AI and Creativity
Can AI become an artist? Can AI be creative?
The first official approach to the arts by AI was Portrait of Edmond de Belamy in 2018. As soon as the product was released on auction, the two adversarial opinions fiercely conflicted. One said it is a revolutionary step of AI generating a solid artwork and the other said this should not be counted as art because the machine can not have the ability and qualification to create arts. Still, these two thoughts are controversial today. A Chinese contemporary artist, Ai Weiwei quotes “Creativity is part of human nature. It can only be taught ‘’ But, is it true that creativity belongs to only humans, not AI?
People who claim that AI can not be creative emphasize that creativity is an exclusive human property. But before moving on to the talk about AI creativity, what “human creativity” really is should be clarified first. Is the way humans create different from the way AI creates? The answer to this question is ‘not really’. Artworks are created through one’s life data. Artists create art by using their experiences, memories, and ideas. In the process of combining all the data together, the new combination can be found. It is reproduced but the way he or she connected. Human creativity occurs in the mechanical act of combining and reproducing ideas. If we consider humans’ process of creation, it sounds quite similar to AI creative process. AI also has probabilistic algorithms that build as much or more data than humans and combine them exquisitely. It is also able to present creatively-looking results through the process of combining data. There is irony and ambiguity when the process of humans and AI are similar. Considering their similarity, AI’s creation process should be considered as one of the creative processes.
However, even though the creative process is similar, AI creativity can not be identical to human creativity for several reasons. First, AI can process and learn within the boundary of the given data. The main difference between humans and AI is the resource(原料) which takes a huge part in forming creativity. Even though the results made by AI seem creative and inventive, they are reproduced from particular data set so it is hard to say that AI ‘create’ them on its own. Also, AI’s creation holds copyright and plagiarism issues but solutions have not been clearly identified yet. This loses specialty and uniqueness which are the main conditions for being creative.
Second, unlike humans, AI can not change or develop voluntarily. For instance, a computer program called AARON is known for producing abstract drawings. AARON was able to draw and paint humans and objects because it has knowledge of the positions of body parts and relationships between objects and humans. It also knows the harmony of colors so exquisite arts were produced. However, the limit AARON couldn’t pass was that it can create arts that correspond to the existing version only. In other words, it is impossible for AARON to be “I’m tired of abstract painting. Shall I draw landscape today?” because AARON is specialized for abstract painting. In addition, the program can not voluntarily modify the model so Cohen should intervene to change or develop the model. The fact that AARON can not escape the structure instructed by Cohan shows why AI can not be creative like humans.
Therefore, the current AI can be approved for being used to assist in the creative process. In fact, AI is standing out in assisting the creative process of humans. The picture below is from the CartoonGAN model that converts photos to graphics. It sounds ordinary but converting according to illustrators’ style is a unique feature of this model. It provides four different styles — Hayao, Hosoda, Paprika, and Shinkai. Each style provides different types of color, brightness, and atmosphere. The fact that the model provides different illustration styles felt creative and inventive to our group. In the future, if more data with diverse illustration types are gathered, not only Japanese anime but also Disney, Pixar, the developed model can present different results.
Our group also generated a GAN model to make dresses that do not exist. Due to time and money limitations, 600 steps were available and the result did not meet the expectation. However, some photos featured the shape of the dress and car. The results are unintentionally very new and bizarre, giving the audience and designers creative inspirations. Moreover, the GAN model shows that the process of AI creating is way faster and convenient than the process of humans. If designers try to design new fashion dresses, they need to start from data collection, meetings, evaluations, etc which would take more than a few months. However, the AI machine can learn a massive amount of the given data by itself and present results within a few days. The result will be different and new for each trial represents the limitless potentials of AI productivity. Although AI’s ability can not be approved as creativity, it can support artists to give inspiration and stimulate their creativity.
Because the whole concept of the AI machine is learning the given data, there is an avoidable limit to create something outside of boundaries and dimensions. However, to challenge creative AI, AI engineers have focused on implanting randomness in AI machines. As people describe creative ideas come from ‘Ah-Hah moment’, randomness is a necessary condition for creativity. Creative works are sometimes created by random chances and accidents, like penicillin discovery by Alexander Flemings. If AI can process how to choose by chance, in the infinite process of learning data, AI with random ability can create infinite combinations of which can surpass the existing limits.
In conclusion, I believe the current AI can not be creative by itself but it can be assistance for the creative process. Although the creation process of both humans and AI machines is similar, the fact that AI can not collect data and learn voluntarily makes it not creative like humans. However, AI has several advantages. It unintentionally creates new and novel outputs that can stimulate human creativity. Also, it takes less time, energy, and money during the creation process. With stable and untiring strength, AI can create different products for each trial. AI industry keeps challenging itself to install creativity in AI machines. AI has not reached creativity but its creations can be the inspirations for the human creative process. To prepare for the day when AI creativity matches human creativity, questions such as “How to distribute between AI works and human works” and “How to enforce AI’s intellectual property rights” should be addressed.
