The Environmental Cost of AI: An Honest Conversation

The Reality

Large language models consume energy. Every story Daily Adventure generates uses electricity - powering servers, cooling data centres, transmitting data. The tech industry often glosses over this, but I won't.

You might have seen headlines claiming a single AI task uses as much electricity as a household does in a day. That's misleading - those figures typically refer to training entire models over months, not generating a bedtime story. But that doesn't mean the impact is nothing. It's something, and it adds up.

Data centres use water for cooling. They take up land. They require infrastructure. These are real costs, and as AI usage grows, so do these costs.

Why I've Made This Trade-off

So why build Daily Adventure at all?

Because I've weighed the value against the cost, and for our family - and now for hundreds of others - it tips in favour of running it.

My son is more engaged with reading because the stories feature him, his interests, and places he recognises. We use the stories for actual reading practice, calibrated to his level. He's motivated to read because it's his story. That engagement with literacy, that love of stories - I couldn't create that myself. I'm a developer, not a storyteller.

We're not replacing professionally written books. They're still a huge part of his diet. But a bedtime story featuring his characters, set in his town, at his reading level? That didn't exist before Daily Adventure.

For children who are neurodivergent, we've added features like consistent opening and closing phrases, progress markers, familiar settings, and social situation practice. These aren't gimmicks - they're genuinely useful for kids who need routine and predictability in their stories.

What We Do About It

I won't claim we've solved the problem. We haven't. But here's how we try to be responsible:

Intentional usage, not frivolous generation. Daily Adventure creates one story per day, not endless content on demand. There's a purpose to each generation, not a "generate until you get something good" approach.

Efficient model selection. We use the smallest model that does the job well. Bigger isn't always better, and smaller models use less energy.

Caching and efficiency. We don't regenerate what we've already generated. Common elements are reused where appropriate.

No pretence of carbon neutrality. We support verified carbon offset and removal projects, but I won't claim that makes us carbon neutral. Offsets are imperfect. They're a contribution, not an absolution.

The Honest Bit

At the scale we operate - generating a story a day for families - we're a rounding error in the global AI energy picture. The real energy consumption comes from training models and enterprises running millions of queries daily.

That doesn't make our impact zero. But context matters.

I lead a fairly green life in other respects. We donate physical books, Yoto players, audiobooks, and board games to local schools. We run a creative writing contest for kids that explicitly doesn't use AI. The point is to add value to reading and imagination, not to replace human creativity.

My Position

AI and LLMs aren't going away. Avoidance won't make them disappear. The question isn't whether they exist, but how we use them.

I'd rather be someone who engages thoughtfully with this technology - building guardrails, being intentional about usage, teaching my son about what AI is and isn't - than someone who either ignores it entirely or uses it carelessly.

If AI can be safely run to help kids engage with stories, expand their imagination, and build a love of reading, then the trade-off is worth it to me. I recognise not everyone will agree, and that's fine. Every parent gets to make that choice for their family.

But I refuse to let this potential to help go unused while pretending the environmental cost doesn't exist. Both things can be true: AI has a footprint, and it can still be worth using thoughtfully.