The Last Frontier

Lately, there has been a robust conversation about ChatGPT, the artificial intelligence chatbot developed by OpenAI. Released just this last November, editorial and social media commentary range from effusive fawning praise to skeptical consternation, and fearful concern. While the large language model routinely produces articulate responses across many domains, at times, it also confidently makes factually incorrect statements, and errors in logic, and displays other disturbing ‘behaviors’ such as, in one publicized case, asserting emotional attachments to its instructor.

More than a little conjecture focuses on what human jobs are in jeopardy. Uh, oh! Who wants to be replaced by a digital instruction set? One thing is for sure, this conversation about AI, the ChatGPT chatbot, and others is not going away. AI applications are making big waves. So, what does it all mean? Who among us knows? Buckle up. It's going to be a bumpy ride.

While my concern focuses more on the impact of AI on the visual arts, it is worth thinking about written art, poetry, and literature, the written words created with lasting artistic intent. Visual and language-based arts are more obviously in the crosshairs of AI than physical manifestations such as sculpture and architecture. Although architecture could conceivably be next because most design these days is digital. Anything that is created by the human hand seems safer from these digital copycats. While AI has seen success in writing job applications and academic essays and other short-form written text, I find it more difficult to believe that AI could challenge human authors in the long form.

When pondering whether AI could create a novel as complicated, messy, emotional, and satisfying as ‘We Begin at the End’ by Chris Whitaker, I do not see it. How could AI create multi-faceted characters and convoluted intersecting story arcs without anything but an essay-length instruction set, and even if then? And, I posit that there is no chance that AI could compete with Philip Pullman’s ‘His Dark Materials’ trilogy, or Adrian Tchaikovsky’s ‘The Final Architecture’ series. Could intricate fantasy world-building be AI’s forte? Perhaps I am too smug, but I doubt it.

Why am I writing about this, let alone allowing it cognitive space? Because I care about art and this is as good an intellectual query as any. The conversation started for me when my dear friend, the recently departed Ron Zimmerman, posted an AI digital ‘painting’ on his Facebook page. The date: November 2, 2022. He wrote: "Painting of a girl with a flute created by Artificial Intelligence from a plain English description.” She looked so familiar. I murmured to myself, “Vermeer.”

And then another friend, Joanne Johnson, commented, “Reminds me, not sure why, of The Girl with the Pearl Earring, by Vermeer. I wonder if AI "learning to paint" involves loading lots of Master's works into the databases?” Ron replied, “I asked it to paint using the style of Vermeer!” So, the mystery was solved.

The rendition of the girl was arresting and disturbingly beautiful. I responded to Joanne, “I was about to make the same comment. It is definitely inspired by Vermeer. These AI images rely on a significant database of image materials as resources. It is not as if there is any artistic inspiration. I find it amazing that these images can be created by AI but I submit that it would be impossible without the vast image database which these programs draw upon (pun intended). What would happen if you sat HAL (remember 2001: A Space Odyssey?) across from a woman on a chair and instructed HAL to paint her image, without digital access to other resource materials? I doubt it would be very impressive.”

Like a dog with a bone, I continued, “Hi Ron. This is a significant subject for considered discussion. Yes, this image is impressive. But remember, you told the machine what to do. I too have tied DALL-E. And after signing up, I asked it to paint a fall mountain landscape with subalpine larches. I got four renditions of what would be considered forgettable motel room paintings, or mediocre postcard art, mundane underwhelming stuff.”

“If I had referenced a specific artist, I may have received a more impressive output. Fortunately, you called upon a master of light, Johannes Vermeer. Not many would know to do so. Viewing the disappointing AI landscapes from my request, I could tell that it had merely copied, and composited from already available online digital images. So, I lost interest in the program as it seemed to be less interpretive and creative than I may have expected. I acknowledge that almost all art is influenced by those who have gone before. Our human brain could be considered the original database from which derivative images are created, as we all remember what influences us. After all, the great masters of the Renaissance diligently copied the art of those who came before them to learn and create a path toward their unique creative expression. As the saying goes, we build on the shoulders of giants.”

At present, AI seems to me more interpretive and highly derivative than creative. And, it may yet surpass its current state. Probably. Who knows? But for now, it seems mostly a parlor trick, a toy. While AI is a very powerful machine language with an extraordinary ability to acquire, compile, and arrange digital data, it remains highly dependent on available source material. 

Ask the machine to create a painting of a sun-drenched swimmer in a Los Angeles swimming pool with a poolside watcher, but NOT in the style of David Hockney. Ask it to create an image of a lonely man at a midnight diner but NOT in the style of Edward Hopper. A tissue paper collage of a 'very hungry caterpillar' but NOT in the style of Eric Carle. An illustration of the tale of Treasure Island, or Uncas battling Magua in the Last of the Mohicans but not in the style of N.C. Wyeth. A gilded portrait of Adele Bloch-Baker but NOT in the style of Gustav Klimt. Or the symbolist image of eighteen gossamer-robed young women with musical instruments descending a spiral staircase but NOT in the style of the famous pre-Raphaelite Edward Burnes Jones. I could go on and on and on, endlessly, as I recount sensational images created by humans, but you understand what I am driving at. Art at its core is a messy endeavor that is borne of inspiration, desire, and often torment. The artwork of the greatest is both unique and groundbreaking. And for the rest of us, a true expression of ourselves. The machine shares and feels none of these inspirations, motivations, and feelings. It merely performs according to an instruction set. Long live the expression of divine art created from human inspiration!”

Three days later, Ron Zimmerman posted another AI work. “Squirrel with a fork and napkin around the squirrel's neck eating a thin piece of cheesecake with a fork on a breakfast table”

I commented, “I like it too. How about a squirrel with a pencil mustache, wearing a fedora, and smoking a fat cigar while pouring a hefty portion of Macallan 12-year-old Scotch whiskey into a crystal tumbler?” Ron responded with another squirrel image. I thought I would try my description in a separate query. I got somewhat different results.

I replied, “Well, my squirrel idea generated some terrible images but my request for a mountain goat painting a watercolor of the Matterhorn in the style of Marcel Duchamp yielded some unexpected, yet charming results. The Tyrolean Alpine Hats look a bit off. But that's okay.” I intended to challenge the AI to use Marcell Duchamp’s painting ‘Nude Descending a Staircase’ but the AI did not take the bait. Too bad.

I then briefly veered off into copyright issues, “Interesting new article on AI artwork and copyright law, how visual artists are ‘questioning the legality and ethics of AI that incorporates existing work.’ Story on Wired.com: This Copyright Lawsuit Could Shape the Future of Generative AI.” Adobe promises that their AI will only access ‘rights-free’ images, but others don’t make that claim. So, it seems that anyone’s copyrighted image is fair game for adaptation in a work created by an AI program, only grist for the mill.

Recently, German photographer Boris Eldagsen was named the creative category winner in the open competition in the Sony World Photography contest. He declined the award as he used AI to create his image. He intended to demonstrate the need for photography and AI images to be judged separately in competition. So, battle lines have been drawn. AI is here and it is already disruptive. Even in-phone photo apps are creating an avalanche of unrealistic photos of Instagram models with digitally altered bodies and facial features. We already recognize rapidly emerging and significant problems when we cannot differentiate between human and machine-created words and imagery. It’s messy out there and it’s only the beginning.

Back to Vermeer and my core issue about fine art. How do humans create art? How does AI proceed? And, will humans hold their current creative edge, their seeming slim advantage? I think so. I hope so. But, then again, I am an optimist.

Vermeer created but a small body of work. Only 37 paintings have been attributed to him. The ones that are best known are his interiors which display his deft use of window light to portray luminous scenes of domestic tranquility. While his fame was not widespread during his lifetime, due to his limited creative output and predominantly local patrons, his exposure in recent years has skyrocketed. And because of his now widespread popularity, there is a deep digital catalog of his work, uncopyrighted digital images ripe for reference material to be incorporated into digital depictions by AI algorithms such as DALL-E.

So what image is most responsible for the AI depiction created from Ron’s description? It is not surprising that The Girl with the Pearl Earring comes to mind. It is one of Vermeer’s best-known paintings. But there is more to it than that. The screen actress Scarlett Johansson starred in a 2003 movie titled ‘Girl with a Pearl Earring,’ an adaptation of a book by Tracy Chevalier based on Vermeer and his subject for that painting.

Roger Ebert’s review awards the cinematic effort with four stars, the maximum. "Girl With a Pearl Earring" is a quiet movie, shaken from time to time by ripples of emotional turbulence far beneath the surface. It is about things not said, opportunities not taken, potentials not realized, lips unkissed. All of these elements are guessed at by the filmmakers as they regard a painting made in about 1665 by Johannes Vermeer. The painting shows a young woman regarding us over her left shoulder. She wears a simple blue headband and a modest smock. Her red lips are slightly parted. Is she smiling? She seems to be glancing back at the moment she was leaving the room. She wears a pearl earring.”

Scarlet’s wonderful promo shoot photo, and her appearances in the film, channel Vermeer’s iconic image of the young woman. However, Scarlett is even more beautiful, ethereal, and arguably more omnipresent in digital media than Vermeer’s original painting. It is clear to me that Dall-E sought out the images of Scarlett Johansson with which to create Ron’s request.

In reading more about AI, it seems that the enormous computing power and data resources, the raw reference materials, are more devoted to solving rules-based problems that are significantly objective. Recent advances include computing, biotech, chess, and technical writing. I submit that all these disciplines are governed by structural components. And, while improvement has been rapid in these arenas, art poses a different situation. I am reluctant to call it either a challenge or an opportunity as I think it is an area best left to humans. Why do I say that and why is art different? Art is amorphous, messy, and highly subjective, a product of emotional and sometimes irrational inspiration, abstract thought, and motivation driven by an internal creative force rather than a problem-solving objective.

I think it imperative to compare, and contrast, the human process against the AI methodology. From my perspective, the human artistic process involves examination, instruction, absorption, inspiration, intention, initiative, creation, introspection, evaluation, and insight. It’s a complex process that starts in the mind with an active brain, observing, thinking, sifting and sorting, and finally activating hands and media to bring cognitive inspiration to life. And, it is not a one-off activity by any means. Creating visual art is an iterative creative process that involves tinkering as one proceeds, always asking questions. “Does this need more darks to bring out the contrast and create more drama?” Writing shares this same process. Creative writing relies on reflection, insight, and rewrites. Truman Capote is attributed as saying, “Writing is rewriting.”

And, once any work of visual art is completed, the process becomes circular as we start again with inspiration and a new concept. Visual art is envisioned in the human mind and manifested by the human hand. Most importantly, the genesis of visual art is a result of a sentient mind.

The AI process shares some of these steps but neglects others. The AI exists as an instruction set that does not consider what it might do next. It does not cogitate on what might be. There is no inspiration. It simply waits. Once it receives written instruction, it goes to work executing a process designed by humans to search via keywords to access images that already exist in a database of published works (created by humans) and assembles these objects in relationships based on the machine’s interpretation of the words and structure of the written request. There is no visualization as with a human artist. While there is a neural process, the machine is not thinking as we do. It is a function of electrons, a complex activity executed in bits and bytes.  I suggest that there is instruction, association, and compiling to create the requested output. But beyond that, nothing. AI is a linear process that, unless instructed to be repeated, is simply a one-off. The machine does not examine its work once finished and question itself, “Could I have done better?”

I tried a smackdown between me and the tool. I called it DALL-E versus Bill-E. My instructions to DALL-E were my description of a pencil sketch that I created earlier this year. “Pencil drawing of Beargrass flower blooms on the steep south-facing slope of Bandera Mountain, looking across an inversion layer of clouds to Mount Rainier on the distant skyline.” I chose pencil as the medium to further challenge the AI as I assumed there would be less digital source material available. The results speak for themselves.

I submit that many of those gushing about the bright future of AI are not artists. And, as ‘not artists’ I doubt there is much understanding of what constitutes art and the creative process. AI has no current capacity for self-reflection in visual art, and no emotional ability to feel. The final hurdle for AI is achieving sentience and until then we own fine art. Our internal process as sentient beings set us apart and enables us to create great literature and visual art.

The human artist begins with motivation and relies on a subject as a source of visual information from which to create. The AI (computer code created by humans) begins with instruction (supplied by a human) and relies on digital source material (created by humans) to interact with and interpret. So, the resulting ‘creation’ more closely aligns with plagiarism than not. The machine does not create. It assembles. It is more a mechanic than an artist. And, if the source material is limited, one could submit that the results will be as well. Again, sentience sets us apart. We, humans, can feel, imagine, dream, stand in awe of a sunset, and be inspired to create that which does not yet exist. And that makes us valuable. And irreplaceable!

And despite my analysis, I still ask myself if I am afraid of AI. When it encounters the domain of art, I realize that I am somewhat fearful. I fear a dumbing down of both the process and the machine-created output. That is bad for everyone. While the appreciation of art is usually considered innate it is also a cultivated skill, a way of seeing, with an educated and critical eye to both learn, interpret, and reflect upon. Human art speaks to mankind. What kind of message will computer-generated art deliver? And worse, if AI gets even better at art will it both devalue the art and the artists themselves, thus disincentivizing human artists. I can just imagine artists throwing up their hands and saying, “Why bother?” At this juncture, it seems to me that fine art represents the ‘last frontier’ in the quest for AI to infiltrate our lives.