Are we failing with AI too?

So far, we failed to adapt our living habits to reduce our carbon footprint or destruction of biodiversity, to quote only a few of humans’ impacts on our planet. Air travel is booming, meat consumption is expanding, overfishing is destroying our oceans, animals species are disappearing at a never seen rate. Despite mass awareness campaigns and activist spectacular actions, behaviors’ change of what many consider as “traditions” proves very hard and remains limited to a few (btw, the younger generations do not look anymore virtuous than their elders).

While many celebrate the convenience and creativity of GenAI, the underlying energy and water requirements and associated carbon footprint often go unnoticed. For example, training a state-of-the-art large language model can consume energy equivalent to that of several households for an entire year. Generative AI has spurred another set of environment-destructive behaviors, using it “just” for entertainment or reduced effort…. And everyday brings its new social media GenAI tsunami. The latest “Ghiblification” trend uses OpenAI’s latest 4o image generation model to recreate photos with Ghibli’s signature elements. Everyday users as well as celebrities or even the White House have joined in, sharing their Ghibli-style transformations. This week, creating one’s mini-me doll in a blister box is flooding social media.

Using Gen AI is not a centuries-old tradition, so engrained that it would be hard to change. I would like to believe that this still new behavior could (hopefully) be influenced to mitigate its unintended consequences on our climate and biodiversity.

---

AI providers need to take collective responsibility

• To reduce energy consumption and carbon emissions in AI Training

A 2021 study estimated that training a single large AI model could emit over 626,000 pounds of CO₂—roughly the same as the lifetime emissions of five average cars. Data centers, even with modern cooling systems and efficiency measures, contribute to the environmental toll if powered by non-renewable energy sources. Providers need to focus on developing algorithms that achieve similar results with less computational power. Streamlining model architectures and using more efficient training techniques can substantially reduce energy demands, as well as transitioning data centers to renewable energy sources.

• To discourage unnecessary utilization of GenAI

While each individual usage to draft short, low-stakes communications or to generate images may seem trivial, the cumulative effect across millions of users can lead to significant energy usage. The ease of access inadvertently increases overall consumption. Providers (and authorities) need to raise awareness about the environmental costs of unnecessary AI usage. Educating users on when and how to use AI responsibly could lead to more mindful consumption.

• To collaborate in solving most pressing environmental issues

Collaboration between tech companies, environmental organizations, and policymakers is necessary to share best practices and innovative solutions to accelerate the transition to more sustainable AI practices, and to redirect some GenAI efforts towards solving environmental problems (such as optimizing renewable energy grids, predicting climate events, or limiting plastic pollution).

---

Empowering individuals for responsible usage

By fostering a culture of mindfulness and accountability, we could hope that individuals too contribute significantly to mitigating environmental harm.

• Educating to help understand environmental impacts

Educators and digital platforms can offer interactive tools that compare the carbon footprint of AI tasks (e.g., generating a meme vs. drafting a comprehensive report) with everyday activities like driving or home energy use.

Corporate and academic institutions can provide training on both the water and energy consumption of AI, and how to most efficiently use it (like batching queries).

• Developing mindfulness on AI usage (one can always dream…)

Individuals need to be encouraged to use their own thinking and creativity for low-impact tasks, where traditional software suffice. And the same social media which encourages viral anecdotal use of AI, could be leveraged to launch AI detox and diet campaigns!

---

We’ve long been failing in our environmental responsibilities—flying more, increasing deforestation, depriving oceans, watching species disappear. Now, with generative AI, we risk repeating the same mistake. In just a few years, AI has become another playground of excess: models trained at enormous energy and water costs, used casually and compulsively to entertain, replicate, and shortcut.

Yet unlike meat consumption or air travel, this is not a centuries-old habit embedded in culture or identity. Generative AI is new. Our relationship with it is hopefully still malleable. That gives us a rare opportunity—before patterns harden—to shape usage with intention.

This isn’t just on the tech companies, though they must lead in designing more efficient models, using clean energy, and being transparent about the environmental costs of their tools. It’s also on policymakers to regulate responsibly, and on institutions to educate users in sustainability literacy—digital and ecological alike.

But above all, it’s on us, the users. We need to ask: Do I need AI for this? Or am I just using it because it’s easy, fun, or trendy? Mindfulness isn’t naïve—it’s necessary. Choosing when not to use AI may become the new environmental discipline we didn’t know we needed.

Because if we can’t act wisely with a tool this new—this flexible—then yes, we are failing with AI too.

But we don’t have to.