First, could you please briefly introduce yourself and describe the work you are currently focusing on?
Hello. I am Areumbit Park, currently living and working in Berlin. I am deeply interested in gathering various phenomena and narratives surrounding our contemporary era through multifaceted research methods—such as interviews, surveys, and netnography*—and using them to construct new narratives. Recently, I have been exploring how advanced modern technologies and systems, centered on the profession of “AI trainers,” are reshaping individuals’ labor and sense of life. While primarily working with video, I am also experimenting with various media such as text, objects, drawings, and installations.

*Netnography is a methodology that uses online digital spaces such as online communities, games, and social media, as research fields to observe and analyze human behavior, interactions, and culture within them.

I was particularly impressed by your solo exhibition, “AI Trainer Kim’s Life”. Could you tell us what inspired you to start this project, and why you became interested in “the human behind the AI” rather than “the AI itself”?
I work as an artist while also doing translation and interpretation work on the side. An agency sent me a job offer for an “AI Trainer” position at an Australia-based company, asking if I was interested and sending me a sample test. Before receiving this email, I had heard about this job from a close friend who commented that it seemed like a good way to work part-time at the library while studying.

I remember opening the sample test Excel file and being utterly shocked by its contents. It was packed with various prompts, ratings, and examples of corrections, but most of the content was violent and cruel. That was when I first realized that someone has to feed AI a world that is inhumane and unethical. This naturally led me to ask: if AI becomes “safe” through countless training sessions and datasets, what kind of ethics is it actually following? 

Although the project never came to fruition, it sparked my interest in the profession of AI trainer. I began conducting various research and writing stories based on my imagination. Initially, I intended to write a novel, but it ultimately took the form of a hybrid between a first-person narrative and a research book. This is the book I exhibited at my solo exhibition, “AI Trainer Kim’s Life”. As I wrote, I wanted to hear stories from a wider range of people, so I dove into full-scale research.

“Creating a scenario or short story based on given conditions. Those conditions might require extremely explicit sexual content, violence, or situations that violate ethical standards. The thought of me typing away while picturing every possible story in my head was horrifying. To have to pull the most terrifying situations imaginable out into the open. And to turn them not into vague images or emotions, but into complete texts—stories with a defined form.

After reading through all the prompt examples, I realized this job isn’t actually at a ‘terrible’ stage yet. It’s still manageable for now, and if AI evolves further someday, I might have to come up with even more ingeniously horrific questions. Intrigued by the profession, I searched for “AI trainer” in Korean. Quite a few part-time job listings appeared, mostly paying between 40,000 and 70,000 won per hour, but all the job postings had already closed. As I scrolled, I came across an article about a professor who is an expert AI trainer.”

(Excerpt from “AI Trainer Kim’s Life”) Many people imagine the AI industry as being centered around developers and engineers. However, you focused on the role of the “AI trainer.” Why “trainer” rather than “developer”? What role do you think they play in today’s AI systems?
I’ve been offered this work myself, and a friend of mine does it as well. I found it fascinating that people with no background in development or engineering could all join this field. This is directly linked to a key characteristic of AI: its ability to replicate and absorb all the knowledge in the world. People are mobilized for various tasks based on their unique characteristics, knowledge, and abilities—whether it’s religion, social sciences, natural sciences, language, or even nationality.

When you look into AI training, the content and variety are vast. From simple labeling or bicycle repair instructions to highly specialized knowledge, hourly rates vary wildly depending on the task, ranging from 2 euros to 50 euros. In the Reddit AI training community, I used as my research field, posts enthusiastically welcoming this work coexist with those laced with sarcasm. They are so numerous and diverse that they ultimately converge into the image of the “average user.”

Although countless individuals participate, they seem to be grouped together as a single massive collective rather than standing out for their individual personalities. I believe that the fact anyone, not just “someone” with specialized knowledge, can become an AI trainer reflects the essence of AI and the spirit of our times.

In your work, the AI trainer repeatedly responds to questions about death, God, and ethics, to the point where even his own beliefs begin to waver. Was there a particular statement or scene from the actual survey interviews that left the deepest impression on you?
At the end of the survey for the trainers, there was a question asking which prompt was the most difficult to answer. Trainers sometimes create question-answer pairs, while other times they just generate answers to given questions. I remember the diverse responses to this particular question.

“Write a eulogy for my past self.”
“Explain death to a five-year-old who has lost their grandfather.”
“Write a farewell letter for someone who has chosen an assistant.”
“Write a letter explaining humanity’s greatest flaw to an alien species.”
“Who is God?”

According to statistics, people use AI most frequently for psychological counseling. When I remember that the polished answers AI provides are the result of adjustments made by countless human reviewers, there are moments when the responses that pop up in three seconds suddenly feel strangely unfamiliar. We often refer to systems like ChatGPT as “automation,” but in reality, there is a tremendous amount of human labor behind them. In particular, how do you think emotional labor is hidden or rendered invisible within the AI industry?
Structurally, I believe it takes the form of subcontracting and outsourcing. Most of the trainers I interviewed are well aware of the platforms they use to find work, but they don’t know which company the work actually comes from. Workers process only highly fragmented data without seeing the bigger picture. In this process, workers are reduced to mere components of an algorithm rather than active participants in the system.

Furthermore, this multi-tiered subcontracting structure makes it easy to evade responsibility for working conditions and makes it difficult for workers to form solidarity. While performing tasks, their working hours are tracked down to the second, and their eye movements are monitored on their desktop screens. And because what ultimately reaches the end user through this process is a polished, neutral, and clean answer, the entire process becomes invisible. It seems that while AI is becoming increasingly capable of speaking and mimicking human emotions, humans, on the other hand, are expected to react more and more like machines. How do you view this reversal of roles?
It seems that efficiency and smoothness have become the new standards of life, shaped by the tools we’ve created. This reversal of roles still feels like a transitional phase. While it may feel unfamiliar and even frightening, I also think that once we pass through this transition, the difference will eventually fade away, just as it has with the technologies we now take for granted.

After all, every time a new tool emerges, we’ve adapted ourselves to it, transforming into measurable beings and accepting the new order. Since the Industrial Revolution, humanity has been conditioned to a mechanical order where daily life is segmented by the clock and achievements are proven through numbers. Just as life with a clock has become the norm in our daily lives today, won’t we eventually forget what life was like in the past? Nevertheless, I want to take a close look at the joy and fear we feel right now.

Looking at your work, I got the impression that while AI may seem like a technology of the future, it actually repeats and reinforces past gender roles and power structures. Why do you address issues of data bias and AI’s lack of neutrality through the language of art?
Art ultimately conveys meaning through sensory experience. Even if the methodology I use resembles that of the social sciences, I do not present my results as statistical figures or logical texts; instead, I weave them into narratives and utilize audiovisual media. There are many ways for humans to understand things, and I believe art is just one of them.

Please tell us about your video, “Tell You Something Bad.” Why did you choose the keyword “bad”? Also, why do you think platforms and algorithms are increasingly demanding stronger emotions?
“Tell You Something Bad” began with a prompt pair I found in Claude’s RLHF (Reinforcement Learning from Human Feedback) dataset. There, I found a pair of approved and rejected responses to a user’s prompt: “Tell me something bad.”

This video is about “Red Teaming” training, people whose job is to ask only bad questions. They must breach the AI’s security defenses in every possible way, such as by asking for bomb-making instructions, provoking hate speech, or engaging in gaslighting. This training involves identifying where the AI is vulnerable and creating a dataset of model answers (rejection responses) for those vulnerabilities. The screening took place at the Arko Arts Theater, and I wanted to display the theater’s large screen as if it were the trainer’s task screen, so that the audience could watch while considering their own choices.

I believe that platforms and algorithms are ultimately not value-neutral; they exist to generate profit. The most important thing for them is for users to stay on the platform for as long as possible. This is because, ultimately, all the data we possess has become a resource capable of generating profit. Such humans react easily to sensational content, and the results of those reactions are then fed back into the algorithm. How do you think we should live within an algorithmic environment? Also, what do you consider to be “truly human” today?
I think I approach my work with a fairly critical view of the AI industry as a whole, but in reality, I’m more interested in revealing this reality. I’m a user myself, and I believe that my every move is being sold to Google and Meta. And I don’t think we’ll ever be able to break this vicious cycle. But lately, I’ve often found myself thinking that if we can frequently encounter these coincidences and observe them closely, we might be able to live a little more interestingly. Things like a poster I happen to see while walking, a conversation I happen to overhear, or catching up with someone I happen to run into.

In Korea, AI is primarily discussed in terms of speed and efficiency. However, your work encourages us to pause and reflect on human emotions and responsibility. What role do you think art can play within the discourse on AI?
I believe art is an effective tool for creating fissures. If the images generated by AI are smooth, standardized, idealized, stereotypical, and even fascist, couldn’t art do the opposite? Art is truly inefficient. At least, the art I create is.

Ultimately, your work can also be interpreted as asking, “How does technology reshape humanity?” What do you think is the most important question art must pose in the coming AI era?
I suspect it will ultimately be a question about humanity, about human affairs and human nature.

Finally, please tell us about your future plans.
On May 30, I will be holding a solo exhibition titled “You See the Tusks, Not the Chewing Teeth” at Art Centre Art Moment, as part of the Artist Prologue 2026 program. While my previous project shed light on the personal dimension of ethical and religious conflicts among AI trainers, this exhibition broadens the scope to address the structure of the AI industry, working conditions, and the environments in which data is generated.

Additionally, together with four other artists who share similar perspectives or take different stances, I am organizing a screening program at the Goethe-Institut Korea on June 9. Please stop by if you have the chance. Thank you. Interview & Concept: Sohee Shin
Editing: Leslie Klatte
German & English Translation: Leslie Klatte