and unstructured datasets and create models for climate futures that traditional methods simply cannot do [2]. AI has the ability to perform machine learning and use large language models to take snapshots of data and create a human driven modeling system that is regionally specific and socially equitable [4].

Most extreme weather prediction models struggle to emulate situations in developing countries or in areas with microclimates and topographical formations that alter the common weather in the region, which leaves many small island nations and these regions in unprecedented natural-disaster related scenarios, especially as extreme weather becomes more frequent [2]. Therefore, AI has the potential to analyze significantly more data and offer predictions for extreme weather and climate mitigation strategies in these regions, in addition to specified disaster response mechanisms based on topographical analysis in these areas [2].

Regional Map of Central America

Many real-world examples of AI being used in climate governance already exist to create regionally accurate models. For example, EMPIRIC_AI is able to simulate tropical cyclones in the south pacific with the goal of assessing the best ways for healthcare-related industries to respond to these cyclones in the event of a natural disaster [2]. In the Solomon Islands, an AI-based mapping program analyzes satellite images in order to create accurate distributions of mangrove forests, which aids conservation efforts and allows for a more specific allocation of resources in the areas that need them the most [2]. Another real-world model is used in Brazil, in which similarly analyzed satellite images are instead used to survey the Amazon rainforest for illegal roads, helping to enable preventative deforestation mechanisms and shut down illegal logging operations [2].

Artificial Intelligence is often connected with the idea of “data colonialism,” in which wealthy corporations and nations often gain access to unprecedented amounts of data, and are therefore able to use it unfairly and inequitably. Therefore, a main goal within AI climate governance must be to ensure that this information and modelling is being shared equitably [8][4]. While difficult given the nature of knowledge to congregate to power, this equitable sharing of benefits is possible through many methods.

Primarily, thanks to AI’s ability to analyze large amounts of data, it is able to take opinions of the public and policy proposals, and efficiently create “group statements” which would likely gain both public and policymakers approval very quickly [4]. This would be particularly effective in instances where climate policy is controversial, such as legislation surrounding carbon taxes. Similarly, the possibility of incorporating Indigenous knowledge with traditional forecasting systems in AI promises to create more culturally sensitive climate modelling [2].

The plethora of AI models available open-source and as public goods can be used to its advantage with climate governance. These systems, thanks to the fact that everyone has access to them, allows the vast majority of people to gain access to the climate governing information instead of it occurring behind closed doors, with little to no input from the public. Additionally, the idea of “digital sovereignty” surrounding AI models allows them to become human-tailored in specific ecological and socioeconomic contexts[2]. This allows AI to create solutions to situations that individual communities face, both environmentally and socially. A potential issue in the training of LLMs is that most datasets, especially environmentally, tend to have an unconscious bias toward men. Therefore, AI, in order to be equitable, must be trained on ungendered data sets and be able to present equitable solutions that lack this bias [2].

AI is a very promising solution for a lot of issues exhibited by traditional climate modelling systems. It can create and blend regionally, socially, and ecologically specific models in order to best help conservation mechanisms, and possesses the ability to better equate the benefits of climate governance than traditional models [4].

Potential Risks

Energy and Water Consumption

The energy consumption of AI, including the use of electricity, water consumption, land use, and mineral resources, are significantly contributing to the material footprint of AI and are key arguments to its limitations [5]. The growth of data centres are the main contributors of energy consumption—1.5% of global electricity are already being used to facilitate these spaces, and are predicted to double by 2030 [5]. However, it’s worth noting that there is limited data on the effects of AI on energy consumption and climate implications outside of data centres [5]. The ODI Global roundtable discussion highlights an effort to model the cross-sectional impacts of AI on climate change, noting more than 30 pathways that AI and climate change are connected, while also emphasizing the limited research on the energy used across the lifecycle of AI [5].

AI Data Centre


Both the embodied carbon footprint and the operational carbon footprint must be included to fully understand the scope of AI’s carbon footprint [6].  The embodied carbon footprint refers to the emissions produced from the manufacturing, transportation, and disposal of physical hardware used by AI, which includes building data centres, but also global shipping of hardware, manufacturing GPU chips, disposing chips and servers, and more [6]. The operational carbon footprint is the carbon emitted due to the electricity of running AI systems [6]. This includes data processing, experimentation, training, and inference of data, the process in which raw data is collected, experimentation of the best model architecture and approaches, the actual training of AI on data, and the deployment of the AI [6]. Furthermore, cooling systems within data centres also take up a significant amount of energy [6]. A study on Meta’s carbon footprint estimates that the ratio between its embodied carbon footprint and operational carbon footprint as 30%/70% for large scale machine learning tasks [6]. Taking into account renewable energy sources such as solar power and Meta’s other carbon removal programs, this cuts the operational carbon footprint significantly, leaving manufacturing as the main source of greenhouse gases [6].

The water consumption of AI models must not be forgotten too. This refers to the water used in producing, operating, and maintaining AI models, such as the water used in cooling systems within data centres, water used to manufacture chips, and the water consumed when producing electricity to run data centres [7]. Compared to the agricultural industry and energy production industry, AI uses significantly less water; however, as the use of AI becomes more prominent, it is necessary to take into account the environmental impacts [7]. Chat GPT is estimated to consume 500ml of clean freshwater every 20-50 questions, a number which accumulates significantly taking into account the 100 million monthly users of the platform [7]. A study in the journal Science Advances estimates the total water consumption of AI models is 3 – 6 million cubic meters per year, which is about the water consumption of 300,000 to 600,000 people [7]. Furthermore, the water use of AI systems also impacts local ecosystems through generating wastewater that can contaminate water flows and decrease biodiversity levels [7].

Finally, AI is also leading to the “gentrification of energy” – where AI systems are increasingly using up renewable energy sources in local grids, leading other users to rely on traditional carbon-intensive sources of energy [5]. Similarly, many developers prioritise short-term growth over long-term sustainability, connecting AI systems to small-scale gas units while waiting to be connected to local renewable power grids [5]. Although these decarbonisation initiatives drive the carbon footprint of AI systems down, they do so at the cost of local regions that unsuccessfully compete for the energy [5].

Security, Privacy, and the Digital Divide

AI offers significant promise, especially in its ability to create regionally distinctive models for nations and regions that struggle to accurately predict the future of the climate. However, many of these regions tend to be in LDCs (Least Developed Countries) or marginalized areas that already struggle with a lack of data and regulatory laws that may create problems with AI’s implementation for climate governance.

While AI promises itself as an exceptional solution to many of the issues that rapid climate governance faces, it faces many issues surrounding implementation. They often face a lack of broadband connectivity and face unreliable electricity still, making it impossible for AI to be used in broad swaths of the country. Additionally, due to the developing nature of these countries, they lack significant historical climate records, which hinders the ability of AI to create solutions, as it has no data to work with [2]. Additionally, most AI processing and development exists primarily in China and the USA [2]. This makes it difficult for developing countries and small island nations to access these systems for implementation, and creates a dependency upon other countries for these systems [2]. Finally, the labour to implement and maintain systems in developing countries is absent. Due to the lack of current infrastructure, there is also a lack of skilled labor that can assess problems, find solutions, and ensure that the systems run smoothly [2].

AI also suffers from cyberattacks, like any other electronic platform. Therefore, in order to implement AI in climate governance, the issue of cybersecurity must be addressed. Cybersecurity is an especially bad problem in places with little to no regulations for it, such as LDCs (Least Developed Countries) [2]. Without paramount importance being placed on data protection laws, security breaches could cripple systems or release sensitive information. Finally, marginalized populations and those who lack digital literacy stand to suffer the worst consequences of this, as their information can be collected and stored, and through a cybersecurity breach can be shared [2].

The previously mentioned barriers have extremely strong political and legal implications. The issue of “digital colonialism” stands to create problems, as the existence of most power surrounding AI resides in the control of globally northern governments and firms. This marginalizes the global south, as it becomes a place for data and labor, rather than innovation and locally distinct models. Models that are designed in the global north may not work as well for the global south, while also excluding them from being able to create their own models [5]. Implementation of regionally distinct models is necessary to allow LDCs digital sovereignty and security [5]. The development of AI models is closely linked with the generalized right to develop. Access to these tools will continue to be misaligned with necessity for these tools, as LDCs lack the infrastructure and finances to properly create, enable, and execute these models in a beneficial way [2].

Using AI for climate governance will require significant implementation globally and safeguards to ensure that entire countries do not become dependent upon other countries or companies to be able to function. Additionally, problems surrounding cybersecurity, especially amongst marginalized regions need to be addressed in order for AI implementation to continue safely. While offering potential solutions, risks surround the actual usage of AI for these tasks, as the data may be too vulnerable or potentially crippling to LDCs.

Bias, Misinformation, and Data Inequities

Because AI is dependent on human sources, it is inherently susceptible to bias, misinformation, and data inequities. Issues like this can be imperceptible to an AI, meaning its learning algorithm will take bias and misinformation as fact. These problems combine to create AIs that favor certain groups or use information that isn’t real, skewing efforts made in good and honest conscience. When looking at large open access forums like social media, this is to be expected and can be planned for; however, it can take root in less obvious ways that are harder to notice or circumvent.

Human bias appears in the data AI is trained on in numerous ways that disrupt its focus on fair usage to serve the interests of certain opinions or preexisting inequalities. It is common to see a favor towards the English online and in the data AI is trained on and uses. This leads to an uneven polling of information from the English speaking world, dampening the voices of non-English speaking nations and people [8]. On a smaller scale, AI used to plan for flooding has been found to have bias that disproportionately leaves out communities “across existing lines of vulnerability and race” [8]. The effects of this could only get worse as other AIs train on the same data and further cement this bias into its algorithm. If put into use, these communities would see more harm with the inequality currently in place deepening. Its dependence on sources that already exist make it likely to continue the discrimination currently in place [8].

Conclusion

As AI and worldwide access to it advance, its role in climate governance will continue to grow and adapt. Already, it is being used to predict, mitigate, and create solutions for disasters and the effects of a warming planet. Its ability to organize, process, and simplify large datasets will allow climate scientists and policy makers to sort data quicker and easier. This simplification process allows decisions to be informed by broader and larger sources than ever before, creating more holistic and rapid solutions. However, this comes with the cost of potential misinformation, bias, data inequity, high water and energy usage, and uneven benefits toward the global north. Despite the risks, climate governance is an ever more pressing issue, and AI proposes itself as a potentially game-changing solution.

Melting Iceberg Due to Climate Change

References

[2]

  1. ↑ Cite error: Invalid <ref> tag; no text was provided for refs named :0
  2. ↑ 2.00 2.01 2.02 2.03 2.04 2.05 2.06 2.07 2.08 2.09 2.10 2.11 2.12 2.13 2.14 2.15 2.16 2.17 2.18 2.19 2.20 2.21 2.22 2.23 2.24 2.25 2.26 United Nations Climate Change Technology Executive Committee. "Artificial Intelligence for Climate Action: Advancing Mitigation and Adaptation in Developing Countries" (PDF). United Nations Industrial Developmen Organization. United Nations Climate Change Technology Executive Committee (date). Artificial  Intelligence for Climate Action: Advancing Mitigation and Adaptation in Developing Countries. United Nations Industrial Developmen Organization https://unfccc.int/ttclear/misc_/StaticFiles/gnwoerk_static/AI4climateaction/f2922b97c4cf431996c468e622127eb5/112f8be560ea447dab5ff2e53ab3f6e4.pdf
  3. ↑ 3.00 3.01 3.02 3.03 3.04 3.05 3.06 3.07 3.08 3.09 3.10 3.11 3.12 Cite error: Invalid <ref> tag; no text was provided for refs named :2
  4. ↑ 4.00 4.01 4.02 4.03 4.04 4.05 4.06 4.07 4.08 4.09 4.10 Cite error: Invalid <ref> tag; no text was provided for refs named :3
  5. ↑ 5.00 5.01 5.02 5.03 5.04 5.05 5.06 5.07 5.08 5.09 5.10 5.11 Cite error: Invalid <ref> tag; no text was provided for refs named :4
  6. ↑ 6.0 6.1 6.2 6.3 6.4 6.5 6.6 6.7 Cite error: Invalid <ref> tag; no text was provided for refs named :5
  7. ↑ 7.0 7.1 7.2 7.3 7.4 7.5 Shaji George, A; Hovan George, A.S; Gabrio Martin, AS (2023). "The Environmental Impact of AI: A Case Study of Water Consumption by Chat GPT". Partners Universal International Innovation Journal. 1 (2): 97–104.
  8. ↑ 8.0 8.1 8.2 8.3 8.4 Cite error: Invalid <ref> tag; no text was provided for refs named :7
  9. ↑ Cite error: Invalid <ref> tag; no text was provided for refs named :8
  10. ↑ Cite error: Invalid <ref> tag; no text was provided for refs named :9

[1][2][3][4]

[5] [6]

This conservation resource was created by Course:CONS200.

[7][8]

  1. ↑ Joshi, D (2026, February 10). "AI's climate consequences: why AI governance needs to confront climate injustice". ODI: Think Change. Check date values in: |date= (help)
  2. ↑ Machen, R; Pearce, W (2025). "Anticipating the Challenges of AI in Climate Governance: An Urgent Dilemma for Democracies". Wiley Interdisciplinary Reviews Climate Change. 16 (2).
  3. ↑ Yu, D; Thompson, BD; Dubey, R (2025). "Leveraging AI to advance psychological research for climate policy". Current Opinion in Behavioral Sciences. 65: 101547.
  4. ↑ Wu, C; Raghavendra, R; Gupta, U; Acun, B; Ardalani, N; Maeng, K; Chang, G (2022). "Sustainable AI: Environmental Implications, Challenges and Opportunities" (PDF). Proceedings of Machine Learning and Systems. 4: 795–813.
  5. ↑ Mehryar, S; Vahid Yazdanpanah; Tong, J (2024). "AI and Climate Resilience Governance". IScience. 27 (6): 109812–109812.
  6. ↑ Nost, E (2024). "Governing AI, governing climate change?". Geo: Geography and Environment. 11 (1).
  7. ↑ Cognuck González, S; Numer, E (2020). "What is climate governance?". Archived from the original|archive-url= requires |url= (help) on |archive-url= requires |archive-date= (help). Invalid |url-status=https://www.unicef.org/lac/media/19651/file/what-is-climate-governance.pdf (help); Missing or empty |url= (help)
  8. ↑ UNDP; IDOS (2024). [Connections that matter: how the quality of governance boosts climate action | United Nations Development Programme "CONNECTIONS THAT MATTER: How the quality of governance boosts climate action"] Check |url= value (help). United Nations Development Programme.
Retrieved from "https://wiki.ubc.ca/index.php?title=Course:CONS200/2026WT2/AI_and_climate_governance:_an_overview_of_opportunities_and_potential_risks&oldid=891828"