Can Artificial Intelligence Be an Impressive Artist?
Nice to meet you, I am Venny. Some AI thought that I am this sweet-looking middle-age man. However, actually, I am a 25-year-old woman.
This doesn’t bother me, since this program called PULSE also thought that my beloved cats are white men.
Of course, the premise of this program is recreating human images. They didn’t expect cats. However, you can see that the eyes, the ears, and the shape of the faces of the depixelized cats are almost identical. Even when you observe all the three pictures including the picture depixelized from a human picture, the eyes and the noses are almost identical. This shows that the original dataset used to create these images is not very diverse. The dataset is crucial in determining the outcomes of AIs based on machine learning.
Here is another example of images generated by AI. The three images below were generated with a machine learning framework called generative adversarial network. Among all the images, the cat on the left side looks quite realistic. The middle one looks quite unusual, and the right one probably can’t be considered a cat. They all seem fluffy, but this example also shows that AIs are not perfect. However, this website also generates a lot of cat images that look natural. In case you need non-existing pretty cat images, you can pick your favorites by yourself. You can have tons of them.
Creating a successful AI is not a piece of cake. For a group project during the Interaction Design class at Keio University SFC held in 2020, our group tried to create a program that detects whether your posture is good or bad. However, because we only used pictures of our group members, of which all had black or dark brown hair, it does not work very well with other hair colors. You can try the website here: https://postures.glitch.me/
As we saw in the cases above, artificial intelligence has its weakness, and many aspects have to be considered to create an effective AI. Can it create art despite these challenges?
Opinions and Researches on AI and Creativity
The main topic of this article is whether AI can be an artist. “Artificial Intelligence (AI)” became a buzz word, but basically it’s an emerging technology that is intended to recreate some capabilities of humans.
Nowadays, probably there are few people who view photography as an AI. However, imagine you are living in the era when the only way to create a picture of yourself was by drawing. Not many people could paint or hire a painter. Photography must have been viewed as a huge innovation.
History repeats itself. Photography, which was once not welcomed by all people, became a significant way of creating art pieces. As written in the paper Can Computers Create Art? (https://arxiv.org/abs/1801.04486), “many people believed that photography could not be art, because it was made by a mechanical device rather than by human creativity. Many artists were dismissive of photography, and saw it as a threat to “real art”.
The author of the paper disagrees that AI can be categorized as an artist. However, it notes the importance of technology in art creation. It says that technology contributes to innovation, which is an important factor for vital art.
There is another research CAN: Creative Adversarial Networks, Generating “Art” by Learning About Styles and Deviating from Style Norms (https://arxiv.org/abs/1706.07068), which concluded that their experiment results showed that human subjects could not distinguish art generated by the proposed system from art generated by contemporary artists and shown in top art fairs.
They propose a computational creative system for art generation without involving a human artist in the creative process, but involving human creative products in the learning process. The computationally feasible theory developed by Martindale was used. It stipulates that artists try to push away habituation and to increase the arousal potential of their art by style breaks. Based on this, researchers propose a modification of GAN (Generative Adversarial Networks) to generate creative art by maximizing deviation from established styles while minimizing deviation from art distribution.
New art generating agent called CAN (Creative Adversarial Network) incorporates the most significant arousal-raising properties for aesthetics such as novelty, “surprisingness”, complexity, ambiguity, and “puzzlingness”, as defined by Berlyne.
As I pondered over artificial intelligence and creativity, one old saying came to my mind. The saying goes like this: The beauty is in the eye of the beholder. This means that the public and the art experts have the final say if a particular piece is surprising and valuable.
Can AI Artists impress humans?
My opinion on whether AI can create are or not is “yes, but not completely in the same way as humans”. As written in the chosen materials, when AI creates art, it is based on an algorithm.
I do think that AI can create new/surprising/valuable art/design/music. However, this would happen only when the art-generating AI is proactively used by humans. In my opinion, the future of AI usage would be very similar to that of photography. As written in the first material, there were people criticizing photography. Despite some oppositions to photography, as time passes photography became more popular for the generation of portraits. In addition, humans have created various ways of expression with photography and now there are photography museums and photography competitions. This shows that many people came to appreciate the technology called photography. Maybe in the future, we will see artificial intelligence museums and competitions.
It has to be noted, however, that photography is a technology to create a 2-dimensional copy of the 3-dimensional objects. The original focus of photography is to create a picture as similar as possible to the original object. The practice of creating creative AI is rather the attempt to capture the algorithm used in creation. Rather than merely duplicating the existing thing, creative AI technology tries to copy and expand the creative process itself.
Both the strength and the weakness of AIs come from its difference compared to the creative process. AI is especially strong in creating “new/surprising” art pieces because it can have a unique process of generating art that is not possible or not yet existing in humans. It is also good in creating “typical” art, because it can find patterns among many examples.
For example, there is a service called “ecrett music”, which is used for generating royalty-free music. I tried it and uploaded the video online (https://youtu.be/hT4QLjcQ79U). It can make good background music. It might not be very novel, but it is useful. Probably, many people will find some of the music generated by this program pleasant. You can also make some changes in the patterns that generate music.
Finally, I think that there is one crucial difference between human artists and creative AIs. It is that AIs don’t have a reward system like humans do. Humans have a reward system in the brain, which gives them the pleasure and the drive to create art. How can AIs create art if it is not pleasant or meaningful for them? Perhaps humans can create an artificial reward system for the brain, but I doubt that it would be identical to that of humans. Photography and AIs could not have emerged without human effort. It is also up to humans whether AIs can create art. A camera will not generate art if it’s just sitting and collecting dust. The same goes for AIs. Like photography, “how” and “why” to use the technology will play a vital role in creating impressive art.
An important attribute of art is its uniqueness and scarcity defined by the creator’s own personality, feelings, and struggle to bring the accomplished piece into life. If we have a stream of endless art pieces generated by AI there is concern that the art pieces will cease to be surprising and valuable. The case may resemble that of diamonds. The most precious of the stones on planet Earth has been cherished by millennia and holds the highest value. And then, human technical genius succeeded in creating artificial diamonds — purer, more transparent, and more abundant than natural ones. However, their value is very low. Will this be the fate of computational creativity? Only the future will reveal.
