If you care about the climate, should you be anti-AI?

💡 This article is featured in The Definitive Reading List: illuminem’s Recommended Sustainability Classics

The techno-optimism surrounding the surge of Artificial Intelligence (AI) is palpable. Enter however a classroom of sustainability Master students, as I regularly do, and you will hear voices of complaint against what is perceived as a massively energy-intensive Silicon Valley toy with limited tangible upside for humanity’s pressing social and environmental challenges. Recent reports that data centres could consume as much electricity as Japan by 2026, and globally could require as much as half of the UK’s fresh water use by 2027, lend evidence to this concern. Consequently, several jurisdictions are considering curbs on new server farms, including Germany, China, Ireland and Singapore. A former French minister went as far as suggesting to ration access to the internet. If you care about climate action, should you be anti-AI?

Two considerations could help answer this question. First, it is important to recognise that the real promise of AI, as a General Purpose Technologies, lies not in offering an immediate fix to all our predicaments but in accelerating technological progress, including in the realm of environmental sustainability. In the context of climate mitigation, the early applications of AI have already begun to demonstrate its potential. For instance, making air travel more fuel-efficient, developing faster electric batteries, or helping reduce the reliance on critical raw materials, essential for the green transition but often associated with significant environmental and ethical concerns. Recently, AI helped solve a crucial hurdle related to plasma containment, representing an important step in the long road towards safe fusion energy use. 

AI could help with climate adaptation too. Consider the fact that the spread of zoonotic diseases is a crucial risk stemming from climate change, and that AI is being deployed to accelerate the development of vaccines. Or that extreme weather events will become more frequent, with AI helping improve weather forecasting allowing life-saving early warnings. 

The second fundamental consideration relates to energy use. Many national decarbonisation strategies include plans to reduce energy consumption. However, as I’ve argued in a recent paper with energy economist Roger Fouquet, energy sobriety was always going to play a junior role in decarbonisation strategies vis-à-vis the imperative to foster innovation. If history is of any guidance, it is worth noting that declines in energy consumption are very uncommon, short-lived, and rarely happen by design. In line with this reality, the International Energy Agency expects sobriety to contribute just 8% to CO2 reductions in its roadmap towards Net Zero by 2050. Innovation offers the only credible pathway to permanently decouple economic activity from climate change and environmental degradation. When assessing the trade-offs between a short-term increase in energy use of AI and a medium-term boost in innovation, this element should be taken into account. 

We are in the early stages of the AI revolution, implying that the technology is still being fine-tuned. If clear incentives are set right now, innovation will be designed also to maximise energy efficiency, prioritise more efficient models, or minimise cooling needs. Indeed, while a command on OpenAI’s ChatGPT requires ten times the electricity of a Google search, it has been shown that it is possible to build a large language model similar to ChatGPT-3 with much lower emissions. 

Which brings us to the importance of regulation. Rather than opting for outright bans to the construction of data centres, clear incentive structures should be put in place to steer AI innovation towards solutions that are not only commercially viable but also aligned with sustainability. This could start with clear environmental reporting requirements for AI companies and could extend all the way to legislation requiring only renewable energy sources be used for new data centres. Aware of the problem, tech behemoths are turning to AI itself also to improve the efficiency of their data centres, or pledging to make them carbon neutral by 2030. To do so, tech companies are developing strategic partnerships with energy companies, be they nuclear, geothermal, solar and batteries. But voluntary measures should be complemented by regulation. Something both the US and the EU are considering in recent bills. 

Electrification was always expected to be a crucial component of the green transition, and therefore demand for electricity was inevitably set to increase as fossil fuels get progressively phased out. But part of the current predicament originates from the fact that it can take as little as one year to build a data centre, and up to five years to build renewable energy facilities. This mismatch is aggravating the climate cost of AI, and in most cases is the result of poor regulation. 

From a climate perspective, AI should neither be glorified nor vilified. Alone it cannot install the necessary solar and wind capacities, promote a circular economy, or restore degraded ecosystems and biodiversity. These challenges require action by consumers, businesses and governments navigating trade-offs between living standards, energy security and sustainability. But AI can augment human ingenuity in finding a solution to these wicked problems.