At the end of May, I published an article titled “Local Art School Undergraduate Exhibition” on the WeChat public platform “Jasagala”. This post about the undergraduate final works of the art academy was entirely generated by machine learning. The student photos, names, work images, work titles, and work explanations were all automatically completed by ChatGPT and Midjourney under my guidance, without any manual modifications afterwards. Unexpectedly, it broke through a hundred thousand reads in a few hours and kept rising. I had also published similar articles before. However, those articles remained only within our existing reader circles and were not brought to the public for discussion. Coincidentally, Mr. Meng Yao of “Pictorial” invited me to write a column based on this event. I took this opportunity to share my humble opinions on issues such as artificial intelligence, academy education, originality, and style.
The process of creating the graphics and text for this post actually deviated slightly from my initial idea. At first, my idea was to directly use existing student works as image prompts, but later decided to use pure text prompts as the only source of generation. The result was surprisingly good. The part that took more time was the author’s photo (Figure 1). Striving for diversity while maintaining authenticity is not an easy task. Ultimately, it might be because faces are the most familiar and sensitive image information to humans.
图一 Figure 1
The painting and sculpture part was actually the easiest to generate (Figure 2). In most cases, I only needed to write a few keywords describing the general content of the work, plus the names of one or two well-known artists, such as Richter, Peter Doig, Liu Xiaodong, David Hockney, etc.
图二 Figure 2
For the work introduction part, I spent a lot of time preparing in advance based on using GPT4. I built a context that met my requirements through continuous dialogue. Essentially, I had to meticulously filter and set up in advance the depth and diversity of the explanation of the work. After that, I would use about three keywords to get a work explanation. For example, for traditional media, I tend to keep the work description fairly simple; however, for new media, I expect its explanation to be “somewhat obscure”; as for humanities colleges, it is “art historical context, needs to be extremely obscure and contain philosophers and academic terms, and be annotated in German.” (Table 1)
然后，我们将引入后法兰克福学派的“消费社会”(Gesellschaft des Konsums)观念，以理解阿伊努族的文化特征是如何在现代消费社会中保持其独特性。进一步的，我们将利用哈贝马斯的“系统与生活世界”(System und Lebenswelt)理念，探讨在现代社会系统压力下，阿伊努族的生活世界是如何应对和转变的。
|Traditional Media||New Media||School of Humanities|
|Work Introduction: “Northern Winter” is a reenactment of my deep affection between sisters and the indoor life of northern China in winter. Everything between my sister and me at home becomes the source of my painting, showing a deep humanitarian concern and love for traditional Chinese home life. In the picture, I try to capture our laughter and those warm indoor spaces: familiar wooden furniture, steaming heating, and ice flowers condensing on the windows. I record these details one by one, trying to convey the atmosphere and emotion of indoor life in the northern winter through them. These details may be small, but for me, they constitute a unique way of our life, a part of my identity as a Chinese, and also my understanding of the world.|
Work Introduction: In “Silence,” I use an empty room as the theme of the picture, trying to capture and express the sense of tranquility that is often overlooked in daily life. The empty room, without noise and clamor, only has quiet furniture and forgotten items, which tell silent stories. I try to reveal these seemingly silent but deeply meaningful spaces through the details of the work, thereby provoking reflection on those overlooked, forgotten corners in daily life. I hope that through viewing this work, people can become more aware of the subtle influences in daily life, and begin to cherish those quiet moments that are forgotten in the noise.
|Work Introduction: “Seasonal Games” is both a performance of performance art and an exploration of the relationship between human perception and environment. Under the hot summer sun, I choose to sell coldness in an attempt to break the conventional perception of seasons. Here, coldness is both a physical state and a symbol, indicating deep emotions and thoughts. Through such a confrontational approach, I try to decode our inherent cognition of seasonal perception and re-examine our interactive methods with the environment. This work is a deep exploration of perception, experience, and meaning, allowing viewers to reflect on how we understand the world and interpret life.|
Work Introduction: “Intersecting Fables” is not a typical visual chapter, but a subtle depiction of existence, trying to find a balance between the explicit and the obscure. Borrowing from the four elements – water, fire, earth, wind, I place myself in a theater full of possibilities, leading the viewer to a field intertwined with elements. There, water and fire coexist, wind and earth rely on each other, metaphorically representing the inner emotional whirlpool, and reflecting the external survival predicament. These elements swim in the depth of the stage, like multiple images of human nature, their existence is a projection of our inner depths. I hope that viewers can catch those fragments of elements that briefly shine under the theater lights while watching. They might lead us to a deeper understanding of the essence of existence and consciousness.
|Title: “Tracing Ainu Ideology and Reflections under the Perspective of the Frankfurt School: A Deep Deconstruction of Heterogeneous Cultural Collisions”|
This paper will base its analysis on the critical theory of the Frankfurt School, providing an all-encompassing and in-depth examination of the unique culture and ideology of the Ainu people. Through this ideological collision, it attempts to reveal the possible hidden connections between them and subsequently reposition both within the cultural patterns of modern society.
Firstly, this paper will start from the socio-cultural phenomena of the Ainu people, attempting to understand how their ideology is formed and transmitted through unique cultural symbols. For this, we will employ the Frankfurt School’s “Theory of Communicative Action” (Kommunikatives Handeln) for deconstruction, aiming to reveal the social forces hidden within this unique culture.
Then, we will introduce the concept of the “Consumer Society” (Gesellschaft des Konsums) from the Frankfurt School, to understand how the cultural characteristics of the Ainu people maintain their uniqueness within the modern consumer society. Further, we will use Habermas’s concept of “System and Lifeworld” (System und Lebenswelt) to explore how the lifeworld of the Ainu people responds and transforms under the pressure of modern societal systems.
Finally, the paper will attempt to further interpret and reflect on the cultural phenomena of the Ainu people from the perspective of critical theory. By doing so, we hope to understand anew the value and meaning of heterogeneous cultures through the lens of “Multiculturalism” (Multikulturalismus), and provide fresh insights into multicultural exchanges and collisions within modern society.
However, this WeChat article, or this work, its starting point is actually not entirely about artificial intelligence, but it’s rooted in the art education system. Several years ago, I was already attempting something like this. However, due to the difficulties of implementation and time constraints, it was always just a thought that couldn’t be realized. I feel a sense of fatigue from art school education, as if I’ve been through it millions of times. This fatigue is not just about the content or form of student creation, but more about the education system itself. I studied at the Central Academy of Fine Arts Middle School, then went to the School of Plastic Arts of the Central Academy of Fine Arts, and later went to UCLA for my master’s degree. I have now been a lecturer in the United States for several years. I have been in the art school education system for more than a decade. I remember when I was studying at the Central Academy of Fine Arts, my classmates and I could almost predict the appearance, style, and themes of works from each department and even each studio. If you went to the exhibition site, you could recognize which studio the work came from almost 20 meters away. In the process of learning about machine learning, I also gained new perspectives to think about and rethink what is new and what is education.
In machine learning, there is a concept called “overfitting.” In most cases, generative machine learning is about finding a function that can generate new data that is close to the original data based on an existing dataset. However, if the input samples are too few or lack diversity, when a model is too complex, it may overadapt to the random errors or noise in the training data, rather than the underlying relationships. Overfitting models perform well on training data and have high accuracy, but their predictive performance often decreases when applied to new, unseen data. The reason is that the model has “remembered” the specific noise and outliers in the training data, rather than learning the real trends and patterns behind the data.
Fundamentally, the reason why these machine-generated graduation works are so close to real works in style and performance is because, in art education, and even in the broader human learning process, we easily fall into a predicament similar to overfitting in machine learning. If we consider the creative trajectory of our predecessors as a training dataset, we usually start by imitating their style. Each style may mean a complex combination of various creative strategies. Critic and art historian Arthur Danto proposed a concept called the style matrix in The Art World in 1964. Simply put, there is a relationship of affirmation or denial between styles within the art world. From a simplified perspective, if we only consider the two dimensions of “representation” and “expression,” then we generally have four styles: Fauvism (representation, expression), Neoclassicism (representation, non-expression), Abstract Expressionism (non-representation, expression), Hard-edge Abstraction (non-representation, non-expression). Using this line of thinking, we can decompose the creative style of a certain school or artist into keywords. For example, we can roughly deconstruct Liu Xiaodong’s creative style:
刘小东 ≈ 现实主义题材+写实主义画面表现+中等尺寸笔触+摄影视角+非制作性+油画材料性+中高饱和鲜灰色调+低差异性饱和度+高色相差异性+中高明度……
Liu Xiaodong ≈ Realistic subject + Realistic pictorial representation + Medium-sized brushwork + Photographic perspective + Non-production + Oil painting material + Medium to high saturation gray tone + Low saturation difference + High hue difference + Medium to high brightness …
Returning to the context of machine learning, these polarized, fragmented styles become the textual cues used to create images. In other words, writing cues in this context is essentially the entire content of artistic creation. Moreover, long before the advent of artificial intelligence, many people were creating art by inputting cues and tuning parameters. For example, a painter who had someone else paint for him after being diagnosed with Parkinson’s, a director who doesn’t directly participate in photography or script creation, and a music director who can’t personally handle the bow.
However, the complex array of choices about various creative methods in this context is more of an artist’s subjective judgment, rather than a golden rule of creation. But when many students start learning, they enter a mode of passive learning that accepts the overall style wholesale. On the one hand, it’s because they can’t discern in terms of creative experience, but more importantly, it is a survival strategy subtly inculcated. Qi Baishi once said, “Those who learn from me will live, those who imitate me will die.” However, in academic education, painting like one’s teacher is not something to be criticized. I’ve personally seen old professors from art academies visiting the group exhibition of young teachers, and then criticizing almost everyone’s work. The only work that was somewhat affirmed was commented as “painted quite like Xu Mangyao.” Many times, what academia fears is not being criticized for painting too much like their teacher, but rather, the fear is that in the process of their work entering the market or the system, they may lose the competitive advantage they could have enjoyed because their work doesn’t resemble their teacher’s. This is an unspoken understanding, and even a collective scheme.
The final presentation of this work in the form of WeChat push notifications is a response to this situation. Unfortunately, most readers fall into the blind confidence or inferiority in specific technologies. They fall into judgments about finding differences in truth, such as “Oh my, can’t you all see that this is obviously AI?” and “Oh dear, I was really fooled, I didn’t expect it to be all AI-generated.” For me, AI is not an entity that is opposed to humans. AI should serve as a mirror, allowing us a better angle to understand what it means to be human.
Another comment is the disdain for the coldness of AI-generated images. Many people firmly believe that machines and algorithms will never have emotions, so the works they generate will definitely not contain emotions. However, there’s a hard-to-ignore issue here, which is that we naturally set “us (humans)” against “them (artificial intelligence).” But from my point of view, I would rather regard “me” and “everything else but me” as two distinct subjects. After all, how do we know for sure that other humans really have emotions like us and are not just simulating them?
Among many comments, there’s one that I feel deeply about. “I naturally ignored all doubts and flaws, and empathized with the AI’s work. I was looking for metaphors, only seeing what I wanted to see.” Yes, rather than obsessing over this, it’s better to “hear it and turn it into sound, meet it with the eye and turn it into color.” The information generated by machine learning is like an endless sea in a parallel time and space. We’ve already had so many discussions about “what is art” in the long river of history, such as: talkies can’t be called movies, photography has to be film, anything not rhymed can’t be called poetry… But in most cases, the essentialists will be beaten by time beyond recognition.
The painter Zhang Zhongge once said something that I’ve remembered for many years, “Painting may be partially considered art.” Besides the so-called art part, painting could also be technology, magic, luxury, an investment tool, a promotion channel, physical labor, time, passive inhalation of toxic paints and mediums. The non-artistic part in painting is just like the post-human elements in AI art. But this doesn’t mean that the final outcome has to be cold, boring, and insignificant. Perhaps we can set aside frames and backgrounds, pause concepts and logic, and return to the simplest and most direct viewing of the work. Art may be partially human.
Another point that concerns me is that among hundreds of comments, not one was about the humanities. Is it because everyone is unfamiliar with the field of art history, or because text naturally doesn’t attract as much attention as images? Or could it be that knowledge itself is actually an illusion?
Finally, if I tell you that this article is actually generated by ChatGPT based on a prompt of about 150 words written by me, would you feel angry, thinking that you wasted several minutes reading for nothing? But what if I tell you it was just a joke, and I still spent a lot of effort to write it, word by word? For this article, for you yourself, what is the actual difference?