How to Use AI Without Compromising Your Sustainability Values

We’ve all heard the warnings: Failure to embrace AI could cause you and your organization to fall behind. Yet AI data centers consume a great deal of land, energy, and water while also polluting the environment. Nonprofit staff and leaders are struggling to figure out how to balance these benefits and drawbacks. 

“The most common question I get: ‘Is it OK for me to use AI given its impact on the environment?’” says Anna Lerner Nesbitt, CEO of Climate Collective, a nonprofit focused on balancing technology use and protecting the environment. 

Nathan Chappell, founder of Fundraising.AI, agrees and says AI’s environmental impact is often a top concern nonprofit professionals express. Chappell is not surprised, he says, because nonprofit staff want to make the world a better place and worry that AI use conflicts with that goal. “We can’t just do good and do harm at the same time,” he says. 

To understand how individuals and organizations can reduce the environmental impact of AI, we consulted experts. Here’s their advice.

Use the simplest tool for the job.

“We’re not anti-AI at all; I think it has incredible value,” says Miriam Aczel, an environmental researcher at the United Nations University Institute for Water, Environment, and Health. “Avoiding AI at this point isn’t possible,” she says, “but you can control your own use and be smarter about your own use.”

The biggest downside of AI is its intense demand for electricity, Aczel says, which increases carbon emissions and water usage. Aczel, who recently published research on AI’s environmental impact, says it’s important to understand the types of AI. 

Generative AI, which creates new content, such as ChatGPT, consumes a lot of energy, but using AI to analyze information uses less. A common example of low-energy AI analysis is spam filters that sort email as it arrives in inboxes, Aczel says. That’s also an example of AI that is built into existing tech tools, which users are often unaware of.

Generative AI is an energy hog, especially when it creates images or video. According to Aczel’s research, a single AI image can require 1,450 times more electricity than an AI chat. A video can consume as much electricity as 200,000 simple AI analytical tasks. The mode matters, too: A deep-research request consumes far more energy than a basic query.

“The key is really being conscious about longer outputs, repeated generation, high-resolution images,” Aczel says. “A good rule of thumb is use the smallest, simplest tool that does the job.”

Users should use AI that is “fit for purpose,” so an AI tool won’t always be the first choice, Aczel says. 

“If you want to write a donor email and you’re trying to write it from scratch, AI could be useful,” she says. “If you just need to shorten a paragraph or check grammar, you might not need a large model. You could use a light model to help edit. And that’s not a very heavy use of AI because it’s not generating. It’s just shortening.”

Another way to reduce the impact of AI is with “eco-prompting,” or “zero-base prompting,” says Beth Kanter, an author and technology consultant. This generally means targeted prompts and specifying what you want, Aczel notes. If you want a three-sentence response, include that in your prompt so AI doesn’t generate a three-page report. 

“You can say, ‘Explain photosynthesis to me like I’m 5 years old and in two sentences,’” Aczel says. “That’s a really short, concise, specific prompt, and I think the answer you’ll get will be much closer to the answer that you’re looking for.”

Saying please and thank you is a waste, notes Aczel. “We’re not saying be rude to your AI models,” she says, “but just be conscious every single additional word can lead to additional electricity usage.”

This strategy for using AI is the approach environmental nonprofits are gravitating toward, says Apolinar Gonzales, executive director of the Climate Advocacy Lab. The organization recently conducted a survey of environmental nonprofits and gave the Chronicle a pre-release peek at the findings. Most environmental nonprofits are open to using AI but are “incredibly resistant” to using it to generate long texts, images, and video, he says.

“An overwhelming number of the folks we surveyed said they were using it for summarizing dense documents or research, doing analysis of data to come to conclusions, and develop summaries so that decisions can be made,” Gonzales says. 

Lerner Nesbitt adds that people must recognize that everything they do affects the planet. For example, she says, trans-Atlantic flights have a higher carbon footprint than AI use. “I don’t think an individual should have to feel guilty about experimenting with AI to understand what it is and see if it can help them do important work.” 

Build the environment into your AI policy.

Organizational policies have a big impact on a nonprofit’s environmental AI footprint, experts say. Start by acknowledging environmental concerns in AI policies, says Kanter. Sometimes staff who speak up are “labeled resisters,” she says, but concerns should be heard and integrated into policy guidelines.

“Welcoming all voices in that conversation is crucial,” Kanter says. “All of those perspectives should be honored.”

Organizational values should shape your AI policy, Kanter says. For example, one nonprofit she advised values “innovation, efficiency, [and having] options,” she says. 

In their policy, “to incorporate options, they said, using AI was a professional choice, not a mandate,” Kanter says. “Staff decided which tasks benefited from AI assistance and which stayed human. Choosing not to use AI for a particular task was viewed as a legitimate professional decision, not a performance issue.”

Nonprofit AI policies that address the environment, Kanter says, typically acknowledge the footprint and commit to using tools thoughtfully. In other words, she adds, not every task warrants a prompt.

Educate staff and choose tools wisely. 

Nonprofits should also educate staff about sustainable AI use, Chappell says. “Train team members on how to evaluate what model is appropriate for the task” so they don’t over-compute on a simple task, Chappell says. While one person’s impact may seem small, if it’s “compounded by 10 million nonprofit professionals in the country all prompting all day long, that’s a pretty big impact.” 

To further minimize negative effects, ask vendors whether the AI tools built into  software are analytical or generative, whether they run automatically, and if they can be turned off, Aczel advises. “There are a lot of places where it is running in the background that can be either reduced to a lower, simpler level or turned off altogether,” she says. “AI features running in the background can be massive. These scale up really quickly.”

Organizations should press vendors for energy use metrics, but Aczel and Lerner Nesbitt say few AI companies are open about it. When people want to buy large appliances like a refrigerator or stove, they can check federal Energy Star ratings to find the most energy efficient options. No federal guidelines exist for AI models.

“We don’t really have a universal nutrition label or energy label for AI, and it makes it hard to compare against the different tasks and tools,” Aczel says.

Once you have an AI policy that elevates environmental considerations, it’s important to revisit it regularly. It may not even last a year due to the fast pace of technology, says Lerner Nesbitt. “Living principles are much more useful,” she says. “You’ll need to update them and tweak them, as the sector evolves so rapidly.” 

In that vein, Kanter recommends organizing a small internal group to pressure-test policies and keep them in line with organizational values. For example, if the policy recommends using technology that’s most appropriate for the task but staff find that online meeting software automatically adds an AI notetaker to every meeting, that’s something they can flag. 

“Those groups are where a lot of learning takes place to figure out where’s the spot that more human judgment is needed,” Kanter says.

Eco-friendly AI models come with caveats.

Some AI models claim they are more energy efficient because they use renewable resources such as solar or wind power to fuel their technology. 

“There are lots of smaller tools that people are starting to use that are performing very comparatively to these large models,” Lerner Nesbitt says. “These groups are often very transparent on what electricity grid their models are trained on and also the energy and water footprints that their models are generating.” 

Some models that claim to be eco-friendly include GreenPT, Clairo AI, and Change Agent AI.

However, as Aczel’s research notes, it’s important to look at the overall impact, not just reducing carbon footprint. 

“We need a full accounting of the land impacts used in generating the electricity,” she says. For example, a large solar field may occupy a large area of land. “Not all renewables are equal,” she explains. ”Sometimes shifting from one type of electricity to another might move the burden from one footprint, say a low carbon footprint, to a high water footprint.”