Generated AI is committed to helping to solve everything from climate change to poverty. But behind every chatbot response there is a deep environmental cost.
Today's AI technology requires large data centers located all over the world. This completely draws a huge amount of electricity and consumes millions of liters of water to keep you cool. According to the International Energy Agency, by 2030, data centers are expected to consume as much electricity as all of Japan, with AI likely to be responsible for 3.5% of global electricity usage.
The continuous massive expansion of AI use and its rapidly growing energy demand make it much more difficult for the world to reduce carbon emissions by switching fossil fuel energy sources to renewable electricity.
So we are left with pressing questions. Can we take advantage of AI without accelerating environmental collapse? Can AI really be made sustainable?
We are at a critical time. Environmental costs for AI are accelerating, and largely unreported by the companies involved. What the world next can determine whether AI innovations align with our climate goals or weaken them.
At one end of the policy spectrum is a path of self-satisfaction. In this scenario, tech companies are unchecked and unchecked, expanding data centers and reviving private nuclear microreactants, dedicated energy grids, and even coal plants.
Dobresum / Shutterstock
Although some of this infrastructure could be run on renewable energy instead, there is no binding requirement that AI must avoid using fossil fuels. Even if more renewable energy is installed to power AI, they may compete with efforts to decarbonize other energy use. Developers may advertise increased efficiency, but these are quickly engulfed by the rebound effect. The more efficient AI, the more it will be used.
The other end has more fundamental possibilities. Global moratoriums or complete restrictions on the most harmful forms of AI are similar to an international ban on landmines and ozone-depleted substances.
Of course, this is politically unlikely. The nation is competing rather than suspending it to control AI troops. The global consensus on the ban is, at least for now, mi-killer.
But between self-satisfaction and prohibition, the windows close quickly due to decisive, targeted actions.
This can take many different forms:
1. Essential environmental disclosure:
AI companies can report the energy, water and emissions used to train and use the models. Having a benchmark helps you measure progress while increasing transparency and accountability. Some countries are beginning to impose larger corporate sustainability reporting requirements, but there are significant variations. Although mandatory disclosure alone cannot reduce direct consumption, they are important starting points.
2. AI Services Emission Label:
Just as the carbon emissions of restaurant menus and supermarket produce can lead people to less impactful options, users can be given the opportunity to know digital choices and the footprint of AI providers, including efforts to measure website carbon emissions. In the US, one of the nation's most recognised environmental certifications, the Blue Energy Star label helps customers choose energy-efficient products.
Alternatively, AI providers could temporarily reduce functionality and explain the various levels of available renewable energy.
3. Usage-based pricing related to impact:
Existing carbon pricing is intended to ensure that heavy users pay their environment share. Research shows that this is most effective when carbon is priced across the economy of all companies, not only targeting individual sectors. However, they rely heavily on digital technology providers who take this environmental burden into consideration in the first place.
4. Sustainability Cap or “Budget Calculation”:
This is particularly targeted at nonessential or commercial entertainment applications. Organizations may limit employee use as well as heavy office printing and methods of restricting corporate travel. As businesses begin to measure and manage indirect supply chain emissions, new business policies may be required for the energy and water footprint of AI use.
5. Water management requirements for areas with water stress:
The simple regulation here is to ensure that AI infrastructure does not uncheck the local aquifer.
Market power alone cannot solve this. Sustainability does not emerge from goodwill or tricks of clever efficiency. Enforceable rules are required.
Consumers are not fully aware
Consciousness helps. However, it is naive to expect individuals to self-regulate in a system designed for ease of use. “Use AI only if necessary” might be something like “Don't print this email” 10 or 2 years ago.
awstoys / shutterstock
The world is building a future that drives AI that consumes like the industrial past. Without guardrails there is a risk of creating convenience technologies that accelerate environmental collapse.
AI will one day solve problems we couldn't do, and our concerns about discharge and water seem trivial. Or maybe we aren't going to worry about them.
The way we engage with AI now blindly, carefully or critically – shape whether it serves or undermines a sustainable future. Policymakers should be thought carefully through regulations and treat AI like other highly profitable resource-intensive industries.
I don't have time to read about climate change.
Instead, get weekly summary in your inbox. Every Wednesday, the Conversation Environment Editor writes Imagine. This is a short email that will deepen a bit about one climate issue. Join over 45,000 readers you've subscribed to so far.
