This hot take comes straight from my email inbox this week.

We’ve, collectively, as OTPs, have broken ourselves into severe siloes when it comes to how we choose to adhere to our professional code of ethics and the OTPF.

When it comes to AI, we tend to land on one of two extremes: no AI, or everything AI.

But some of us are in the messy middle, and I’d argue that’s exactly where our profession should be. We’re advocating for public health initiatives. We’re voting against more data centers and for clean water. We’re fighting for stronger accountability of existing infrastructure. AND, at the same time, we’re learning how the systems actually work. We’re teaching ourselves how to minimize our own AI footprint through smarter, more efficient use.

Those things aren’t in conflict. One is macro, the other is micro. You can believe generative AI, as it’s being built and forced on us today, is deeply flawed. You can also accept that it is already embedded in the systems we work in, and that refusing to understand it does not protect our patients. It just leaves us less equipped to advocate for them.

This week’s headline provides a concrete example of what “smarter use” even means, because it’s more tangible than people realize. It comes down to something called tokens.

Every word you feed an AI has an energy cost. Fewer words = smaller footprint. The “caveman speak” virality isn’t a myth.

A recent framework called the One-Token Model, developed by the team at Antarctica, is trying to solve a problem most people never think about: the true environmental cost of AI is invisible. Companies spend millions running these models, but the actual physical work behind each token stays hidden, which distorts cost and obscures the environmental impact of AI at scale. They argue that if we keep treating AI as a magic black box, we will keep building infrastructure that is needlessly expensive, inefficient, and environmentally damaging.

To understand why this matters, you first need to know what a token even is.

When you type something into ChatGPT, Claude, or Gemini, the AI does not read your sentence the way you do. It breaks everything down into small pieces called tokens. Think of tokens like '“tiles in a game of Scrabble.” A short, common word might be a single tile. A longer or more unusual word might take several. Every response the AI generates is built by laying down those tiles one at a time, and every single tile requires computational effort, which means electricity, which means heat, which means water to cool the machines.

Here is the part that connects straight to the environment: in most AI models, the energy consumed is directly proportional to the number of tokens processed. More tokens in and out means more electricity burned and more carbon emitted. And it adds up fast. According to the National Renewable Energy Laboratory, cooling systems alone can account for up to 40% of a data center's total energy use. Multiply that by billions of exchanges happening every day, and the footprint becomes enormous.

This reframes something we usually treat as purely technical into something ethical. Every bloated, rambling prompt has a cost that lands somewhere, often in the communities hosting these data centers. But it also means individual users have more agency than they think. You cannot single-handedly fix AI’s climate problem. You can, however, stop wasting tokens, and when millions of people do that, it’s influential.

Watch for “token efficiency” and AI energy transparency to become bigger parts of the conversation, especially as enterprises face pressure to report their AI carbon footprint. Also, watch how this shifts the framing of prompting skills. What I’ve been teaching as a way to get better outputs turns out to also be a way to waste less energy. Those two goals point in the same direction.

This was generally my response to one of the questions in that email I received this week: how can you teach people to use AI “effectively” when AI is harming the planet? This is how. Teaching effective prompting is one of the few concrete, individual-level things that actually reduces harm. A clinician who writes tight, specific prompts with 0-1 reiterations due to good prompting consumes fewer tokens than one who dumps in paragraphs of unnecessary context and regenerates ten times to get it right.

We talk constantly about occupation, environment, and the fit between them. The digital (virtual) environment is part of that. If our patients and students are using these tools regardless, then teaching them to use them safely and effectively is a small act of environmental stewardship layered on a clinical skill. It will not save the planet at a macro level. It is also not for nothing at the micro level. And I think that’s defensible as an individual and a profession.