Behind the chat box: The environmental impacts of artificial intelligence

When you use ChatGPT to give you a movie recommendation, the first thing on your mind probably isn’t how these generative AI models are affecting the environment. However, their overlooked environmental impact is widespread.

The massive data centers behind these AI chat boxes contribute to about 2% of electricity use worldwide. Beyond their impacts on electricity, data centers also contribute to significant water usage to maintain operable temperatures and emit carbon byproducts as they run 24/7.

Caroline Earls, a senior majoring in strategic communication, discussed her outlook on AI’s environmental effects.

“Climate change isn’t really alluded to AI, and it scares me that such a giant tool that everyone uses has negative environmental impacts. These impacts give me more reason to be scared of [AI],” Earls said.

Balancing those environmental negatives while discussing recent advances in AI’s energy efficiency is key for James Walden, a Miami University professor of computer science and software engineering and the director of Miami’s Center for Cybersecurity, who expressed a more positive outlook on combating AI’s energy consumption.

“There is an enormous amount of effort and research that’s going into making AI more power efficient. One way is through model distillation . . . in which AI models are hundreds of times smaller and hundreds of times more energy efficient,” Walden explained.

To visualize model distillation, Walden drew upon the contemporary example of Apple Intelligence — Apple’s upcoming 2025 launch of personalized artificial intelligence tools.

“To run [Apple Intelligence] on a phone in eight gigs of RAM is amazing, you know, going from having multiple data centers of computers to getting it down to fit on a single phone,” Walden said. “So there’s a tremendous amount of work in distilling models using just a few parameters.”

This distillation of data is also seen when going from data centers to one’s laptop screen. Less than a month ago, Miami’s Information Technology Services announced that Google Gemini, Zoom AI Companion and Webex AI Assistant are now available to everyone with Miami credentials.

However, these tools, especially when they are still in their learning and inference stages, contribute to lots of energy consumption that tends to fly under the radar. Just this year, Google stated in its Environmental Report that in 2023, its “total data center electricity consumption grew over 17%” in response to the recent AI boom.

“I personally never would have thought of AI and its environmental and spatial impacts because I would just think that it’s just a Google search engine,” Earls said.

But, as AI advances, so do we — especially in regard to our knowledge of AI and our ability to address AI’s environmental effects.

Enjoy what you're reading?
Signup for our newsletter

“I’m hopeful that AI offers a lot of opportunities to make things more efficient,” Walden said. “There’s a huge amount of people in all sorts of fields working to build more efficient AIs.”

Kwiatksk@miamioh.edu