AI and the Environment
I decided to write this after having a fairly hefty debate with a friend about the environmental impact Artificial Intelligence (AI) is having on our world. I'm used to having these conversations about the ethics around AI but while I'm not ignorant to the impact, this was the first time I had really dug into a conversation around it.
The gist of it is that while AI offers some incredible benefits, it also comes with environmental costs we can’t ignore. From the energy it takes to power massive data centers to the resources needed to produce its hardware, AI has a footprint that’s bigger than most people stop to consider. Let’s break down some of the biggest environmental impacts of AI, what we can do to reduce them, and whether AI’s benefits could ultimately outweigh the damage.
Biggest Environmental Impacts of AI
Energy Use Training AI models requires a huge amount of electricity. Often, this electricity is generated by burning fossil fuels, which pumps a lot of carbon dioxide into the air. For example, training one large AI model can produce more CO2 emissions than driving a car for its entire lifetime. That’s a lot of energy!
Water Usage When AI servers run, they get hot—really hot. Cooling them down takes a lot of water. In fact, Microsoft’s water use went up by 33% between 2021 and 2022, partly due to AI research. And this is happening all over the world as AI research ramps up.
Hardware Production AI needs powerful hardware to run, and making that hardware isn’t exactly eco-friendly. Components like GPUs (the brains behind AI) are made using metals like cobalt, gold, and silicon. Mining these materials can cause pollution, soil erosion, and other environmental damage.
Electronic Waste AI hardware, like all electronics, doesn’t last forever. When it’s time to upgrade, a lot of this equipment ends up as electronic waste. Sadly, many electronics aren’t recycled properly, which can lead to even more pollution.
Other Impacts AI can also have unexpected effects, like making certain industries, such as fossil fuel production, more efficient. It can even encourage overconsumption by creating super-targeted ads that push us to buy more stuff.
How We Can Reduce AI’s Environmental Impact
Use Energy-Efficient Hardware One way to cut down on energy use is to switch to more energy-efficient hardware. GPUs, for example, use less energy than traditional CPUs when doing AI tasks. Also, multi-node servers (which let servers share resources) can lower energy use per server.
Cool Data Centers More Efficiently Instead of using traditional air cooling (which eats up a lot of energy), liquid cooling is a better option. It’s quieter, easier to maintain, and can cut down on energy use significantly. Yes, I know I've listed water consump as an environmental concern, but liquid cooling is more energy-efficient than air cooling. The key is using sustainable water sources and recycling, which can balance the environmental trade-offs.
Optimize Data Centers There are lots of ways to make data centers greener. For example, using free air cooling, which brings in outside air to cool servers, or upgrading to more efficient power supplies can help. Titanium or Platinum power supplies waste less energy, which makes a big difference over time.
Switch to Renewable Energy By powering AI data centers with renewable energy like solar or wind, we can reduce the amount of carbon emissions they produce. Many tech companies are already making moves in this direction, but more needs to be done.
Use Smarter Algorithms Instead of building every AI model from scratch (which takes a ton of energy), we can use pre-trained models that have already been built. Plus, new technologies like quantum computing and spiking neural networks could make AI training a lot more efficient in the future.
The Long-Term Outlook
The big question is can AI be a net positive for the environment? While it does have a big environmental cost right now, there are also ways AI could help solve some of our toughest environmental problems. For example, AI can optimize energy use in other areas, improve climate predictions, and even help with conservation efforts. But for AI to be a force for good, we absolutely need to make sure we’re reducing its environmental impact as much as possible.
What Individuals Can Do:
Everything I've outlined above is great. But unless you are a large business, government, or a developer of these tools, they don't focus on what you and I can do. And there are options for us!
Use AI Mindfully Be aware of when and how you're using AI. Avoid running large models or computations for tasks that don’t require them. For example, don't use powerful AI tools just for fun or trivial tasks if simpler methods will work.
Choose Simpler Models for Simpler Tasks Not every task needs the most advanced AI model. Opt for smaller, more efficient models when handling less complex problems, like basic language translations or text summaries, to save energy.
Limit AI Use to Essential Tasks Prioritize using AI for tasks where it adds real value. This reduces unnecessary computation and energy consumption. Consider if you really need to use AI or if traditional methods would suffice.
Use AI During Off-Peak Hours Some cloud computing services offer off-peak usage discounts or lower environmental impact due to reduced demand. Schedule heavy AI tasks during off-peak hours to take advantage of less strained energy grids.
Choose AI Providers with Sustainable Practices Some AI platforms and cloud providers prioritize sustainability by powering their data centers with renewable energy or optimizing efficiency. When using AI, choose services that have a proven commitment to reducing their environmental impact.
Reduce the Frequency of Retraining Models If you're working on AI-related projects, reduce the frequency of model retraining. Retraining models too often, especially with minimal data updates, uses unnecessary energy. Instead, schedule model updates only when the new data significantly enhances performance.
Experiment with 'Relaxed' Generation Methods Some platforms allow users to tweak AI model settings for faster, less resource-intensive outputs. If your task doesn’t require high precision, opt for these relaxed methods to cut down on energy use.
Share Pre-Trained Models and Use Community Resources Instead of training models from scratch, look for pre-trained models shared by the AI community. These can be fine-tuned for your needs without expending the resources to build an AI from the ground up.
In short, AI is changing the world, but it’s also guzzling energy and resources along the way. By focusing on more energy-efficient hardware, optimizing data centers, using renewable energy, and asking ourselves what we as individuals can do about it we can start to reduce its footprint. AI has the potential to help solve environmental challenges, but only if we take its own environmental cost seriously. The goal should be to make AI part of the solution, not just another problem.
UN Environment Programme. “AI Has an Environmental Problem – Here’s What the World Can Do About It.” United Nations Environment Programme, 2023. https://www.unep.org/news-and-stories/story/ai-has-environmental-problem-heres-what-world-can-do-about.
HBR Staff. “The Uneven Distribution of AI’s Environmental Impacts.” Harvard Business Review, 2024. https://hbr.org/2024/07/the-uneven-distribution-of-ais-environmental-impacts.
Wikipedia Contributors. “Environmental Impacts of Artificial Intelligence.” Wikipedia: The Free Encyclopedia, 2023. https://en.wikipedia.org/wiki/Environmental_impacts_of_artificial_intelligence.
HBR Staff. “How to Make Generative AI Greener.” Harvard Business Review, 2023. https://hbr.org/2023/07/how-to-make-generative-ai-greener.
Yale Environment 360. “Can We Mitigate AI’s Environmental Impacts?” Yale School of the Environment, 2023. https://environment.yale.edu/news/article/can-we-mitigate-ais-environmental-impacts.
MIT Sloan Management Review. “How Developers Can Lower AI’s Climate Impact.” MIT Sloan Management Review, 2023. https://sloanreview.mit.edu/article/how-developers-can-lower-ais-climate-impact/.