What Does Responsible AI Actually Look Like? I'm Still Working It Out.
Both of my parents were adopted. Both are Black American, and like so many African Americans, neither knew the full story of where their family came from. The slave trade didn't just separate people from their families. It deliberately severed entire communities from their cultural memory, their languages, their origins. That loss didn't end with emancipation. It echoes forward.
This is personal for me. It's the reason I do this work at all.
And it's also the reason I can't simply wave away the criticism of it.
I work in AI. And I want to say something that doesn't get said enough: the people who are worried about this technology are not wrong.
The fears around job losses, environmental damage, cultural erasure, and the flood of soulless machine-generated content filling the internet are not the concerns of people who simply don't understand it. In many cases, they're the concerns of people paying the closest attention.
I think about them constantly.
The environmental question is the one I find hardest.
Training and running large AI systems requires extraordinary amounts of energy and water. The infrastructure behind these tools operates at a scale most people still don't fully grasp. And as AI becomes more embedded in everyday life and work, those demands grow with it.
What tends to get lost in these conversations is that not all AI is built the same way, or for the same purpose.
The industry tends to treat scale as progress. Bigger models, more data, more automation. But in cultural work, restraint is a form of intelligence too.
Research increasingly shows that smaller, purpose-built models trained on carefully selected datasets consume dramatically fewer resources than the large frontier models most people picture when they hear the word AI. Studies suggest up to 60 to 70 percent less energy and water. Some can run locally on a laptop or desktop, without data centre infrastructure, without water-intensive cooling systems running continuously in the background.
That gap matters. The choice isn't simply between using AI or not. How you use it, and what kind of system you reach for, makes a real difference to the environmental cost.
When I build tools for cultural organisations, I'm not feeding community histories into massive public systems and hoping for the best. I work with contained models trained on approved material, with clear limits and human review throughout. Someone is always responsible for what comes out.
Using AI responsibly, in my view, means being deliberate about it. Using the smallest tool adequate for the task. Asking whether AI is genuinely necessary, or just convenient. Those questions don't come naturally to an industry built on growth. But they should be asked every time.
The job question is real, though it needs more precision.
The work at the heart of cultural heritage, interpretation, community relationships, the slow process of understanding what a collection actually means, is not work that should be handed to a machine. It's work a machine might help make possible, by clearing away some of the weight around it.
A curator spending less time manually searching fragmented archives and more time in conversation with communities is a good outcome. Replacing those communities with autogenerated interpretation is something different entirely, and we shouldn't dress it up as the same thing.
The risks are not built into the technology itself. They come from decisions, commercial pressures, and what institutions choose to prioritise. There's still room to push back, if people are willing to.
On the quality of AI content, I have less patience for nuance.
The internet is filling with material generated without care, without cultural knowledge, without anyone truly thinking about what they're making. In cultural heritage contexts that's not just an aesthetic problem. Complex histories get flattened into summaries built for speed rather than accuracy. Communities get misrepresented. Context disappears.
This is already happening.
The answer isn't abandoning these tools. It's refusing to use them carelessly. Everything that comes out of my practice has a human being responsible for it. That's not a marketing line. It's the baseline for doing this work honestly.
Some days it feels like the technology moves faster than our language for talking about what it's doing to us. We're all working this out as we go.
I don't think that uncertainty is a weakness to hide. It might be the most honest thing any of us can offer right now.
The fears aren't irrational. They're pointing at something real about what we owe each other when we build tools that touch memory, identity, and culture.
I'm trying to follow that. That's the best I can say.
