Generative AI and sustainability: Promise and paradox

This article is authored by Shailesh Tyagi, partner and leader, Climate Change & Sustainability, Deloitte South Asia.

Generative Artificial Intelligence (GenAI) highlights a key paradox of the modern age. While it enables rapid progress in climate action and supply chain transparency, it is also highly energy intensive. The real question is not its potential to support sustainability but whether it can deliver a net positive impact.

GenAI (Representational image)

There are two ways to use GenAI for creating a sustainable future. First, “GenAI for sustainability” applies technology to climate risk modelling, extreme weather prediction and optimisation of energy, water and logistics systems. Second, it enhances internal sustainability operations through Environmental, Social and Governance (ESG) reporting, compliance, benchmarking and supply chain due diligence. While both are valuable, they differ in maturity and risk.

GenAI is transforming climate intelligence by shifting from iterative simulations to rapid climate scenario generation. This enables faster, more localised climate risk assessments and makes climate stress testing more accessible using hydrological models and predictive analytics.

GenAI can also allow organisations to produce a breakdown of complex information of their footprint, as illustrated by Deloitte with a global nutrition, health and beauty leader, helping improve the accessibility and usability of sustainability analytics for decision-making. Examples such as these highlight the strong impact of hybrid solutions that combine AI with human expertise.

However, GenAI is most effective as an augmentation tool rather than a replacement for human judgment.

GenAI is also useful in helping users synthesise analytics in a format understandable by decision-makers. Climate challenges and sustainability in general often require integrating complex climate data, infrastructure dependencies and scenario planning. This challenge can be addressed by using GenAI as an intermediary that helps translate complex analytics into actionable insights. GenAI thus complements analytics rather than being a substitute for it.

Despite the benefits, GenAI carries a significant environmental footprint. It has been estimated that the energy needs of AI-related data centres will soon surpass the total power consumption in individual countries. For instance, India can expect a sharp increase in energy consumption related to the deployment of AI systems. In addition to energy, water is a critical resource required to train large-scale models, posing challenges in water-scarce regions.

Carbon intensity is another important factor to consider. The amount of carbon dioxide related to GenAI is entirely dependent on the energy mix used. Even in the best-case scenario where GenAI queries are highly energy-efficient, billions of queries per day mean that there would still be sizable emissions. The challenge is that GenAI can reduce emissions within businesses while simultaneously contributing to a rise in global emissions.

While increased productivity is certainly the key selling point of GenAI, certain challenges, such as biases, misuse, misinformation and unclear ownership, should be accounted for. Cybersecurity concerns related to data security and privacy should also be addressed due to the necessity of large datasets for training models. Lack of transparency prevents users from tracing errors and also from understanding how outputs are generated. Finally, with an increasing number of autonomous systems being deployed, establishing appropriate controls becomes essential.

Sustainability principles should underpin the development and operation of GenAI to ensure it remains credible. The focus should be on embedding sustainability from early-stage development onward, through a “sustainability by design” approach. Both efficiency and accuracy should become one of the key measures of success. Energy consumption, carbon emissions and water usage must be among the main performance indicators. Instead of depending on large models, using the right combination of technologies with data optimisation for model training could result in less computational power requirements with better accuracy in outputs. Similarly, infrastructure choices, including the location of data centres, should be compatible with environmental goals.

It is critical to focus on “Green Data centres” that are powered by renewable power, optimised to improve efficiency. A stronger emphasis should be placed on ethical AI practices to improve data security and privacy, and to reduce bias. In addition, implementing governance mechanisms for monitoring and reporting is imperative for developing sustainable Gen AI, ensuring accountability and auditability. This helps to unleash the full potential of Gen AI in a safe, transparent and environmentally friendly manner.

As GenAI continues to scale, efforts must also be undertaken to address its drawbacks. True pioneers in the field will ask how GenAI can be used to improve sustainability in a responsible manner. As GenAI continues to mature within the sustainability landscape, there remains considerable opportunity to influence how it develops and scales responsibly. There is only a limited window of opportunity before meaningful change becomes significantly more difficult. While laws protecting data and intellectual property rights do apply, a comprehensive regulatory framework for GenAI has yet to emerge. Responsible utilisation of GenAI must become a priority of all stakeholders.

(The views expressed are personal)

This article is authored by Shailesh Tyagi, partner and leader, Climate Change & Sustainability, Deloitte South Asia.