Synopsis
The presentation examines the environmental costs of generative AI, including energy use, water consumption, resource extraction, and e-waste. It also explores AI’s indirect effects on education, showing that while AI can enhance certain analytical competences, it may weaken critical thinking, ethical engagement, and sustainability-oriented agency.
Abstract
The rapid expansion of generative artificial intelligence (GenAI) is underpinned by an equally rapid expansion of material infrastructure: energy-intensive data centres, water-dependent cooling systems, complex semiconductor supply chains, and accelerating hardware turnover that contributes to the global e-waste burden. These material conditions have made the environmental impact of AI increasingly visible in public debate. Environmental organisations (The Guardian, 2025), student initiatives (The Times, 2025), and civil society campaigns (AlgorithmWatch, 2025; Associated Press, 2026) have begun to challenge the routine expansion of AI infrastructures and services, drawing attention to their ecological and social costs (Green America, 2025). At the same time, the environmental significance of AI cannot be reduced to a simple narrative of harm. Alongside critical perspectives, a growing body of academic, institutional, and industry literature argues that AI can also generate substantial environmental benefits when deployed responsibly and within appropriate regulatory and organisational frameworks (Luccioni et al, 2025).
From this perspective, AI may contribute to sustainability by optimising complex systems such as electricity grids, transport networks, agricultural production, and building operations, thereby reducing emissions, energy consumption, water use, and material waste (UNEP, 2025). It can also enhance environmental monitoring and prediction, including climate modelling, biodiversity tracking, and methane leak detection, and may accelerate sustainable innovation in areas such as low-carbon materials, batteries, and resource-efficient supply chains (Reuters, 2025). The environmental evaluation of AI is therefore necessarily ambivalent: its direct ecological footprint may be significant, yet its broader systemic effects may also include environmentally beneficial applications. Any serious assessment must hold these two dimensions together rather than privileging either techno-optimism or blanket rejection.
The first part of our presentation addresses these direct environmental impacts of AI through a lifecycle perspective. Rather than focusing narrowly on simplified indicators such as “emissions per prompt,” we situate GenAI within a broader socio-technical system and examine five interrelated dimensions: (i) electricity use and associated carbon emissions across model training and inference, (ii) water withdrawals and water consumption linked both to data-centre cooling and upstream electricity generation, (iii) the water and chemical intensity of semiconductor manufacturing, (iv) resource extraction and dependence on critical minerals, and (v) waste and disposal challenges associated with accelerated hardware refresh cycles. This part of the session is designed in an interactive format, including a quiz comparing selected environmental costs of AI infrastructure with other forms of human activity. The aim is not merely to provide information, but to encourage participants to move beyond intuitive assumptions and adopt a more systemic understanding of AI’s direct environmental implications.
However, a focus on direct impacts alone is insufficient. It overlooks a second, less visible, yet important dimension: the indirect impact of AI on how people think, learn, judge, and act in relation to environmental problems. As AI becomes more embedded in higher education, its role is no longer merely technical; it is also pedagogical. It shapes students’ competences, habits of inquiry, relationship to knowledge, and capacity to engage with socio-ecological challenges. AI may therefore influence environmental outcomes not only through what it directly consumes or optimises, but also through how it affects those who will later analyse crises, design policies, and respond collectively to ecological problems.
The second part of our presentation examines these indirect effects through the lens of sustainability-related competences. To explore this issue, we draw on empirical research conducted in an environmentally oriented university course focused on environmental economics, policy, and tourism. The study examined students’ reflections on AI through the GreenComp framework. Methodologically, it was based on qualitative semi-structured interviews with nineteen students who had experienced teacher-assisted and active use of AI, predominantly ChatGPT. AI was integrated into a structured pedagogical process.
The findings indicate that AI’s indirect environmental relevance through education is real, yet uneven. According to the interviewed students, AI has the strongest potential to support Problem framing, Holistic thinking, and Individual initiative. Students described AI as useful for generating multiple perspectives, structuring complex issues, identifying solution pathways, and supporting orientation in unfamiliar domains. It appeared particularly valuable where environmental learning required opening up a problem space, comparing viewpoints, and searching for actionable responses.
At the same time, the findings were far from unequivocally positive. Systems thinking, Critical thinking, Futures literacy, and Adaptability were perceived more ambivalently. Although students recognised benefits, they also identified risks: uncritical acceptance of AI-generated outputs, superficial understanding, and dependence on convenient but insufficiently verified answers. These concerns suggest that AI may support selected cognitive operations while simultaneously weakening the habits of judgement and verification that sustainability education ought to cultivate. Its educational effects therefore depend less on the technology itself than on the pedagogical conditions under which it is embedded and guided. The lowest perceived potential was associated with Promoting nature and Political agency, while Collective action, Supporting fairness, and Valuing sustainability also remained weak or contested.
We therefore argue for a more balanced framework for evaluating AI in relation to environmental protection and sustainability. Direct impacts such as energy, water, and material consumption must remain, but they should be complemented by a systematic analysis of indirect pedagogical effects.
References
AlgorithmWatch (2025). How to Resist Data Centers: A Guide For Local Communities in Europe. https://algorithmwatch.org/en/a-guide-to-data-centers/
Associated Press (2026). A chatbot entirely powered by humans, not artificial intelligence? This Chilean community shows why. https://apnews.com/article/ai-human-chatbot-chile-quiliai-data-centers-water-a798d7d1ae3b2fd7646ec5dbfc8a18dd
Green America (2025). No “Dirty Data”: New Campaign Urges AI Companies to Use Cheap Renewable Energy, Stop Propping Up Fossil Fuels and Nuclear. https://greenamerica.org/press-release/push-clean-ai-end-fossil-fuel-nuclear-dependence
Luccioni, A. S., Strubell, E., & Crawford, K. (2025). From Efficiency Gains to Rebound Effects: The Problem of Jevons’ Paradox in AI’s Polarized Environmental Debate. Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency, 76–88. https://doi.org/10.1145/3715275.3732007
Reuters (2025). AI economic gains likely to outweigh emissions cost, says IMF. https://www.reuters.com/sustainability/climate-energy/ai-economic-gains-likely-outweigh-emissions-cost-says-imf-2025-04-22/
The Guardian (2025). More than 200 environmental groups demand halt to new US datacenters. https://www.theguardian.com/us-news/2025/dec/08/us-data-centers
The Times (2025). Gen Z students won’t use ChatGPT — but not because it’s cheating. https://www.thetimes.com/uk/education/article/gen-z-students-wont-use-chatgpt-but-not-because-its-cheating-v8rffjlc0
UNEP (2024). AI has an environmental problem. Here’s what the world can do about that. https://www.unep.org/news-and-stories/story/ai-has-environmental-problem-heres-what-world-can-do-about