Is AI Bad for the Environment? A thought For Earth Day

I have given the impression that I’m not a fan of AI. In fact, I find as a photographer that it can be a very useful under the right circumstances. And I promise, I’ll write about that soon. 

But not today. 

Today, I’d like to talk about the environmental impact of AI. Right now, it's awful, and getting worse…but there’s a silver lining.

According to Scientific American, when OpenAI trained ChatGPT, the power required to run the program produced the equivalent of around 500 tons of carbon dioxide. As of February, ChatGPT’s generative AI uses enough energy to power 33,000 homes a month. Midjourney, which is used to create AI-based images that look a lot like photographs, uses even more computing power and energy resources because there is so much more data to process.,

In other areas, ExxonMobil used Microsoft’s Azure platform, which relies heavily on AI, to optimize fuel extraction operations. These fuels will be burned, and create more CO2. Ironically, carbon foot print for optimizing energy resources is large.

Infrared photo of lake, Edison, NJ, 2022 © Mason Resnick

Big Data? Big Costs

Processor-heavy operations, such as AI-driven generative fill or creation of photo-realistic AI images, use a lot of energy. These features are becoming more popular, which means the number of generative fill requests has caused a rush of demand to power the data centers that run AI.

According to Nature magazine, AI data centers need lots of fresh water to cool their processors and generate electricity. A lawsuit by residents of Des Moines, Iowa, claimed that their local AI data center used 6% of their district’s water, causing major increases in the cost of both power and water to the Des Moines communitie.

According to the Financial times, Sam Altman, CEO of OpenAI, warned that future AI would consume so much electricity it would require an energy breakthrough, such as nuclear fusion, to power it.

In fact, while crypto uses a boatload of energy, AI, right now, is even worse. The bottom line, according to MIT Sloan Management review, is that the AI industry could soon be one of the largest contributors to carbon emissions if trends continue.

Why Is This Happening?

An oversimplified explanation: Whenever you give an AI prompt, it gets sent to a computer somewhere else--not on your computer. The request is processed, and the result is sent back to your computer. The processing may be spread across multiple databases as AI scours the internet for relevant content: words, images, etc. All of this comes at a price, which is in the form of energy use. Even if the AI tool you are using is free to you, someone is paying for it.

A silver lining?

On the other hand, perhaps AI can help scientists figure out how to deploy nuclear fusion to run AI, as well as other ways to reduce its use of energy.

In fact, the AI industry HAS to tackle this “elephant in the room” before it costs more communities in increased energy bills due to demand, and plays even more havoc with the environment.

But here’s the good news. Google has used machine learning from DeepMind to help reduce energy use for cooling its data centers by 40 percent, according to the Financial Times.  Other major data users such as Amazon are exploring ways (with the help of AI) to reduce their carbon footprint. The industry is clearly aware of the problem, even though it is currenty lagging behind the growth in demand.

Governments are looking at regulations that will help move the AI world to a more sustainable approach.

As photographers, what can we do? I know it’s tempting to play with AI a lot. It’s the new crayon in the box. But think about its environmental impact and use it sparingly. Be aware that your experiments in seeing (some may call it "ripping off other photographer's work without compensating them--but thats another topic for another time") may be causing some community’s electric and water bills to skyrocket. 

Cyberspace can sometimes feel separate from the real word but in this case, it has real-world consequences. Understand that the link between your computer and real-world impact is real. Compute responsibly.