AI as an Artist?
In order to determine whether or not AI can be creative in itself, one must consider the definition and boundaries of “creativity” itself. There are cases where creations made completely by an AI system are treated as examples of a creative process at work, such as paintings and artworks made through a GAN system. There are also cases where an AI system acts as a source of inspiration that leads to creative work, such as the generation of patterns, which can be used within a design of any kind.
There are three simple categories to creativity according to Margaret A. Boden from the University of Sussex: combinational, exploratory, and transformational. Combinational creativity can be described as combinational in theory — ideas are formed from familiar concepts and ideas. Exploratory creativity focuses on the generation of new ideas by exploring the boundaries of a certain idea and bring out new potential. The last one, transformational creativity, attempts to transform concepts through a new dimension, resulting in something completely different.
It can be seen through these categories that current AI systems usually fall under the first two categories. Data is fed to the system, and something new is created out of the set data that has been received. The third category of creativity, however, can be argued that it requires human effort. The ability to create a concept that reaches outside of its own set dimensions requires human intelligence. In this way, the current state of AI can be creative within a limited boundary. What is most important about this concept is that the AI can be vastly creative in a combinational sense, in that it can achieve faster results compared to humans. Having a creative concept that is created through an AI-assisted process will be significantly greater than one without in terms of produced data sets.
In this example below, my group and I created a GAN model using RunwayML that generates an image that mimics Japanese manga covers. These generated images would have been difficult to replicate solely on human effort. On top of that, the generated images mimic the overall color schemes and location of the lettering for the manga titles but do not spell anything out. Although this model could be improved upon by separating various aspects of the manga covers into variables and running a GAN model through each of these variables, the results from this model itself show images that are unlike anything the world has ever seen.
Although the GAN model was unsuccessful at creating a “fake” manga cover, it shed light on the aspect of creativeness within AI. Despite the images produced by the GAN model is unique, it is difficult to declare these images as “creative” in nature (at least, in the transformational perspective). If these images were used as, for example, an album cover or a book cover, then perhaps it can be considered creative in the sense that we, as humans, have used the images and placed creative meaning into it.
A prototype GAN model for this manga cover maker used birds as a base model, resulting in the manga cover images undertaking some bird-like aspects. The colors seem to take on a specific color scheme, with patterns of feathers seen in some of the images. In cases such as these, although unintentional, it can spark new ideas for designs. In this way, humans can take advantage of the boundless nature of AI’s combinational and exploratory creativity and use it to create something transformative in nature.
A more direct example of AI being used as a tool for creativity are programs that revolve around the camera.
The website above utilizes the camera to detect what kind of hand gesture is being made. In response, the program would throw out a sign that makes it so that it would lose every time in rock-paper-scissors. The original model was a rock-paper-scissors machine that would always win against the player. Although relatively simple, this website illustrates how beneficial AI can be in creating new and exciting concepts. Although the program itself is not necessarily creative, it can lead to creative creations when combined with human intelligence.
This project, in turn, branched into another interactive page where the camera detects the facial gesture that is made and sends feedback in the form of an emoji. The concept of this program is to show how simple it can be to integrate facial detection and recognition through AI. The creative aspect of this requires human interaction, however.
With all of these observations, however, the argument of whether or not AI can be creative in an artistic sense can be traced back to what creativity is in the first place, and what art is in itself. Seeing AI designs by itself is different from seeing a design that is AI-assisted. Does art require a deep and meaningful background for it to be considered valuable? By that logic, AI designs can never be considered creative. Although the topic of “art” can be subjective, modern-day AI technology is similar to a pen, in that by adding a human aspect to it, the tool becomes a means to creative output. The bigger question lies in the future of AI, however. As AI develops into something that is more and more difficult to distinguish between human creation, the question of whether or not AI can be considered an artist must be reevaluated.
