Artificial Intelligence (AI) is changing the world as we know it. High school kids are getting it to write their assignments for them, our cars are using it to drive themselves, we now have AI chatbots and digital assistants on almost every device, and it helps run search engines like Google. Elon Musk is even making an AI-powered humanoid robot for factory work and to do his hoovering! There is almost a sense that, soon, everything will be automated by this seemingly magical technology. But, AI has a dark side, one that could threaten our precious Earth if left unchecked. But, conversely, it can also be used to save the Earth from our climate sins. So, is AI good? Or bad?
Let’s start with a super quick and simple explanation of how AI works. AI is based on simulated neural networks that function similarly to the synapses in your brain. These neural networks can be excellent at recognising patterns in data, or even creating data that follows a pattern. However, like a biological brain, they must be taught how to do this first.
To achieve this, they ingest vast amounts of data in a process known as training. For example, an AI that can identify cats and dogs in images will be trained on a dataset of images of cats and dogs. Simple AIs like this cat and dog detector only need a relatively small dataset to work correctly. But their data demands skyrocket for more complex AIs like self-driving cars, language models like ChatGPT and digital assistants!
However, for these complex AIs, data demands don’t stop after training. Once these complex AIs are trained and operational, they need to access yet more large datasets to function correctly. Take ChatGPT; it doesn’t store all of its knowledge in the AI; instead, it is stored in a massive remote data centre that it can access through cloud computing.
It’s these data demands of training and running AIs that are so problematic. Let me explain.
These data centres have racks and racks of high-speed storage and servers. All of this computing power takes a lot of energy to run, which in turn creates a lot of heat. This heat would easily damage the delicate circuits if left unchecked, so most data centres are cooled with energy-intensive air conditioning units. As such, a single data centre can consume the equivalent of 50,000 homes worth of electricity!
But it isn’t just data storage that increases AI energy demands. Training AIs requires a vast amount of computing power and, in turn, a vast amount of energy. In fact, training just one simple AI can use as much electricity as 100 US do in a year.
But what about the more complex AIs out there? Well, Bloomberg reporters recently discovered that Google’s AI branch is responsible for 10–15% of the entire company’s energy consumption! That means Google’s AIs currently use a massive 2.3 TWh per year! That’s almost as much energy as the country of Malta uses in a year, and Google’s AI branch, and almost every other AI company, are only set to grow exponentially from here.
All of this energy usage has two significant problems, current emissions and increasing demand.
Let’s be conservative and assume all of these data centres are in the US. Well, only 41% of US energy comes from clean sources like wind, solar and nuclear power. As such, each kWh of US energy emits, on average, 0.389 kg of carbon emissions. This means that Google’s AI branch is likely emitting around 894,700 tonnes of carbon dioxide per year! And that’s just one AI company; there are literally hundreds of them out there now.
As I said previously, the AI industry is set to grow dramatically over the coming years as it becomes both more capable, and able to take on more tasks, but also more integrated within our everyday lives. So, it isn’t hard to envision a future where the AI industry emits hundreds of millions of tonnes of planet-wrecking carbon dioxide, or possibly more, into the atmosphere.
AI advocates may say that this problem will be mitigated as our energy grids transition to cleaner sources, like wind and solar. In a sense, this is a problem that will solve itself. While this viewpoint is partially correct, it doesn’t consider the realities of our energy transition.
You see, renewables can only be built and deployed so fast; there are bottlenecks. For example, solar panel factories are limited to how many units they produce in a year, and there is only enough material supply and trained professionals worldwide to keep a certain number of solar panel factories open.
This means that AI’s growing energy demand could quickly reduce how fast renewables can replace polluting fossil fuel energy, as energy grids might be forced to keep old coal or gas plants open to meet demand. You can already see this in China as their expanding economy drives energy demand through the roof, so even though they are building more renewable infrastructure than any other nation (by quite some margin), they are still having to rely heavily on coal power. In fact, China’s coal power is having to expand to ensure it can meet demands, despite their climate targets.
So, AI does have the potential to both pollute and delay our crucial planet-saving energy transition, threatening the world as we know it. But conversely, it can also be used to help save the planet.
Many AI applications help make our society both more efficient and more resilient to climate change. For example, it can make climate models way more accurate, enabling us to predict where climate change will affect the most and helping us adapt to looming challenges. AI can help design more efficient mechanical designs, leading to more efficient cars and wind turbines. AI can be used to manage renewable energy across massive smart grids, balancing the delicate interplay of consumer demand and keeping grid-level batteries charged, which in turn leads to a far less wasteful energy grid. AI can even monitor agricultural conditions and how well crops and livestock are growing, enabling farmers to make better choices and dramatically boost efficiency. AI can even accurately detect and track deforestation, floods and methane leaks in real time, enabling us to act quickly to stop or mitigate these dangerous events.
In fact, AI can even optimise itself by finding ways to make its programming, the computer it runs off, and the data centres it uses to be more energy and time efficient.
So, is AI good or bad for the planet? Well, as always, with any tool, it depends on how we use it. If we let it permeate every corner of our lives, and don’t do anything to address the inherently high-energy demands, then it can definitely hold back our progress to net-zero, and our climate target will sail past us. But if we are aware of this, and put things in place to mitigate it, or prioritise the use of AI in areas which overall will increase our efficiency, then AI has the potential to be a potent instrument to save our world from ourselves.
But we have to take human and corporate nature into account. If the AI industry is set free, then startups and technology companies will try and force it into every aspect of our lives, looking to sell the promise of autonomy for a quick buck, and AI’s power demand will soar! What we need to do is find robust and legitimate ways to keep the AI industry accountable for its part in driving up energy demands. If we can do that effectively, then AI will become the climate saviour many want it to be. But as of yet, I’m not convinced we as a society can do such a thing.
Thanks for reading! Content like this doesn’t happen without your support. So, if you want to see more like this, don’t forget to Subscribe and follow us on Google News, Flipboard, Threads, TikTok, YouTube, and Twitter, or hit the share button below.
Sources: OilPrice.com, IEA, Penn Today, Cyber News, Columbia Climate School, EIA, Nature, DW, WE Forum, Carbon Fund, EU Parliament, Planet Earth & Beyond