The use of AI is more wasteful than the aviation sector. But we don’t know exactly how much, because the AI companies keep it a secret. To create enough new energy, the way is now being paved for nuclear power, which AI companies are investing heavily in. Here’s what you can do to reduce your consumption.
‘One prompt on ChatGPT is the same as ten searches on Google. Training an AI model produces the equivalent of 300 return flights between New York and San Francisco and five times as much pollution as a car produces in an entire lifecycle. We don’t need AI art, we don’t need AI shopping lists, we don’t need AI-powered cars, we don’t need ChatGPT, Gemini, Groq or Dall-E or any other revolutionary technology that already exists inside our human brains. We need the planet.’
This is what American activist Matt Bernstein wrote on Instagram recently, according to the podcast Hard Fork, accusing AI of being one of the causes of the catastrophic wildfires in Los Angeles. While there are certainly a myriad of causes for the fires, he hit the nail on the head for hundreds of thousands of people who liked and shared his post, before it was taken down.
For many people, the speed with which generative AI is being introduced into businesses, government agencies and private homes is mind-boggling. Not only because the new services are built on top of other people’s work without permission (violating their copyright), because there are inherent flaws in the services as they are probabilistic machines, or because they are widely misused along with voice cloning and deep fakes to manipulate and abuse other people. But perhaps most of all because they are water and electricity vampires. They are swallowing our energy ressources.
Bernstein’s numbers probably don’t hold up. There are an incredible number of figures in the media that refer to researchers who guess at how high the consumption is. For example, DR wrote that a prompt could be up to 50 times as energy-intensive as a traditional search. But we have no idea how much more AI wastes, because AI companies like Microsoft, OpenAI and Google are so-called black boxes and they simply keep it a secret. They admit that their energy consumption has increased dramatically in recent years – for example, Microsoft’s water consumption increased 34% between 2021 and 2022. And they recognise that they can no longer meet their climate targets, but they don’t tell us how their generative AI technologies are making a mess. We know that generative AI requires lots of data processing and thus lots of energy in data centres and lots of water to cool those same data centres. We also know that AI as an industry is probably dirtier than the aerospace sector.
The aviation sector is estimated to be responsible for just over 2% of our global carbon footprint, while the energy consumed by the world’s data centres accounts for 2.5-3.7% of global greenhouse gas emissions, according to a World Economic Forum report. But unlike airplanes, there is no transparency about the environmental waste of AI companies’ data centres, and unlike airplanes, there is little scolding of AI for its waste. The upcoming EU AI Act will likely require these figures to be disclosed, unless the tech lobby gets it completely watered down in the negotiation process of a common understanding of the new law, an AI Code of Practice.
Machine Learning Led The Way
We have been using AI for many years. AI is the broad technology in computer science that aims to create intelligent machines that can perform tasks that typically require human intelligence. A subset of AI is machine learning, ML, which we have had for many years. ML is a technology that can recognise patterns and is used to make predictions. It’s used in healthcare and industry, for example with energy optimisation, so we become smarter in our resource consumption. Then, we have generative AI which is trained on vast amounts of data and uses machine learning to create something new, but unlike machine learning, which always requires human involvement, the goal of generative AI is for machines to operate without human involvement.
AI – or ML – is already being used to do some good for the climate, and there is certainly plenty of potential to do even more. But for AI to actually save our planet, as some technology optimists believe, is more sales pitch than reality. Not least because generative AI is so thirsty and so rapidly expanding.
Some heavy users of generative AI tend to say that they’ll just eat less beef or fly less so they can prompt more with a clear conscience. But that argument doesn’t add up, because generative AI is a new thing that we didn’t realise we needed and that we could do without to take better care of our planet.
Paving The Way for Nuclear Energy
AI companies like Microsoft, Amazon, Nvidia and Google are busy finding solutions to satisfy their energy thirst. Current green energy production using solar and wind is not enough, which is why the idea of nuclear power has been revived. They are thus partnering with energy companies, buying into them and pushing them to revive decommissioned nuclear power plants and build new ones.
At the same time, the CEOs of the same companies, the tech billionaires, are investing in new energy sources, for example Sam Altman, reportedly worth around a billion dollars, is focusing on advanced fission through Oklo, a startup that aims to commercialise microreactors that can run on nuclear waste. While Bill Gates, with a net worth of over $100 billion, has founded TerraPower, a company developing next generation nuclear reactors, according to reports from Yahoo.
Nuclear power is now considered a sustainable energy source, apart from the fact that we don’t know how to safely get rid of the ways, but nuclear power has been set on pause for many years due to concerns about nuclear scandals such as Chernobyl and later Japan.
What Can You Do?
Should we be ashamed of prompting? Yes and no. Yes, because humans are lemmings and when some do something, most others follow suit. No, because it’s a structural problem and it’s not fair to blame individuals. The crazy thing is that AI companies have been allowed to launch so many environmentally harmful services unhindered in just two years, and most are just jumping on the bandwagon for fear of losing relevance and competitiveness.
If you suffer from AI shame, there are things you and your company can do, and here I have been inspired by, among others, Rasmus Steiniche from Neuro Space, who teaches AI and the environment:
- Use smaller models as much as possible as they often don’t have access to unlimited data centre power.
- Use traditional search engines first and see if you get a response so you can save prompts.
- Don’t use generative AI as a calculator or translator when you can use other more environmentally friendly services. For example, deepl.com is an excellent translator.
- Use pre-trained models – build on top of others. The base model is the same and you put an application model on top. It’s the training that costs a lot of C02.
- Use the European base model Mistral instead of American counterparts like ChatGPT/CoPilot. It’s good and it’s regulated by EU law, so it’s more likely to comply with EU regulations and future environmental requirements.
- Choose a cloud provider where you can select a region that is low on C02. For example, French Scaleway and American Google are examples of cloud services that give the environment extra focus.
- Avoid duplicate datasets and processing.
- Keep track of data lifecycle and deletion policy.
- Use sustainable storage – i.e. retrieve data once in a while rather than all the time.
- Have good file formats and compression algorithms.
Translated partly with the help of DeepL.com (free version)
Photo: Markus Spiske Unsplash.com
This article was first published in Danish in Copenhagen Review Communication
Read more: White Paper for a Global Sustainable Development Agenda on AI and AI’s carbon footprint – how does the popularity of artificial intelligence affect the climate?