AI is not my enemy. But it is learning from me without asking.

Using AI on my own work is my choice. Using someone else’s work without asking is not. That’s the line I draw.

I’m not against technology. Resisting it doesn’t help — finding out how it can make life easier does. I use AI to help me with bookkeeping, to research materials, and as one of many tools in my artistic process.

I’ve also experimented with it directly. A few years ago I made the Bearbold Collection, a series of NFTs, because I wanted hands-on experience with both NFT art and AI. I’ve even generated a few (ugly) AI images to illustrate my posts on social media. More recently, with Mirror Dissonance, I used AI to generate images based on my own artwork — because I wanted to understand what it can actually do.

And sometimes I use it for references. Before a painting, I often work with 10–12 reference photos to understand how a subject looks from different angles and in different light. I use my own photos, and I work with a network of photographers who provide high-quality images I can base my work on. When I needed Ayrton Senna in a very specific position, with an expression and lighting that didn’t exist in any photograph, I used AI to build it from existing photos. Not to make the painting — to make the reference I needed.

I rarely end up with a fully AI-generated image — if ever. AI can’t give me the precise result I want, so it becomes a mix of AI, Photoshop and handwork. The image with this post shows exactly that: the mood board behind Growth Tea — my own references, the AI attempts marked “AI”, and the finished painting.

Artists have always manipulated references. This is just a new tool.

But there’s another side to this.

AI models are trained on images scraped from the internet — including artists’ work, without permission and without payment. The more distinctive your style, the more valuable you are to those models. And the more likely someone, somewhere, is generating images that look like you.

So I’m running all images of my work through two tools developed at the University of Chicago. Glaze protects my style from being mimicked. Nightshade adds nearly invisible changes that corrupt an AI model if the images are scraped into its training data.

It won’t stop everything. But it gives me peace of mind.

The question I keep coming back to: I used AI to adjust the expression on existing photos of a real person to get the reference I needed. The result is a painting — not a photograph, not a claim about how Senna actually looked. But I still wonder: is that okay?

I think it is. But I’m not certain. And I think that uncertainty is worth sitting with.