Can Computers Create Art?

Over the past five years, an explosion of research and development in the field of artificial intelligence (AI) has led to incredible breakthroughs in human-to-machine communication. Through a training process whereby AI chatbots (such as OpenAI’s ChatGPT) were fed billions of human-authored words, these massive systems are now capable of carrying on strikingly human-like conversations – conversant in virtually any dialect of any spoken language on Earth. As the debate rages on whether these AI entities are creating novel work or merely “artful plagiarists,” other research groups are training AI systems on the massive troves of artwork available digitally on the internet. In 2022, this led to the emergence of systems such as Stable Diffusion, Midjourney, and DALL-E: AI entities that can create, or “synthesize,” a brand-new image based on a text prompt. This represents a unique new frontier in the world of art; can works that were produced algorithmically by a computer program be considered art?

In Art Matters, Gordon (2024) admits it is difficult to provide a single, precise definition for art, instead including several, each capturing different aspects – an aesthetic experience, and a creative work among them. We may consider each of these individually through our exploration of the computer-as-artist conversation.

The aesthetic experience is typified by its expressiveness, beauty, and/or emotional impact to the viewer; in fact, due to its inherent subjectivity, the viewer represents a key participant in the measure of aesthetics. And here we encounter a significant challenge in assessing whether AI-synthesized images can be considered art: human bias. Through a series of experiments where subjects were asked to evaluate artwork without knowing whether it was generated by an AI or a human, followed by a similar exercise where the origin was disclosed, Norwegian psychologists Simone Grassini and Mika Koivisto (2024) found that there is a measurable negative bias toward AI-generated artworks. Elise Kjørstad reports on the first trial that “[p]eople had a clear tendency to prefer the AI images, and felt they gave them more positive feelings.” Further, when the subjects believed an image to be AI-generated, it was consistently rated lower than those assumed to be produced by humans, and classified as “uglier, and of lower emotional value.” (Kjørstad, 2024). However, the fact that AI-generated art can, when provenance is masked, be thought of as attractive on its own merits, satisfies Gordon’s definition of supplying an aesthetic experience.

The bias uncovered by Grassini and Koivisto may be attributable to a scarcity assumption; items produced mechanically are generally considered to be less valuable than handmade alternatives. (In addition, the notion of scarcity is exacerbated by the fact that an AI is capable of generating new images in seconds – a great deal faster than humans will ever be capable of producing.) More likely however is that this bias is due to a predisposition that art requires a creative spark – or what jazz musician Yosvany Terry calls “a sense of interplay, or the ability to react in the moment” – which, at least historically, has been a phenomenon assumed to be uniquely human (Mineo, 2023).

This topic of creativity dominates the debates surrounding AI-synthesized images’ qualifications as art. This “anti-creative” bias may stem from the somewhat dispassionate origins of the AI system itself. To build an AI capable of producing images, a scientist (not an artist) trains it by feeding it millions of images. The AI system evaluates, or “models” these images not on their aesthetics, but numerically – treating them merely as data. Similarly, an AI generates images by generating representational data algorithmically, not through the same compositional sensibilities that a human would. However the technique, though, in an abstract sense this is not altogether unlike the process that (human) artists undergo when training to become an artist: they internalize many different works produced by a variety of artists in their chosen medium, and create new pieces that include their own interpretation or reflect their own style.

As an example, consider the art of Picasso; mention of his name typically conjures up images done within the Cubist style he pioneered. However, in his early career he modeled his art after the Spanish old masters Velázquez, El Greco, and Ribera, as we see in his Realist-styled work, Salmerón (1895) [Figure 2]. In such a way, Picasso trained through a fusion of emulation and imagination, basing new work on the techniques and style of those that came before. While it is impossible as a human to (directly) relate to an AI, it may be that a computer’s own creative process – an essence of originality contained within its ones and zeroes – is nascent, waiting to develop.

Originality in art – its novelty, and uniqueness, where an artist creates something new and distinct from that which came before – indeed has a direct analogue in the digital realm. In spite of its potential for algorithmic precision, the typical process of creation through AI synthesis deliberately introduces randomness, and hence uniqueness as part of its recipe. This will cause an intentional variety with every output it produces, even if given the same starting ‘prompt.’ This element of chance – which may indeed be compared to the “spark,” or Terry’s reaction-in-the-moment – embodies a sort of creativity, in spirit if not absolute.

While AI image synthesis has seen major, unanticipated, and surprising advances over the past several years, it is evident that this technology is still in its infancy. New breakthroughs in the form of advanced models, or those capable of producing multimedia works (audio and video) are arriving at a breakneck pace. Irrespective of which side of the debate over computers-as-artists one comes down on, the lines will assuredly get blurrier. Even so, we are already struggling with a means of categorization for today’s pastiche of art and science. Whether viewed through Gordon’s lens of aesthetics or creativity, AI-generated images are challenging traditional beliefs about art, and these works can be viewed on their own merits as both aesthetic and original. Through its decidedly inhuman process, the art produced by artificial intelligence has arrived at something arguably human-like: a reflection of human society, and a humane embodiment of technology.

References

AI and the Aesthetics of Art: A New Era for Collectors | Psychology Today. (n.d.). Retrieved March 19, 2024, from https://www.psychologytoday.com/us/blog/the-mind-of-a-collector/202311/ai-and-the-aesthetics-of-art-a-new-era-for-collectors

AI-Generated Art: The Intersection of Technology, Creativity, and Ethics. (n.d.). Humanists of Linn County. Retrieved March 29, 2024, from https://hlcia.org/blog/2023/2/27/ai-generated-art-the-intersection-of-technology-creativity-and-ethics

Chamberlain, R., Mullin, C., Scheerlinck, B., & Wagemans, J. (2017). Putting the Art in Artificial: Aesthetic Responses to Computer-Generated Art. Psychology of Aesthetics, Creativity, and the Arts, 12. https://doi.org/10.1037/aca0000136

Epstein, Z., Hertzmann, A., & THE INVESTIGATORS OF HUMAN CREATIVITY. (2023). Art and the science of generative AI. Science, 380(6650), 1110–1111. https://doi.org/10.1126/science.adh4451

Gordon, P. (2024). Art Matters. Oxford University Press, USA.

Grassini, S., & Koivisto, M. (2024). Understanding how personality traits, experiences, and attitudes shape negative bias toward AI-generated artworks. Scientific Reports, 14(1), 4113. https://doi.org/10.1038/s41598-024-54294-4

Kjørstad, E. (2024, March 31). People liked AI art – when they thought it was made by humans. https://www.sciencenorway.no/art-artificial-intelligence/people-liked-ai-art-when-they-thought-it-was-made-by-humans/2337417

Mello-Klein, C. (2022, September 9). Artificial intelligence is here in our entertainment. What does that mean for the future of the arts? Northeastern Global News. https://news.northeastern.edu/2022/09/09/art-and-ai/

Mineo, L. (2023, August 15). Is art generated by artificial intelligence real art? Harvard Gazette. https://news.harvard.edu/gazette/story/2023/08/is-art-generated-by-artificial-intelligence-real-art/

Picasso: The Early Years. (n.d.). Retrieved April 1, 2024, from https://www.nga.gov/features/slideshows/picasso-the-early-years.html#slide_2