By Thilina Balasooriya
For centuries, philosophers have contended that one fundamentally unique quality of humans is their innate creativity. Often paired with intelligence, natural artistry—the ability to conjure emotional and imaginative ideas, often from thin air—is often said to distinctively define humankind. For instance, Socrates believed that poets were divinely inspired, while Imannuel Kant asserted that the imagination and genius of artists followed no rules. However, Aristotle had a different theory; he believed that the poet was rational and goal-oriented in his execution, using almost algorithmic precision to evoke the desired response from the audience. This idea, though unpopular at the time, is taking new meaning in the present-day. A new dilemma is emerging concerning this seemingly inherent human characteristic: can artificial intelligence be creative?
AI and Images
First, take a look at these three paintings:
Only two of the above are human-made. Can you guess which is AI-generated?
If you guessed the first one, you would be correct. But was it a guess in the dark? Were there any defining characteristics that shouted “artificial”? If so, consider applying to be a judge at your next local art competition: this AI-generated painting, called “Théâtre D’opéra Spatial” was created by Jason M. Allen, an AI-art enthusiast who entered this painting, made using the software Midjourney, into the Colorado State Fair art competition in 2022. His (their? its?) work ran undetected by panelists and eventually won the blue ribbon, garnering backlash in the artistic community, who felt they had been cheated.
His image was created using Computational Creativity (CC), a novel concept in the artificial intelligence space that describes the algorithms designed to emulate creative and artistic processes. According to Ramón López de Mantarás at the Artificial Intelligence Research Institute in Bellaterra, Spain, “Computational creativity is the study of building software that exhibits behavior that would be deemed creative in humans.” Similar to many other artificial intelligence algorithms, CC functions by feeding computers large sets of training data (in this case, artistic data like photos, musical scores, paintings, etc.) and use iterative algorithms to analyze the data for patterns, generate predictions, and ultimately create new solutions that fit within the constraints of those patterns. Integration of visual art and AI is one of the fastest growing topics in computer science, with new research being published in places like the Association of Computational Creativity’s Journal, which is launching just this year.
A common real-world example of CC is Dream, an app developed by Wombo, that takes a user input picture combined with a selected artwork style and literary prompt, which it then uses to create an AI-enhanced version of the picture. One particularly interesting feature about the app is that the same inputs can yield different versions of the picture. Combining the picture below of Riverside Park with the art style “Cartoonist” and the prompt word “Warrior,” you can see the transformation of this picture into art using the AI algorithm. Although the “warrior” (abstracted from the traffic cone on the left picture) lacks concrete elements like facial features, you can see general elements like a sword and shield that have been newly generated, as well as a cartoonistic rendition of the background trees and environment.
AI and Music
With implementations like Google’s reverse image search, image analysis is quickly growing in familiarity and popularity. However, visual art is not the only kind which is being revolutionized using AI. A large subset within CC is musical composition: computers can now analyze scores of music and generate novel music that imitates the style of composers and artists around the world.
To understand how this works, we can take a deep dive into DeepBach, a music composition AI program developed by Gaetan Hadjeres and François Pachet of the Sony Computer science laboratories that reharmonizes musical scores by the famous Baroque era composer Johannes Sebastian Bach while maintaining style, upon being given a soprano melody. Bach is known for his creation of polyphonic chorales, short compositions sung in four harmonizing parts that articulate syllables simultaneously to create a cohesive and understandable piece. By defining certain data of the score like notes, rhythm, lyrics, and key signature, DeepBach generates parameters that are then inputted into the program’s internal neural network—a set of layered nodes of information that aggregate data to create predictive outputs. Specifically, DeepBach uses Recurrent Neural Networks, which analyze sequential data from the chorale before and after a given time to predict the best harmonizing note at that instant. It also incorporates two Non-Recurrent Neural Networks: one accumulates data about the current harmony at the given time while the other takes data from the other three neural networks to create a final prediction for the proper harmonizing notes in each vocal part. The results are enticing; in a test of 1272 subjects, 646 of whom were classical music lovers and 365 of whom were musical composition students or experts, more than half believed the computer generated music to be created by Bach himself.
What does this mean for the future of art?
If computers can fool humans by mimicking the creative genius of Bach, it is worth reconsidering what it means to be creative at all, and what implications this holds for the creative future of humankind. One may argue that AI art cannot be considered original because it uses training data of previous art to generate novel works; in the same sense, however, the human experience is molded by what we observe around us—we may not be as different as we think from our algorithmic counterparts.
AI, at least in the modern age, is far from perfect. Many AI-generated pictures and compositions can be identified due to recurring patterns and a lack of holistic structure that can be improved upon in human conceptualizations. But as artificial intelligence improves, many may not be able to tell the difference between AI-generated and manmade art. By then, even if we concede that computers may never be able to utilize the same subjective and indeterminate methods that humans apply in artistic creation, is there still any value in arguing that AI can never be “truly creative”?
Ultimately, artistic creation has been at the forefront of human innovation and culture ever since its conception and this is unlikely to change moving forward. But next time you encounter a picture on the internet or hear a new song you like, it’s worth wondering whether it is the work of a creative genius or artificial intelligence.