Artificial intelligence in sustainability

Thirty papers, five domains, and one clear takeaway: the relationship between AI and environmental sustainability is not settled [1–20]. It involves genuine and growing promise alongside documented risk, and both sides of that picture are now grounded in evidence rather than assertion. AI has improved the accuracy and cost-effectiveness of renewable energy forecasting [5,6], enabled biodiversity monitoring at scales that conventional methods cannot match [10,11], substantially reduced resource use in agriculture [13], and helped convert waste into a recoverable material [15,17,18]. These results come from working deployments in Barcelona, Jaipur, Nairobi, and Bratislava — not modelled projections. The risks are equally concrete. AI's rising energy and water demands [1,2] could offset a meaningful share of the environmental gains that AI-optimised systems are simultaneously working to deliver. Access to those gains remains concentrated in regions with capital, infrastructure, and technical expertise [3,4], while the places facing the most acute sustainability pressures continue to be poorly served. Looking ahead, green AI architecture, federated learning for data-sparse environments, and mandatory environmental disclosure for AI systems all warrant treatment as research and governance priorities on par with application development [1,2,19,20]. What the field needs most is not more applications but better accounting: a robust, standardised way of determining whether AI's net contribution to sustainability is positive, and under what conditions it reliably can be. Developing that framework is the foundational work that the next generation of AI sustainability research needs to take on first. Three priorities emerge from this synthesis as especially pressing for the next phase of research. The first is lifecycle environmental accounting. The field cannot credibly claim to be advancing environmental sustainability while remaining agnostic about the full environmental cost of the tools it is using. This requires developing standardised methodologies for measuring not just operational energy use but manufacturing emissions, water consumption, and end-of-life waste for AI hardware and infrastructure. Several frameworks have been proposed; none has yet achieved the regulatory backing or institutional adoption needed to make reporting routine. The precedent set by financial accounting standards — where independent verification of reported figures became mandatory over several decades of regulatory development — offers a plausible roadmap, even if the timescale is daunting. The second priority is equity by design. The pattern identified across multiple domains — in which AI sustainability tools deliver their largest benefits in contexts that already have significant technical and financial capacity, while providing limited value to the communities and ecosystems under the greatest pressure — will not correct itself through incremental improvement of existing approaches. It requires deliberate design choices: building for low-connectivity environments from the outset rather than treating connectivity as a given; partnering with local institutions and knowledge holders rather than applying externally developed tools to local contexts; and funding research that prioritises the specific problems facing smallholder farmers, under-resourced conservation organisations, and municipal governments in lower-income settings. The on-device irrigation classifier [13] and the Nairobi waste management pilot [16] demonstrate that these design choices are technically feasible; what is missing is the institutional and funding infrastructure to make them the norm rather than the exception. The third priority is longitudinal evaluation. The evidence base for AI in sustainability is wide but shallow: many studies are reporting promising results from short-term pilots, and relatively few that track outcomes over multiple years, across diverse sites, and under conditions that were not partially controlled by the research team. Building that deeper evidence base requires sustained investment in monitoring and evaluation that goes beyond the typical grant funding cycle, and it requires the kind of cross-institutional collaboration — between AI researchers, domain specialists, and the communities affected by both sustainability pressures and AI interventions — that academic incentive structures do not always make easy. None of these priorities is technically demanding in the way that building a better neural network is technically demanding. They are primarily institutional and political challenges, and that is precisely why they have so far received less attention from a research community whose skills and incentives are oriented primarily towards technical innovation.

[1]        A. A. Waoo, D. D. Rao, M. P. Singh, P. K. Pareek, S. Kamal, and S. V. Pandit, "Strategizing IoT network layer security through advanced intrusion detection systems and AI-driven threat analysis," Full Length Article, vol. 12, no. 2, pp. 195–207, 2024. [2]        A. A. Waoo and B. K. Soni, "Performance analysis of sigmoid and ReLU activation functions in deep neural networks," in Intelligent Systems: Proceedings of SCIS 2021, pp. 39–52, 2021. [3]        J. K. Jain and A. A. Waoo, "An artificial neural network technique for prediction of cyber-attack using intrusion detection system," Journal of Artificial Intelligence, Machine Learning and Neural Network, 2023. [4]        M. Tiwari and A. A. Waoo, "Transforming healthcare: The synergistic fusion of AI and IoT for intelligent, personalized well-being," in Revolutionizing Healthcare: AI Integration with IoT, 2024. [5]        R. Nayak, P. S. Patheja, and A. A. Waoo, "An enhanced approach for weather forecasting using neural network," in Proc. Int. Conf. Soft Computing, 2012. [6]        S. Soni, R. Soni, and A. A. Waoo, "RFID-based digital door locking system," Indian Journal of Microprocessors and Microcontroller, vol. 1, no. 2, pp. 17–21, 2021. [7]        J. K. Jain, A. A. Waoo, and D. Chauhan, "A literature review on machine learning for cyber security issues," International Journal of Scientific Research in Computer Science, 2022. [8]        R. Vinuesa, H. Azizpour, I. Leite, M. Balaam, V. Dignum, S. Domisch, A. Fellénder, S. D. Langhans, M. Tegmark, and F. Fuso Nerini, "The role of artificial intelligence in achieving the Sustainable Development Goals," Nature Communications, vol. 11, no. 1, p. 233, 2020. [9]        D. Rolnick, P. L. Donti, L. H. Kaack, K. Kochanski, A. Lacoste, K. Sankaran, A. S. Ross, N. Milojevic-Dupont, N. Jaques, A. Waldman-Brown, A. S. Luccioni, T. Maharaj, E. D. Sherwin, S. Khorram, K. Khattar, L. Neufcourt, C. Guha, P. Sheridan, S. Bhatt, H. Rolnick, and Y. Bengio, "Tackling climate change with machine learning," ACM Computing Surveys, vol. 55, no. 2, pp. 1–96, 2022. [10]      E. Strubell, A. Ganesh, and A. McCallum, "Energy and policy considerations for deep learning," in Proc. AAAI Conf. Artificial Intelligence, vol. 37, no. 13, pp. 15140–15148, 2023.