When Is an AI Picture Actually Artwork?

AI generated pictures are actually seeping into promoting, social media, leisure, and extra, due to fashions like Midjourney and DALL-E. However creating visible artwork with AI truly dates again many years.

Christiane Paul curates digital artwork on the Whitney Museum of American Artwork, in New York Metropolis. Final yr, Paul curated an exhibit on British artist Harold Cohen and his laptop program AARON, the primary AI program for artwork creation. Not like right now’s statistical fashions, AARON was created within the Seventies as an professional system, emulating the decision-making of a human artist.

IEEE Spectrum spoke with Paul about Cohen’s iconic AI program, digital artwork curation, and the connection between artwork and expertise.

How do you curate digital artwork?

Christiane Paul: Curating digital artwork isn’t that completely different from some other artwork type. Whether or not portray or pictures or print, all of us take a look at the sophistication of an idea and the way it’s translated right into a medium. So my curatorial decisions aren’t pushed by the expertise. In the event you’re a curator of portray, the choice of a piece for an exhibition wouldn’t be pushed by a particular paint or approach for a brush stroke.

In 2001, Harold Cohen produced AARON KCAT as a part of his experiments in producing figures with the AI mannequin. Cohen taught the mannequin the way to deal with overlapping objects in a composition, which he did by having the mannequin fill in objects from the foreground to the background.Whitney Museum of American Artwork

That being mentioned, in fact, there have been reveals about pointillism as a particular approach in portray. And, there might be an exhibition centered on AI applied sciences as an inventive medium. However the common standards would nonetheless be the sophistication of idea and its implementation.

Do you collaborate with engineers as a part of your work?

Paul: Sure, in fact. Many artists even have a background in engineering, notably in relation to the older technology of digital artists. When there weren’t any digital artwork applications or colleges, digital artists typically would have a background in engineering or programming. So you’re employed with builders and software program engineers, and plenty of artists are programmers or coders themselves—I’d say a lot of the artists I’m working with. They often need to outsource, simply as a result of quantity of labor, however most of them are additionally very deeply within the weeds.

What are the challenges of gathering and preserving digital artwork?

Paul: For artwork establishments or collectors, it is very important have requirements and greatest practices for archiving and maintaining monitor of the applied sciences, as a result of computer systems and methods change at such a fast tempo. Within the ’90s individuals began paying extra consideration to implementing conservation approaches, and there are a number of methods. Certainly one of them is storage and {hardware} conservation. That is used for items that conceptually depend upon {hardware}. After which there may be migration, emulation, and re-creation.

There is no such thing as a silver bullet. One has to have a look at the person art work to see which strategy could also be the most effective one. Within the Harold Cohen exhibition, for instance, we principally re-created one of many earlier items from scratch based mostly on Cohen’s notebooks and printed out code that we discovered, and his son truly recoded that in Python. We reconstructed the unique BASIC however then additionally recoded in Python.

What impressed the Cohen exhibit?

Paul: I had identified Harold Cohen for fairly a while. We labored collectively on an exhibition in 2007, and AARON is an iconic work. All people learning digital artwork is aware of this as one of many basic items.

We had introduced a few of his works into the gathering of the Whitney Museum, so showcasing that was one level. However I additionally thought that it could be notably fascinating to revisit the primary AI software program for artwork making within the gentle of present text-to-image fashions. Their processes are radically completely different, and authorship and collaboration play out in a really completely different manner.

AARON discovered the way to generate pictures of crops, as seen in AARON Gijon from 2007, utilizing guidelines that Harold Cohen offered about their dimension and patterns governing their branching and leaf formation.Whitney Museum of American Artwork

Harold Cohen wrote AARON from scratch. He was fully in control of constructing that software program, which he advanced throughout 5 completely different languages over his lifetime, so the composition of a picture was fully underneath his management. He moved from evocative types, to a figurative part, to a plant-based part, after which returned to abstraction. Later in life, he taught the software program coloration composition and he additionally constructed the drawing gadgets that may execute AARON’s work. He actually thought-about AARON a collaborator, and AARON encapsulated Cohen’s sensibility and aesthetics.

In the present day’s AI software program is basically statistically based mostly, and lots of the authorship and company occurs within the company black field. The artist has no management over that, even when artists practice and tweak their very own fashions. Artists working with AI are very invested in manipulating the software program and dealing with it, however there at all times is a element that’s created by companies that they don’t have management over.

Can AI-generated pictures be artwork?

Paul: Not all visuals created by text-to-image fashions are artwork. It’s fantastic that individuals can use AI to generate pictures and play with it, however I wouldn’t name that finish end result artwork.

AI artwork makes use of synthetic intelligence as a device and medium in a conceptual and sensible manner, critically partaking with these applied sciences and questioning them, be that from an moral or aesthetic perspective. Most of right now’s AI artists are partaking with these applied sciences in a really deep manner. They’re placing collectively their very own coaching datasets. They practice the fashions. They query the biases embedded in AI. So it’s fairly an advanced and concerned course of, and it’s not merely a textual content immediate producing a picture.

This text seems within the Might 2025 subject as “5 Questions for Christiane Paul.”