You might be surprised to learn that AI is already proving to be a powerful weapon in the fight against climate change. The UN, for example, has incorporated AI into its UNOSAT satellite imagery center – using algorithms to analyze images of flooded areas and assess where disaster response teams are needed.
AI can also help to preempt future disasters and help vulnerable communities anticipate what may be coming their way. Indeed, the UN works with vulnerable communities in Burundi, Chad, and Sudan – using AI to analyze past environmental change in displacement hotspots and build future projections.
But isn’t AI, with its heavy energy usage, adding to the climate problem, you ask? Well, yes, that’s also true. (I never said there was a simple answer to the title question!) One of the biggest concerns is that with the rise of generative AI tools like ChatGPT, the energy cost of AI is only going to increase.
Back in 2019, researchers at the University of Massachusetts Amherst decided to look into the carbon footprint of large AI models – of the sort that power today’s tools like ChatGPT. They found that training a single AI model can emit five times the lifetime emissions of the average American car (including the manufacture of the car).
That in itself isn’t ideal. It's not a disastrous number, but it’s not ideal. However, things have moved on drastically since then. At the time of the study, OpenAI’s GPT large language model was GPT-2. Nowadays, we’re up to GPT-4 – meaning the large language model has grown a whole lot, well, larger and more advanced. GPT-2 was trained on 1.5 billion parameters (the number of variables fed into the model), but GPT-4 reportedly has a whopping 1.8 trillion parameters. That’s an enormous leap. Naturally, creating such a vastly bigger model uses more energy than the 2019 study found.
And, let’s not forget, that’s just the energy needed to train the AI. There’s also the energy associated with people actually using these tools. Asking ChatGPT a question uses more energy than a standard internet search, and this, across the approximately 180 million users that ChatGPT has amassed so far, really adds up. Working out the energy cost of these conversations is tricky, but some estimates suggest that ChatGPT searches could use as much power as 33,000 American households each day. Each day. That’s a staggering estimate. Each conversation with ChatGPT also consumes roughly 50cl of water (a small bottle).
If tools like ChatGPT, Google Bard, and Microsoft’s Bing AI chatbot continue to amass users, future numbers will undoubtedly be even more alarming.
Well, it’s not quite that simple. Because, while generative AI tools are well on their way to being major sources of carbon emissions, they can also help us find the solutions needed to solve our biggest climate problems.
There are multiple ways generative AI can boost efforts to tackle climate change. It can be used to better forecast and manage energy demand, for example. It can be used to generate more accurate forecasts about future energy production, allowing providers to allocate resources more effectively. It can be used to design better renewable systems – systems that are more efficient and scalable. And it can be used to design better carbon capture and storage solutions – where emissions from factories and other industrial processes are captured and stored in the ground instead of being released into the atmosphere.
Of course, this doesn’t offset the enormous energy cost associated with generative AI (and AI in general). We absolutely need AI companies to commit to reducing their carbon footprints and invest in sustainable AI. But we can’t look at the climate cost of AI without also recognizing its potential benefits.
The bottom line is that the climate crisis is one of the most pressing problems that humanity has ever faced. And generative AI is one of the most powerful technologies we’ve ever had access to. Bring the two together, and we can hopefully arrive at some innovative solutions and, ultimately, aid the fight against climate change.
