Part 5: Using other people's art

Websim and the practice of generative art

Some of my experiments were adapted from works by established artists. This has been a fascinating way for me to learn about generative art. It reminds me of how budding painters practice their craft by studying the masters. In this part of the tutorial, I’ll share my explorations in this area.

However, before getting into that, I think that it's important to begin with some reflections on how such practices might be performed in an ethical manner. To this end, Egg has graciously agreed to help me start a conversation with all of you.

These reflections were written by Egg as the sole author, expressly for this blog post (as opposed to content derived from text messages that they sent me), to help situate my work in the broader historical context of generative art:

What is generative art? Definitions have varied widely over the years, but one that many in the field have settled on is roughly: 'art that has been created with the use of an autonomous system'¹. The Art Newspaper expands that as follows: 'the term generative art describes works based on an algorithmic code or a mathematical formula. Generative art is made through a set of rules that automates the output where there is usually randomness embedded in the algorithm. The process is defined by the artist, and the output is a sort of machine and artist collaboration'².

I have a longstanding interest in the topic, and have made a fair amount of generative art as a hobbyist over the past twenty years. Few in the community have seen my work, although I had the great pleasure of discussing it with Frieder Nake some years ago, and have gotten to spend some time with Ben Fry and Scott Draves. I was surprised when my friend Ophira started showing me some recent pieces she'd made that were clearly code-based. I've used current state-of-the-art LLMs to write workaday code, and of course I'm aware of the controversy surrounding the use of text-to-image 'generative AI' systems like Midjourney³. But it had never occurred to me that LLMs could be used to create generative art..

Ophira has been experimenting with LLMs and art for some time⁴, and just recently that started to involve the LLM writing code. It would be easy to dismiss her work as just another instance of AI stealing the work of real artists -- but we shouldn't be so quick to judge in this case. After all, while generative art is most often expressed in code these days, there's a long history of creative algorithms written on paper, or spoken aloud, or communicated in other ways. All of them fit the definitions we see above. And despite recent advances, who's more familiar than generative artists with being dismissed as 'not real artists'?

I've seen Ophira laboring over her craft for endless hours; this isn't a quick or mindless process. Certainly some of these first steps into algorithmic art are based on the work of existing generative artists, but I believe that's in the sense that painters have always done studies based on the work of the masters. And as she's started to become familiar with the history and techniques of generative art, she's started producing some genuinely original work. It was one thing when she coaxed an LLM to produce art based on a generative art tutorial, and quite another when she started basing her work on pure descriptions of the algorithms she wanted to see. At what point do studies shade over into original art?

I don't expect most generative artists to shift their practice in this direction -- for those with a deep understanding of how visual phenomena are represented in code, writing the code directly provides a level of power and control that LLM-based work can't match. But I can personally see myself using these techniques like a sketchbook, a way to quickly try out the result of feeding a few processes into each other. And as with other new mediums, over time this medium may develop in its own directions that couldn't have been achieved otherwise. One possibility I can imagine is that this could lead to a sort of meta-generative art, where an entirely new generative art piece can be created for every viewer, with meta-parameters created by the artist that establish carefully-chosen boundaries of the range of new pieces created.

And for those who don't know how to code -- this technique provides a doorway into the medium. Budding generative artists have always looked at the source code of pieces they liked, and started tweaking it to make it their own. These LLM-based techniques, similarly, provide a way to generate new source code quickly, code that's wide open for new practitioners to start tweaking. This approach is, to my knowledge, less than a month old, and already there's starting to be a lively community of people remixing and building on each other's work.

My hope is that, rather than dismiss work like Ophira's out of hand, established generative artists will consider adding these tools to their already wildly eclectic toolboxes, and judge work like this on its originality and vitality and aesthetic richness. After all, that's all we ever asked of the rest of the art world.

¹ Avant Arte

² The Art Newspaper

³ The term 'generative AI' has annoyed me from the beginning, mostly because it's bound to get conflated with 'generative art', causing endless confusion about the difference.

⁴ Including, fascinatingly, getting LLMs to produce visually interesting ASCII art despite the fact that they've never been deliberately led to have any understanding of space, and researchers were extremely surprised when they did.