Art and AI

A year ago, a man by the name of Jason M. Allen entered the Colorado State Fair with a beautiful art piece. The juries were all over Allen’s artwork and awarded him the blue ribbon for emerging digital artists. Unbeknownst to the other participants, however, Allen didn’t create this artwork himself. This work was created by the Artificial Intelligence program made by the research lab MidJourney, which can create hyper realistic images of whatever prompt you give this machine. Allen took this victory to Twitter, where he received both praise and backlash from artists, leading him to hit the artist with a tweet: “Art is dead, dude. It’s over. A.I. won. Humans lost.”. But has A.I. really taken over our art already? Is there no sense in us humans creating art anymore? Have we really lost to the machines in this area of our lives already?

To start off, let’s look at the inner workings of an art AI. Developers of an art AI generally use certain datasets to train their models, which is very similar to how other artificial machines get trained. The dataset used for the model depends on the developer, causing the generated images to look different depending on which art AI you use for your prompt. These data sets consist of general stock photos of things like cats and trees, but also of other people’s artwork so that the art AI will be able to create prompts in certain artists styles, like the distinct style of Van Gogh, or certain time periods of art, like the Renaissance.

Through deep learning, the AI will learn to separate different objects from each other by comparing them on a huge space of variables. To illustrate this learning process a bit better, let’s say that the AI wants to differentiate between a rugby ball and a football. The AI will read the pixels from both the images with footballs and rugby balls and will come to notice that footballs are significantly rounder than rugby balls are. It now knows that footballs are rounder than rugby balls, but what if the next picture is a black and white balloon? This object is round and could by the knowledge of the AI be classified as a football, which we know isn’t the case. This means that the AI needs more variables to be able to differentiate between this balloon and a football, and even more variables to be able to differentiate between a football and an elephant. These variables are what are being formed in the process that we call deep learning, leading to an AI correctly being able to not just identify objects, but also styles, color schemes and time periods.

With these variables, the AI creates a dimensional space with more than 500 dimensions, meaning that each point in there representing an object is a point with 500 coordinates. When generating an image, the AI will trace the words from the prompt to a point in this huge 500-dimensional space and with this try to form a picture that is coherent to the human eye.

An interesting study into the attitude of humans towards AI-generated art has shown that we as humans have quite the prejudice toward art made by AI. In this particular study researchers had a group of people judge 30 artworks: 15 of which were created by humans and 15 of which were created by AI, or so the participants thought. Unbeknownst to them however, all the artworks were created by an AI by the name of Artbreeder. The artworks the participants thought were made by humans scored significantly better on fields such as beauty and worth, while the participants seemed to dislike the AI-generated ones quite a bit.

I think it’s possible to confidently say that we as humans have a natural prejudice towards AI-generated art, which isn’t all that weird in the end. New technologies come with new fears; people see their jobs disappearing right before their eyes, since the new cool machine can now do what they have done for years, except a thousand times faster and sometimes even better.

On one hand, we find it amazing that new technologies can make things like art so much more accessible to people who have never learned to draw and paint before. It opens up a whole new type of creativity where people can think of the weirdest and most complex things and actually very accurately visualize this within seconds. On the other hand, we should ask ourselves if we can even begin to call this art. It is absolutely undeniable that the piece made by Allen is stunning, but isn’t there a huge difference between typing a few words into a text bar and taking hours to create a painting or a sculpture?

Artists have come forth to say that they feel like AI-generated art “is the exact opposite of what they believe art to be”, which is very understandable, also looking back at the response of the people from the study into our attitude towards AI-generated art. We as humans feel as if art isn’t real art without emotion or meaning, because it’s exactly that emotion and meaning that makes a painting or sculpture mean so much to us.

One of my personal favorite art pieces is Anguish by August Friedrich Schenk, since I feel like it is a beautiful portrayal of grief. We can share this emotion with the artist and relate to them, while AI has nothing in common with us: it’s a machine devoid of any emotion. Seeing your own emotion replicated in a painting knowing that the artist has felt the same way as you did once is a feeling that an AI will not be able to replicate.

So, was Allen right in saying that art is dead? It’s an impossible question to answer factually, but I personally don’t think it is. I think people value the connection made with the artist through art way too much to accept that AI has the potential to create everything we want and replace artists. Humans are quite conservative in that way: art is something that has always been used as a tool of communication between humans, so no technology should have to interfere with that. But will we eventually stop caring and still let AI take over? That’s something only the future can tell.

References

[1] https://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html
[2] https://www.youtube.com/watch?v=SVcsDDABEkM
[3] https://doi.org/10.1016/j.poetic.2023.101839
[4] https://cs.uwaterloo.ca/~jhoey/teaching/cogsci600/papers/Roose2022.pdf
[5] https://www.youtube.com/watch?v=SVcsDDABEkM&ab_channel=Vox
[6] https://sciencedirect.com/science/article/pii/S0304422X23000797