This piece belongs to a series of student posts written during the Neuroscience and Behavioral Biology Paris study abroad program taught by Dr. Karen Rommelfanger in June 2018.

By Coco Cao

An example of AI-generated art Image courtesy of Flickr

An article published on New Scientist entitled, “Artificially intelligent painters invent new styles of art” has captured my attention. The article discussed a recent study conducted by Elgammal et al. (2017), who developed a computational creative system (the Creative Adversarial Network) for art generation based on the Generative Adversarial Network (GAN), which has the ability to generate novel images simulating a given distribution. Originally, GAN consisted of two neural networks, a generator and a discriminator. To create the Creative Adversarial Network (CAN), scientists trained the discriminator with 75753 art works from 25 art styles so it learned to categorize art works based on their styles. The discriminator also learned to distinguish between art and non-art pieces, based on learned art styles. Then, the discriminator is able to correct the generator, a network that generates art pieces. The generator eventually learns and produces art pieces that are indistinguishable from the human produced art pieces. While ensuring the art piece is still aesthetically pleasing, CAN generates abstract arts that enhance creativity by maximizing deviation from established art styles. 

After learning about AI’s ability to be “creative” and generate art pieces, I was frightened. Unlike AI’s application in a scientific context, AI in an art context elicits human feelings. Is it possible that AI artists could replace human artists in the future? Considering the importance of the author’s creativity and originality in art, the critical ethical concern regards the individualism of AI artists. Can we consider the art pieces generated from AI as expressions of themselves? 

In 1738, Jacques de Vaucanson, a French watchmaker, generated a life size mechanical duck with feathers. The mechanical duck could eat, move and flap its wings. Therefore, audiences refused to believe it was artificial, since the mechanical duck exhibited all of the behaviors of a real duck (Glimcher, 2004). If the mechanical duck was a real duck, like audiences believed, then the behaviors must have been generated by the mechanical duck. However, this situation may not be the case. The mechanical duck was likely programmed by Vaucanson, causing Vaucanson to be the generator of the mechanical duck’s behavior. In the case of an AI artist, Elgammal et al. (2017) stated in the paper that CAN involved human creative products in the learning process while the creative process was carried out by AI. Therefore, the AI creative process was hugely dependent on pre-exposure of artwork created by humans. Does this mean the artwork was originally created by AI? I don’t think so. During my recent visit to the “Artists and Robots” exhibition, which was held in the Grand Palais in Paris and presented the applications and implications of AI in arts, I noticed that there were some robot paintings with the human artist/programmer’s signatures on the paintings. In this case, the credit of the AI-generated art pieces still belonged to the human artists and programmers. Therefore, AI are not considered as an individual in an art context for now. 

Vaucanson’s duck along with two of his other creations Image courtesy of Wikipedia

Moreover, because AI is programmed to think like humans, does it mean humans already understand the biological basis of creativity and individualism? Moreover, are we able to program creativity and individualism? Currently, research suggests that three brain networks, which are the default mode network, the executive control network and the salience network, are related to creativity. Those three brain networks are scattered through the frontal and parietal cortices (Brenner, 2018). Regarding individualism, Chiao et al. (2009) suggests that neural activity in medial prefrontal cortex positively predicts individualistic and collectivistic views of self. However, all brain areas are interconnected, and we still don’t know the specific neuronal interactions involved in creativity and individualism. Without actually understanding the human neuronal basis of creativity and individualism, we are unable to program creativity and individualism into AIs. Therefore, art pieces generated by AI are not original. 

Other than the originality of art, the meaning of art is also crucial. We can evaluate meanings of art in two contexts: the meaning to the audience viewing the art and the meaning to the artists who created the art. Ted Snell (2018, May 04), the Cultural Precinct director of University of Western Australia, concluded that the evaluation of art depends on audiences’ knowledge and experience. Considering the subjectivity in art evaluation, what is the meaning of art to an artist? 

Image courtesy of Pixabay

There are different kinds of art. As a dance minor student, I am more familiar with performing arts. After years of training in classical ballet, I witnessed my technical improvements as I put more effort into my dancing. However, dance is not only about physical growth in overcoming technical challenges. It also helps me to grow mentally: as I started to accept my imperfectness and I became more humble and persistent in training. If art brings artists mental growth, we still have no way to measure the meaning of growing for AI. Until now, AI’s learning process has been largely guided by humans and AI is currently developed under a human societal context. Therefore, if we consider AI as an individual but biologically distinct from a human, do human societal values apply to AIs? 

So far, we can neither consider those AI artists as individuals nor do we have any way to measure the AI mental growth experienced while producing those works. However, we cannot neglect the infinite possibilities of art pieces generated by AI. Also, considering the immortality of AI, AI could someday exceed us by continuously learning and improving. “Robot” originates from the Czech word “Robotnik,” which means slave. It is possible that AI could “enslave” us in the future. While it is interesting to experiment with AI’s ability in creating arts, we need to evaluate the consequences of accepting AI art pieces. Nevertheless, it is truly fascinating to see the artworks created by an AI artist! 

References: 

Baraniuk, C. (2017, June 29). Artificially intelligent painters invent new styles of art. Retrieved June 14, 2018, from https://www.newscientist.com/article/2139184-artificially-intelligent-painters-invent-new-styles-of-art/ 

Brenner, G. H. (2018, February 22). Your Brain on Creativity. Retrieved July 5, 2018, from https://www.psychologytoday.com/us/blog/experimentations/201802/your-brain-creativity 

Chiao, J. Y., Harada, T., Komeda, H., Li, Z., Mano, Y., Saito, D., Iidaka, T. (2009). Neural basis of individualistic and collectivistic views of self. Human Brain Mapping,30(9), 2813-2820. doi:10.1002/hbm.20707 

Elgammal, A., Liu, B., Elhoseiny, M., & Mazzone, M. (2017). CAN: Creative Adversarial Networks, Generating” Art” by Learning About Styles and Deviating from Style Norms. arXiv preprint arXiv:1706.07068. 

Glimcher, P. W. (2004). Decisions, uncertainty, and the brain: The science of neuroeconomics. Cambridge, MA: MIT Press. 

Réunion des musées nationaux – Grand Palais. (n.d.). Artists & Robots. Retrieved July 5, 2018, from https://www.grandpalais.fr/en/event/artists-robots 

Snell, T. (2018, May 04). On judging art prizes (it’s all subjective, isn’t it?). Retrieved June 14, 2018, from https://theconversation.com/on-judging-art-prizes-its-all-subjective-isnt-it-38430