Beginning in 1996, Rhizome was an email list where artists shared resources and developed language around emerging media. Terms and neologisms were often introduced, tested out, and discussed; they were sometimes adopted and later discarded. The kind of work we know today as “generative” was described in distinct ways; one popular term was “artificial life.” “Emergence” was another key term.
Still, the term did make appearances and was often discussed with considerable sophistication. In 1997, Simon Biggs wrote about his work in a way that seemed to mirror today's arguments around AI and consciousness. He argued for "an avoidance of the notion of an 'artificial author'"—the idea that the computer is taking the place of the human author—which has been part of computer art discourse since the earliest days. Instead, he argued for linguistic models that generate patterns "that can be interpreted as meaning in the mind of the reader."
Biggs’s argument from a quarter-century ago mirrors a contemporary debate around AI and the role of machines in the creative process: do AI tools have true intelligence and artistry, or are they just generating patterns that we interpret as meaningful?
Today, the term “generative art” is increasingly also associated with generative AI. While generative art has long been concerned with code as aesthetic material, generative AI makes use of complex models that analyze large data sets and generate new images, sounds, and texts derived from its training data.
With more kinds of artwork being described as “generative,” a further definition suggested by Marius Watz in 2005 may be of use. For the term “generative art” to have any meaning, he argued, artworks must have a dominant focus on generative systems, not merely use a generative tool along the way. Whether it involves AI or Java Applets (or both), art can be most readily described as generative when it foregrounds the generative system through which it comes into being.