Insights
It should be relatively simple to build decent government technology. Care about your problem, design with your users, build in small iterative increments, and release often. The second love of my life is to explain these to governments who don’t ask for it, vendors who should know better, and large accounting firms who are experts in technology modernization.
But when the work day is done, I paint. I’m terrible at it. For a painting to be good, it must convey empathy and emotion through palette choice, treatment of light, the thickness of paint, and the quality and intentionality of your brush strokes. The viewer is the last part of the painting, as much a part of the whole experience as the painter.
I promise you, there were no aliens in the studio that day.
To push some paint around is the third love of my life. My American dream is to invite you to a gallery showing eventually. One of my paintings will hang in a forgotten corner by the restrooms. I’ll wear all black, we’ll sip Costco prosecco, and discuss how my painting is a nihilistic self-experiential post-modern expression of the illusion of hope and predictability during PI planning. Paint what you know.
But I’m many years away from achieving this. Between now and then, there is a lot of practice, theory, skill building, and despair. I’ve turned to Reddit, searched on the internet, and even borrowed books from the library. Learning this way is disorganized, and it takes a long time to get answers to specific questions or problems immediately.
And so like everyone else, I decided to ‘AI until you make it’.
A couple of years ago, I started by using AI to generate images I could paint. It was easy to prompt the AI to create paintings that were easier, with simpler colors and brushstrokes.
Early generators were pretty bad at generating scenes that made sense. The hands or a figure would be wrong, doors would lead to nowhere, or in one case, upper bodies in the distance seemed to be missing their legs. The output also lacked emotion. And there was the issue of appropriation: AI trains on contemporary artists’ work without compensation or attribution.
So now I paint the objects around me, or models in the studio.
The pear wishes it were in some other painting.
I just use AI in different ways instead. It’s proved to be an incredible resource in discovering new artists. Looking at real paintings is a great way to get inspired, learn about trends in contemporary art, or observe how an artist solved a particular problem.
Recently, I came across Sasha Hartslief, a contemporary figurative artist from South Africa. I love her work. So, I asked Claude to “Suggest some contemporary figurative artists similar to Sasha Hartslief”. I was introduced to Casey Baugh, Hollis Dunlap, and Zooey Frank, among others. And helpfully, Claude followed up with a question: “What aspects of Hartslief’s work do you particularly appreciate? Her handling of light, her subject matter, or perhaps her painting technique?”
Answering that yielded another list of excellent artists. As a new artist without formal training, I’d never stumble upon these alone.
Before starting a new painting, I also asked it to compile the steps I’d need to follow: “What are all the steps to paint an oil painting of a figure? I want an overall process that includes sketching in the beginning, underpainting, etc.” The resulting response is a sensible path forward. I’m trying my best to follow it.
I’m discovering artists faster and learning quicker. At this introductory level, it’s enough to get started. I’m hoping that soon I’ll be able to upload the image of a painting in progress and have it provide a kind, personalized critique appropriate to my skill level. That day will come.
AI works well for learning new topics because it learns from every piece of content it can get its grubby paws on. It generates the answers to your questions through statistical wizardry, advanced algorithms, and a tremendous amount of computing power. For topics where I’m an expert, like civic technology, it hasn’t told me much I don’t already know. But for things that I don’t know much about, like art, it is a tremendously valuable resource.
What I’m trying to learn can happen through carefully constructed Google searches. But with AI, the clean, no-nonsense interface and conversational tone are part of the appeal. Backed by billions in venture capital, there are no ads. This will change, and our access to knowledge will again be noisy, filled with ads and other visual junk. Until then, I encourage you to AI it until you make it.