The impact of AI initiatives on the environment

Following on my recent post regarding the divine right to progress in AI, I'd like to take the time to highlight the environmental impacts that companies are making for the sake of progress. It is high time we bring some of these issues to light and ensure that more people are aware of the dangers we're creating as a result of this new technology. Don't get me wrong, I think AI is cool. I just think we can have cool things and not kill the planet or create a void of socioeconomic divide as a result.

Admittedly I was inspired by talking with a recent colleague of mine who brought this up at a local meetup. Check out Matt (Kelly) Williams and his work on Making Software Greener and the Green Computing Foundation. Another tip of the hat to an endless sea of people who inspire me to speak up and do something in the AI space. I wouldn't be writing without y'all!

Let's get into it.

Energy, between you and me

That's right, we have a soundtrack to this madness!

All jokes aside, energy (and as a result, carbon emissions) has been the talk of the town when it comes to AI. A single query to ChatGPT is equivalent to 20 minutes of light from a bulb. What about an image generation prompt? Consider your smartphone drained. We seem to have an issue of large-scale energy consumption and the idea that it will all be worth it, at some point, eventually. Sam Altman notes as much in his blog post: "The Intelligence Age."

Mind you, there is good coming from the recent AI push. Many innovative companies and individuals are making waves across industries using this technology. Notably in healthcare, there's real benefit that's possible with streamlined diagnoses, administrative improvements, and even wound recognition. My worry is not the benefits that AI can provide, but whether the sacrifices we give to achieve this progress will leave us with anywhere to celebrate.

It is not just carbon we're messing with.

Although carbon emissions is a worthwhile metric to note in terms of the impact compute is making globally, it's only part of the problem. A big facet of these conversations needs to be with one thing Coloradans know very well: water. There are significant reports when it comes to the impact AI training has on freshwater, like GPT-3 taking over 700,000 liters (185k gallons) of water to train. Worldwide, it's currently looking to be an estimated 200 billion gallons of water this year alone for data centers.

Another point: raw materials. Is it sustainable to be churning through silicon and other materials for AI chips, that of which take up 20% of the $450 billion dollar superconductor industry? From what I can tell, we're not even doing anything in the way of recycling to mitigate this resource consumption. I am concerned what will happen when only certain countries in the world have access to the materials needed to continue building more data centers and GPUs.

Are companies doing anything about this?

Big Tech is constantly pledging to go zero emissions or carbon neutral by 20230 or something, which is great but only part of the equation. What happens when these companies miss their mark? Are there any systems in place to incentivize them to care? Sort of. They could also be subject to lawsuits and "greenwashing" investigations, but I'm not sure that would solve the problem.

Another point worth considering: is there a need to lie when it comes to environmental impact?

There are also resources available that speak on ways you, as either a business leader or engineer, could be more mindful of AI usage in your organization. Google Cloud offers a Carbon Aware Computing for GenAI Developers that covers some actionable APIs and methods to somewhat optimize cloud compute for less carbon emissions. Microsoft has a learning path for AI for leaders in sustainability.

I had to go and find these resources myself after some digging, and I think that they are a good start. However, it speaks to the onus on the consumer to ethically source what they do. As a collective, society has a lot of power to make a difference in the climate crisis. This difference should start, namely, with businesses that can take more than Seattle's entire power grid worth of energy to operate. We can make more impact that way than I can by myself implementing these practices into my own life.

Conclusion

We need to be better. The climate crisis is not going to solve itself, and properly identifying the issue is the first step in a meaningful solution. It's clear that there isn't a meaningful lever for companies to be either incentivized or reprimanded in regards to how they handle climate impact initiatives.

All depressing talk aside, here's Cha Cha! The tenacity of his want to eat early is exemplified by his snaggletooth and bombastic side eye.