Artificial intelligence presents a profound environmental paradox: whilst promising to optimise energy systems and model climate change with unprecedented precision, AI itself consumes staggering amounts of electricity and generates substantial carbon emissions. Training a single large language model can produce as much carbon as five cars over their entire lifetimes—a carbon footprint that makes even the most profligate SUV driver look environmentally conscientious by comparison (Strubell et al., 2019). Data centres powering AI systems already account for roughly 1% of global electricity demand, a figure projected to grow exponentially as AI deployment accelerates. This creates an uncomfortable question: can AI help solve the climate crisis if it simultaneously exacerbates it? The answer requires nuanced examination of both AI’s environmental costs and its potential contributions to sustainability (Cowls et al., 2021; Crawford, 2021).
The energy demands of AI stem primarily from the computational intensity of training and running sophisticated models—processes that make traditional computing look positively frugal. Modern AI systems require vast arrays of processors running continuously, often for weeks or months, consuming electricity at rates that would make cryptocurrency miners jealous. The cooling systems necessary to prevent these processors from melting into expensive puddles add further energy overhead (García-Martín et al., 2019). Moreover, as AI capabilities advance, computational requirements grow exponentially: each new generation of models demands orders of magnitude more processing power than its predecessor. This creates a worrying trajectory where AI’s energy appetite grows faster than improvements in energy efficiency can compensate (Schwartz et al., 2020).
Yet AI also offers genuine potential for environmental benefit, provided we can resist the temptation to use it for every conceivable application regardless of necessity. AI-optimised energy grids can balance supply and demand more efficiently, reducing waste and integrating renewable sources that fluctuate with weather conditions (Rolnick et al., 2019). Climate models enhanced by machine learning can predict extreme weather events with greater accuracy, enabling better preparation and response. AI systems optimise industrial processes to minimise resource consumption, design more efficient buildings, and even accelerate development of sustainable materials. In agriculture, AI-powered precision farming reduces water usage, fertiliser application, and pesticide deployment whilst maintaining crop yields—benefits that sound almost too good to be true, and may well prove so if implementation lags behind promises (Kaack et al., 2022).
The environmental balance sheet for AI thus depends critically on how we deploy the technology—a dependency that requires more careful consideration than it typically receives. Using AI to optimise renewable energy systems represents a sensible application where benefits likely outweigh costs. Deploying AI to generate personalised advertising or recommend social media content offers environmental costs with negligible societal benefit, though profitability apparently trumps sustainability in corporate decision-making (Cowls et al., 2021). The geographic location of data centres matters enormously: facilities powered by renewable energy in cool climates requiring minimal cooling carry far lighter environmental footprints than those burning coal in tropical regions. Yet economic incentives often favour cheaper, dirtier locations over sustainable ones (Crawford, 2021).
Addressing AI’s environmental impact requires regulatory frameworks that account for both costs and benefits—frameworks currently conspicuous by their absence. Carbon taxes on AI training and deployment could incentivise efficiency, though implementation faces political obstacles from industries wielding substantial lobbying power. Standards requiring environmental impact assessments before deploying large-scale AI systems could prevent the most egregious waste. Investment in energy-efficient AI hardware and algorithms promises incremental improvements, though breakthrough innovations may prove necessary. Ultimately, reconciling AI with environmental sustainability demands conscious choices about which applications justify their environmental costs—choices that current market dynamics fail to encourage. The irony of using a climate-damaging technology to address climate change would be amusing if the stakes weren’t so high (Hao, 2019; Rolnick et al., 2019).
References
Cowls, J. et al. (2021) ‘The AI gambit: Leveraging artificial intelligence to combat climate change’, AI & Society, 36, pp. 1035-1055.
Crawford, K. (2021) Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. New Haven: Yale University Press.
García-Martín, E. et al. (2019) ‘Estimation of energy consumption in machine learning’, Journal of Parallel and Distributed Computing, 134, pp. 75-88.
Hao, K. (2019) ‘Training a single AI model can emit as much carbon as five cars in their lifetimes’, MIT Technology Review, 6 June.
Kaack, L.H. et al. (2022) ‘Aligning artificial intelligence with climate change mitigation’, Nature Climate Change, 12(6), pp. 518-527.
Rolnick, D. et al. (2019) ‘Tackling climate change with machine learning’, arXiv preprint arXiv:1906.05433.
Schwartz, R. et al. (2020) ‘Green AI’, Communications of the ACM, 63(12), pp. 54-63.
Strubell, E., Ganesh, A. and McCallum, A. (2019) ‘Energy and policy considerations for deep learning in NLP’, Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 3645-3650.