Computing

Tweet

AI and Creativity

Can AI Create Truly Original Art?

Image Source: Marcos Silva/Stock.adobe.com

By Becks Simpson for Mouser Electronics

Published December 27, 2022

It seems like every other day a new, highly powerful artificial intelligence (AI) model is released that can produce extremely impressive content that rivals the work of human artists. All of the creative fields, from visual arts and literature to music and video, are being disrupted with models like DALL-E, Stable Diffusion, and GPT-3 (Generative Pre-trained Transformer 3). As these AI models win competitions and produce artworks that sell in auctions for half a million dollars and are often easily mistaken for works done by their human counterparts, the question of whether AI can really create truly original art arises: Is AI genuinely creative? That question itself opens a can of worms that leads to deeper thoughts about what constitutes originality and creativity in the first place and whether something needs either of those to be considered art. Further down the rabbit hole, the notions of what can and should be considered art are thrown into debate. By virtue of how AI models are trained using reams of past data, human outputs, and history, to say they are creating truly original works might be a stretch, but the way that elements are combined and produced could be considered creative depending on the definition of creativity. However, despite that uncertainty, given the popularity of these models for producing new visual art, literature, and music, creativity and originality are seemingly no longer prerequisites for creating art. Some argue that without a human element of intent and imbued meaning, AI-generated art is not true art, while others state that since human input is still needed, art created by an AI model is genuine.

Art = Creativity; Creativity = Originality, but Is It That Simple?

Ask almost anyone or run a quick Google search: Creativity is often paired directly with originality. Creativity is seen as the act of using imagination to create totally new ideas, typically of artistic expression, that produce novel and unique interpretations. To be creative, one must likely develop previously unseen concepts and outputs for something to be considered art. People primarily opt out of creating art owing to this notion that originality is supposed to go hand-in-hand with creativity; their ideas aren’t “creative” because they’ve emerged before. However, psychology has uncovered three kinds of creativity: Exploratory, transformational, and combinatorial (Ness Labs).

Exploratory creativity involves trying new things within a given set of rules, while transformational creativity involves the creation of new rules (or even ignoring certain rules) within the genre to generate seemingly impossible, but highly creative, new ideas. Most novelists engage in exploratory creativity, while the Cubists, for example, were at least in part transformational. Combinatorial creativity, however, involves combining existing ideas, motifs, and elements in unexpected and interesting ways. While people most often think of exploratory and transformational creativity as the pinnacle of artistic expression, it is combinatorial creativity which actually produces most new ideas. Of course, most great art was created through a combination of the three. Even among the transformational artists, influence/combination was a vital element to their creativity.

The debate around creativity and originality aside, another important piece to consider when it comes to defining art is the role of intent and awareness of meaning. Art is supposed to convey something from within the artist—perspectives, emotions, or worldview—and create a connection with the experiencer, who then interprets those elements. The interplay between the artist and the experiencer can be complex, too. Experiencers of art can interpret things differently than the artist intended, including identifying meaning in elements that weren’t meant to carry any. Conversely, the deep meaning imbued in a work of art can also be totally overlooked or misunderstood by the experiencer. In cases where the artist’s or author’s original intent cannot be identified (e.g., because the person has passed away), the only meaning left is whatever the experiencer discovers. All of these points muddy the water when determining whether AI can truly create art.

How Close Does AI Come to Achieving Creativity and Originality?

In terms of pure originality, arguing that AI is making anything truly unique and novel is more difficult because AI literally is building on the shoulders of giants. “Originality is undetected plagiarism,” said author William Inge. Given that all of these impressive new models are typically trained on vast swaths of publicly available data from existing artworks to classic literature, these influences almost certainly will leak through in some way, even if it’s not obvious. For example, the State Fair–winning entry by Midjourney is said to be reminiscent of Gustave Moreau, a late 19th–century artist associated with the Decadents, who influenced Edgar Degas and Henri Matisse (Washington Post). Additionally, the human intervention usually required in the form of a prompt is also a limiting factor on the originality that AI could demonstrate. In cases where the prompt is highly detailed, most of the originality comes from a human with the initial idea.

Creativity, though, is less cut and dry. Of the three types of creativity, many of the pieces produced by AI across different artistic domains could easily be considered to show combinatorial creativity. Particularly, AI-generated works have shown remarkable ability to combine ideas and concepts into unique and interesting perspectives that their human handlers claim they couldn’t have created without the AI’s help (The Conversation). Even when the original prompt is from a human, AI models like Stable Diffusion can take some creative license to produce images of varying styles and content. These models can also suspend reality in unexpected ways such as in “Coffee Cup with Holes,” a surreal scene depicting a mug with holes and coffee that inexplicably stays in the cup. With the debate around these two points in mind, another important question remains.

Regardless, Is It Still Art?

Human artists can almost certainly be accused of the same shortcomings as this new generation of AI models. Most artists build on historical work and current themes, which are then termed “influences,” in much the same way that AI models are trained. So, if neither humans nor machines are producing 100% original and creative outputs, can their work be considered art? For any artistic pursuit, the creator’s intent and the experiencer’s interpretation are also considered vital. The dance between the expression of the artist’s thoughts, emotions, and intuitions with the evoked thoughts, emotions, and reflectivity of the experiencer makes something a work of art. By this kind of definition alone, what AI is creating cannot really be considered “art” because the algorithms generating the piece are not yet sentient or capable of displaying individual intent. AI models’ inspiration is not internal but rather comes from the human world they’ve learned about or prompts directly from humans. They don’t have thoughts and emotions derived from their experience in the world in the same way humans do.

Having said that, however, others claim that art is in the eyes of the experiencer—if the work of art evokes some kind of thought, emotion, or reflectivity in the person experiencing the art, then it can be considered art. This sentiment in particular is interesting because a sufficiently vague prompt that asks a question but does not suggest an answer can yield results that on the surface indicate the AI has something deeper to say. The more expressive and abstract art from such inputs often has enough elements and motifs to evoke an emotional response in the experiencer and cause reflection on a potential meaning. The AI models cannot explain their artistic choices, resulting in a similar situation to artists who can no longer explain the meaning behind their art, which is left to experiencers to decide. Other definitions of art indicate that both sides of the equation are needed. As the old adage “if a tree falls in the woods” implies, if a painting hung in a gallery has no one there to see it, is it still art?

Human and AI—Making Art Together

While the jury is still out on whether AI can really make genuine art, a recurring theme in this heated debate is that AI artists are more like co-pilots or tools in the artistic process. The lab in Japan that was shortlisted for a national literary prize with a novel co-written by AI describes the model as the co-author that gives input and direction along the way but contributes meaningfully to the final result. Others describe AI artists as being more like tools for helping human creators. The notion of “medium and skill” as important for artists is being shattered quickly with the AI now available. Anyone with an imagination and some bright ideas can produce images that look like watercolor, oil paintings, or charcoal; music that sounds like it came from a professional band; or prose that mimics the classics. The barrier to entry of being an artist and making art is lower when people who can't paint, draw, or write but have an imaginative spark can work with AI artists to produce something that neither of them could do separately. Maybe in this sense the art that AI creates can be defined as “real art” because, in the end, it has the human element that drove its creation in the first place.

Conclusion

The question of whether AI can create truly original art is a complex one that depends on the definition of creativity and originality. AI models can produce impressive content that rivals the work of human artists, but it is unclear whether their work is truly original or creative. On the one hand, AI models are trained on vast swaths of publicly available data from existing artworks to classic literature, which makes it hard to argue that they are making anything truly unique and novel. On top of that, human input is still needed for AI-generated art to be considered genuine. On the other hand, many of the pieces produced by AI show a remarkable ability to pull ideas and concepts together into unique and interesting perspectives that their human handlers claim they couldn’t have come up with without the AI’s help. In the end, it seems the jury is still out on whether AI can create truly original art. What is becoming obvious, however, is the way in which AI is transforming how humans create art, especially by making the field more accessible to those with imagination but without practical artistic skills. Perhaps in the future, the question will no longer be whether AI can create art but whether AI or humans can create great art by themselves?

Becks is a technical lead for machine learning at Imagia, a Montreal-based startup putting AI in the hands of clinicians to drive medical research. In her spare time, she also works with Whale Seeker, another startup using AI to detect whales so that industry and these gentle giants can coexist profitably. She has worked across the spectrum in deep learning and machine learning from investigating novel deep learning methods and applying research directly for solving real world problems to architecting pipelines and platforms to train and deploy AI models in the wild and advising startups on their AI and data strategies.