Sustainability and AI — ethics at the edge of innovation

As COP30 marks a decade since the Paris Agreement, we find ourselves confronting the opportunity to apply AI directly to climate challenges: optimising energy systems, reducing emissions and accelerating climate adaptation are just a few possibilities. But these potential benefits come with ethical and moral trade offs, including the growing carbon footprint of AI infrastructure, energy intensive data centres and large scale model training — and the invisible layer of human labour.

This article explores the dual ethical dimensions of AI in sustainability: its potential to drive climate positive innovation and its hidden environmental and human costs.

How smarter systems create greener outcomes  

The increasing impact of renewable energy around the world is creating an ever more complex energy system, and balancing the variable generation of energy with variable demand for it is creating new challenges for infrastructure and planning. Optimising energy systems for robustness and reliability is a prime area for AI to have a transformative impact; smart grids, smart energy systems and smart charging at both enterprise and domestic level can incorporate the charging of vehicles and even buildings and rely on effective AI systems.

Beyond energy production, AI can optimise logistics, ensuring that goods and people move in ways that reduce energy demand and environmental impact. This complements smart energy systems by reducing unnecessary consumption. The integration of Internet of Things (IoT) devices and big data analytics further enables predictive planning, more effective reuse and recycling, and minimisation of waste. These applications not only reduce emissions but also mitigate the environmental damage caused by resource extraction and disposal.

The Earth Charter provides a values-based framework to align such AI solutions with global social justice, intergenerational equity and broader sustainability goals. As AI becomes embedded in infrastructure and supply chains, ethical design and deployment must ensure that technological efficiency does not come at the cost of fairness or ecological integrity.

The carbon footprint of AI

Like other computing systems, AI has significant environmental impacts — from the raw materials needed to build the hardware to the energy needed to operate the systems.

For you

Be part of something bigger, join BCS, The Chartered Institute for IT.

With the huge uptake of generative AI, users’ requests must also be counted as part of the overall energy consumption, as well as ‘hidden energy costs’ such as resilience and network infrastructure. Even training AI systems carries a significant energy demand; training Bloom, a 176 billion parameter model, took 118 days and 433,196 kWh, which amounts to just over one million GPU hours. 

Google has attempted to analyse the full AI stack energy consumption and found figures exceeding what was previously thought (https://tinyurl.com/n7er77nv). Head of AI, Innovation and Energy Sector Digitalisation at the UK’s National Energy System Operator (NESO), Carolina Tortora, says: ‘The projected growth of data centres in Great Britain could add up to 71 TWh to the UK’s total annual electricity demand by 2050, as businesses continue to digitise and adopt cloud-based services.’ In 2024, the entire energy consumption of the UK was 319 TWh.

Moreover, NESO ‘strongly believe in the value that artificial intelligence can bring to Britain’s energy system, which is why we’re members of the AI Energy Council. The Volta programme aims to show how AI driven insights and automation can improve real time decision making in our control room — tackling future energy system challenges.’
Another significant demand is the water needed for cooling these systems, a problem shared with other cloud solutions. The UK Government Digital Sustainability Alliance (GDSA) estimates the current volume needed is 40 billion litres per year.

Then there is the human cost — the ‘ghost work’, or invisible repetitive human labour required to power machine learning (https://tinyurl.com/mrfeewkh). An ever-growing manual workforce is essential to performing the millions of hidden microtasks behind the scenes to create this smart technology.

The changing role of IT professionals

There are many different ethical frameworks for sustainable AI development, covering aspects such as transparency, lifecycle analysis and green computing standards.

IT professionals play a key role in shaping environmentally and socially conscious AI systems and practices, including using impact analysis to inform decision making when selecting the best AI tools for a task. Finding a sensible balance is critical. The challenge may well be one of ‘right-sizing’ rather than maximising; some AI projects or solutions may be faster or appear better, but their overall impact at scale can be harmful or wasteful. Different models must also all be considered — large language models versus small language models, and cloud solutions versus edge computing, for example.

Challenges also exist at solution design level, such as defining ‘performance’, which can often be a balance between accuracy, delivery speed and, of course, energy consumption. Design and development decisions must consider all the different options in the context of responsible AI.

AI for a resilient future

AI has the potential to support all of the wider Sustainable Development Goals and the Paris Agreement’s long term goals — something explored in BCS’ Getting Started with Tech Ethics: An introduction to ethics and ethical behaviours for IT professionals. However, that potential depends on IT professionals embedding sustainability in the core of their AI decision making. Ultimately, if the costs of AI are to be worthwhile, it must be delivered in a way that creates a more sustainable world for humanity.