The question I get asked most often, when people learn I make generative art with code, is whether I'm worried about AI replacing what I do.
It's a reasonable question, and the honest answer is: not really, and also that's not quite the right question.
What I'm actually thinking about is different and less dramatic. Not whether AI replaces the work, but where it enters the work thoughtfully, and where it doesn't belong. That's a harder question, and a more interesting one, and I don't have it fully figured out.
Let me start with what I do use it for, because the answer isn't glamorous.
I use AI writing tools to help me draft and edit: newsletters, documentation, communications at work. I use it to synthesize research faster than I could alone. I use it for code assistance when I'm working on scripts and hit a wall on syntax or approach. I use it for brainstorming when I'm stuck and need to think out loud with something that will respond. These are the unglamorous applications that account for probably 80% of my actual AI usage, and they're genuinely useful. They give me time back.
At Non-GMO Project, where I lead information systems and our approach to emerging tech, we're doing something similar. We're asking: which AI tools serve the mission without compromising integrity? Where does efficiency gain outweigh risk? Where does the technology's uncertainty about facts or sourcing create problems that matter more than the time it saves? We're being deliberate about it rather than adopting by default, which means moving slower than some organizations and being more confident in the decisions we make.
None of this is dramatic. It's just thoughtful adoption, which sounds boring but is actually hard to do in an environment that rewards moving fast.
Where I don't use AI is more interesting to talk about, and more specific to the art practice.
The generative scripts that produce my work are code I write. The system is mine: the seeds, the parameter spaces, the shape grammars, the color logic. When I run the algorithm and generate dozens of outputs waiting for one to click, that process is between me and a system I understand and built. It's not between me and a model trained on other people's images.
This distinction matters to me, and I've thought about why. It's not purity or gatekeeping. It's that the whole practice is built around the idea that my intuition is doing something real. When I read a collector's note and adjust the parameters, I'm making a judgment that I'm responsible for. When I recognize the output that's right, I'm exercising something I've developed over years of looking. If I handed that process to a model, even a good one, the thing I'm selling would be different in a way I'd have to be honest about.
So the generation stays mine. The selection stays mine. The parameter intuition stays mine.
Where I'm genuinely still figuring it out is the harder part to write.
I've been a technological early adopter my whole career. Flash in 1999. Social media tools for nonprofits in 2012. NFTs in 2021. Generative art in 2020. Each time, the pattern is roughly the same: something new arrives, I engage with it deeply before most people, I discover real value, and then I slowly discover that value doesn't come for free. There are usually costs that take a few years to become legible.
I'm holding that pattern consciously with AI. I find the tools genuinely useful. I also know that widespread AI adoption has implications I can't fully see yet, including implications for artists and for the value of human creative labor. I don't want to be naive about that, and I also don't want to perform concern I don't actually feel as a way of distancing myself from tools I'm using.
The honest version is: I use these tools, I find them valuable, I'm trying to be thoughtful about where and how, and I'm watching what happens to the broader ecosystem with real attention. That's not a satisfying resolution. I don't have one yet.
What I keep coming back to is the question of irreplaceability.
The parts of my practice that matter most are the parts that can't be automated: the years of learning to see, the intuition about what a particular person's story wants to become, the feel for when PVA glue is right, the judgment about which output resonates. Those aren't things I could hand to a model even if I wanted to.
The parts that can be automated, I'm increasingly comfortable automating, because they were never the point. The point is the work, and the relationship between me and the collector, and the object that ends up on a wall somewhere catching afternoon light.
AI helps me get there faster on the parts that are just logistics. Everything that matters is still happening by hand.
Shawn Kemp is a generative artist and Information Systems Renewal Officer at Non-GMO Project, based in Bellingham, Washington. Learn more at shawnkemp.art.