The Hidden Costs of AI: Are the UK's Tech Ambitions Sustainable?

The UK is pushing ahead with ambitious plans to become a global leader in artificial intelligence (AI). While the potential benefits of AI are vast, there's a growing concern that the environmental costs are being overlooked.

Today, the UK government aims to "unleash AI" to boost economic growth and improve public services. The AI Opportunities Action Plan includes plans to create "AI Growth Zones" to stimulate development and jobs, integrate AI into the public sector for greater efficiency, use AI to improve infrastructure, such as identifying potholes and provide AI tools to teachers and small businesses. While these plans are intended to drive progress, they risk exacerbating environmental problems if not managed sustainably. There needs to be a greater focus on the full life cycle of AI, including software and hardware, to understand the complete environmental footprint.

Emissions

AI systems, especially large language models (LLMs), require immense computing power, which translates directly into high electricity consumption. Much of this electricity is generated from fossil fuels, contributing to greenhouse gas emissions. Data centres, which house the servers for AI, consume vast amounts of electricity and this demand is projected to increase exponentially. The International Energy Agency estimates that by 2026, data centers, cryptocurrency, and AI could use 4% of global electricity, which is equivalent to the annual electricity consumption of Japan. Training AI models can produce significant carbon emissions. For example, training the GPT-3 LLM produced around 500 tons of carbon dioxide. A study estimated the carbon footprint of training an early LLM at about 300,000kg of CO2 emissions which is comparable to 125 round-trip flights between New York and Beijing. The "inference" phase, when AI is actually used, also has a carbon footprint, although it's less than the training phase, it is still significant when scaled up to millions of users. The generation of electricity from fossil fuels also results in local air pollution and thermal pollution.

Resources

Beyond electricity, AI's environmental impact includes the resources used to create hardware and store data. The production of servers, GPUs, and other hardware requires substantial amounts of raw materials like cobalt, silicon, and gold. The mining of these materials can cause soil erosion, pollution, and harm to biodiversity. The disposal of outdated hardware generates electronic waste, which is a fast-growing waste stream that contains hazardous chemicals that can contaminate soil and water supplies. Only 22% of e-waste is recycled properly. Data centres require water for cooling, especially in hot climates, which can lead to the evaporation of significant quantities of freshwater. The global demand for water resulting from AI may reach 4.2–6.6 billion cubic metres in 2027.

Wider Impacts

The impacts of AI extend to wider societal issues. AI-driven advertising can promote consumerism and the purchase of fast fashion, leading to more waste and emissions. The use of AI in e-commerce can also encourage rapid and frequent delivery of goods which increases waste and emissions. AI's environmental costs are also not evenly distributed and can disproportionately impact vulnerable regions. For instance, data centres located in regions that rely on fossil fuels for energy contribute more to air pollution than data centers that use renewable energy sources. Additionally, data centers in water-stressed areas put added pressure on local freshwater resources. The decisions made by AI systems can be biased due to inaccurate training data. Generative AI models can also be used to create and spread misinformation.

Conclusion

The UK's ambition to be an AI superpower must be tempered with an understanding of AI's environmental impact. Their plans need to prioritise establishing methods for measuring AI's environmental impacts, so that the public is aware of the trade-off they are making in the wider adoption of AI. They should also require companies to report the environmental impact of their AI products and services. Research on energy-efficient AI algorithms and promote the use of green data centres and renewable energy sources should be a key feature.

It is essential that the UK develops a strategy for AI development that aligns with environmental sustainability so that the country's ambitions for AI will not come at the cost of the planet.