The environmental impacts of AI: when doing good does harm

Something calls you to the work you do. Whether you have a very specific mission or just feel a general desire to help others, there’s reason you show up every day.

Fundraising is not usually that reason. Fundraising typically occupies a corner of your life because it’s necessary. It’s what pays for the good stuff, the life-changing stuff you really want to do. 

Fundraising is also where idealism (what you dream is possible) and realism (how the world actually works right now) can crash headlong into each other. Moral and ethical dilemmas are pretty much inevitable.

It’s been this way for as long as people have been trying to do good. The need for support and resources raises some challenging questions: How do you raise money? Who should you take money from? What should you do in exchange for that money? Should you ever compromise one of your ideals to secure resources you need to survive?

In other words: when do the ends justify the means?

The new kid on the fundraising block — Artificial Intelligence — raises old questions

Artificial intelligence (AI) is being touted as a cure-all for nonprofit fundraising and communications. The bright, shiny promise of AI is that it will raise you more money, reduce your fundraising workload, make your programs cheaper and more effective, and give you back some much-needed time.

Backed by the wealthiest institutions in the world, AI is everywhere. Whether you intentionally integrate AI tools into your office or your staff are just using them quietly on the side, AI is already in your organization.

The thing is, ease of use comes with a lot of hidden costs, including AI’s environmental impact. When you use it, you could be contributing to the social and environmental problems your organization is trying to solve.

Which begs the question: Is this this futuristic technology sending us backwards in time?

Through a sustainability and equity lens, AI looks old-fashioned

There are a lot of concerns around AI, ranging from the use of people’s work to train the models to the privacy issues that arise when you share confidential information with AI tools. But it’s the environmental impact of AI that looks the most like an old-fashioned natural resource grab.

The dirty secret of many artificial intelligence tools is that their environmental footprint — and yours if you use them — is enormous. ENORMOUS.

LLMs, the most popular day-to-day AI tools, seem to be the worst offenders.

When you use AI, your environmental footprint grows, big time

These LLMs are power-hungry, water-hungry beasts. It takes a huge amount of power to train the initial models and then process the millions of requests people submit every day. As it turns out, the machines of the 21st Century seem to gobble up many of the same resources as the machines of the 19th Century.

An ordinary ChatGPT interaction consumes about two cups of water. Use it to write one 100-word email for you every week for a year is like dumping 1 and a half water cooler jugs of water down the drain. It’s hard to imagine buying a few jugs of water in the supermarket then dumping them in the parking lot, but that’s basically how most LLMs work.

That same 100-word email uses roughly the amount of energy that 14 LED light bulbs consume in an hour. Most people don’t turn every light in their house on for hours for no reason, but that’s kind of what you’re doing when you use ChatGPT to answer a question that a search engine or a colleague or a book could have answered instead.

AI Adds Up

With millions of users, this resource-gulping adds up. All over the US, coal and gas projects are firing up to support the growth of AI. The organization behind ChatGPT told the US government it “needs” five new data centres. Each data centre would require more power than a city, meaning they’re asking for the power equivalent of five new nuclear power plants. Meanwhile, Microsoft has made a deal to restart the Three Mile Island nuclear plant to power its AI efforts.

The demands on water systems are just as alarming. Microsoft, Amazon, and Google are expanding data centres in places like drought-stricken Querétaro, Mexico, where local residents and businesses are struggling for survival.

Is this really what progress looks like?

Ends vs means

For decades, most countries have been moving towards cleaner, sustainable energy systems. The rush to inject artificial intelligence into everyday life is now turning the clock back forty years.

After years of progress, we’re going backwards. Because of AI, the world’s largest tech companies won’t meet their climate goals. Meta’s annual emissions have increased by 65% in two years. Microsoft’s annual emissions increased around 40% in the last three years. Google’s emissions increased nearly 50% in four years.

It’s a reminder of how fragile progress really is.

AI should make you think

Artificial intelligence might not have been at the top of your list of things to think about when you got out of bed this morning, but the practical and philosophical questions it raises are central to your work:

Do you have a sustainability policy that helps you manage your organization’s footprint? (And does that policy cover your use of AI and LLMs?) How does your organization navigate situations where the means to an end — like how you raise money — might undermine your core purpose?

AI isn’t the first issue that asks social impact organizations to think about their impact on the world. It won’t be the last. This is an opportunity to be clear about your organization’s social and environmental values and to show the world what responsible leadership looks like.

There may be a case for using AI in your organization, but it’s your job to define the why, when, and how. It’s your job to look at the inconvenient stuff others want to ignore, to understand the impacts of your choices, and to act wisely.

If we want the rest of the world to pay attention to the impact of their choices, then we need to do the same.