I was chatting with a friend over coffee the other afternoon about nothing in particular, and inevitably the topic turned to AI. After all, I coordinate a United Nations-backed group of AI experts, and he makes his living creating AI-generated videos. ¡°Can¡¯t live without it,¡± he said casually, like he was talking about electricity or running water.

Because of the work that I do, I asked what felt like an obvious question, ¡°Do you ever wonder what AI might be costing the planet?¡± He paused, looked away, and thought for a moment. ¡°No idea,¡± he confessed.

I nodded, because the truth is, most days, neither do I.

We talk about AI in big words. Transformative. Disruptive. World-changing. I am guilty of describing AI in these superlatives, as if it were a force of nature rather than a set of systems engineered by people.

But for all the AI awe, one word rarely finds its way into coffee table conversations: sustainability. Colleagues debate the future of work. Friends compare the best AI features on their smartphones. But too often, sustainability appears as a footnote, if at all.

This absence matters. Talking about AI without sustainability is like learning to cook without ever talking about turning off the stove. Things might seem fine for a while, but the electricity bill grows and the system overheats. Then, something starts to burn. AI is no different. Large systems run on massive amounts of energy, trained on nearly all the text humanity has created, and deployed in settings marked by inequality. Ignoring sustainability does not make these things disappear; it just hides them until they are too big to ignore.

Across the conversations we have been having in the United Nations University Institute for the Advanced Study of Sustainability¡¯s podcast series ¡°AI and Sustainability,¡± a common thread has emerged. When sustainability is placed at the center, AI stops being seen as a shiny object and starts becoming a tool for real development. One that can be truly transformative, disruptive and world-changing.

Real development demands tools that work on the ground. Climate scientist David Daou emphasized this in a recent interview on the podcast. AI tools, he noted, do not need to be enterprise-scale systems developed by tech giants. ¡°Small¡± and ¡°medium¡± AI can do the job with modest computers, offline kits or even USB sticks when the internet is slow. They can do the job just right at a fraction of the environmental cost. He added that models get smarter when local knowledge is incorporated: wind direction, tree species, animal sounds. His message was simple. AI is not a silver bullet, but when co-created with communities, it becomes more powerful and sustainable.