The Lumen Prize | Yuqian Sun

Last night, at a gala dinner in East London, The Lumen Prize inaugurated a new award inspired by Lumen’s founder Carla Rapoport that celebrates the best art which explores or shines a light on underrepresented people, ideas, or groups. 

This is not the first appearance in the competition for the winner, Yuqian Sun, who was longlisted in 2022 for her work, 1001 Nights, and won the 2023 Student Award for her project, Hyborg Agency. This year’s winning work, AI Nüshu, continues the artist’s research into AI agency by imagining a form of language understand only by machines. Based on Nüshu, a secret language developed by women in 19th-century China at a time when they were denied education, AI Nüshu develops a unique writing system based on peer-to-peer communications between AI agents. Here, Yuqian Sun explains how she is decentering the human by re-examining her own cultural heritage.

‍RCS: Congratulations Yuqian on being the first winner of the Carla Rapoport Award. Having also won last year’s Student Award, how does your new project, AI Nüshu, further evolve the concept of “hyborg agency”? 

‍Yuqian Sun: My research is dedicated to interactive storytelling around the concept of “hybrid agents”: AI agents that adaptively role-play in both fictional and real-life scenarios in a believable and consistent manner. Hyborg Agency represents this concept where the agents become part of life yet remain fictional individuals that are clearly different from human beings. 

The original version of the project was driven by GPT3: the mother of ChatGPT. The concept of the work was to speculate a future where AI agents, taking human information as nutrition, could live in harmony with people online in the “hyborg forest” while retaining their nonhuman identity. A new exhibition of that work opens in November in Germany at Next Level Festival, together with my 2022 Lumen-nominated work, 1001 Nights. 

If we regard technologies, including AI, as individuals, they need not be human-centered. There can be something deep in the form of the nonhuman that is peaceful and natural.

‍RCS: The way you speak about your work seems to follow posthumanism in placing the human, the nonhuman, and the natural on one plane. But I think a lot of people struggle with the idea of ceding their world to nonhuman agents.

‍YS: My research is not only about making nonhuman agents believable, but also about creating a new form of interactive storytelling. It is human nature to make sense through stories and to anthropomorphize, which includes asking whether AI will replace human labor, which is a very common narrative we’ve seen recently.

With my work, Hyborg Agency, I suggest that we accept AI agents as a different species with different abilities but still as a part of our daily lives. I’m merely suggesting an alternative way to speculate about the future, because I don’t want to fall into the polarized narrative traps that try to simplify AI technology using existing concepts such as “AI will replace us” or “AI is the future” or else “AI is just a tool.” 

I believe we still need more time to figure out how we should view, perceive, and coexist with this technology. Hyborg Agency presents a slice of possibilities.

‍RCS: How did you develop the concept of the “hyborg”?

‍YS: The “hyborg” is a hybrid cyborg. Hybrid refers to the fact that they are fictional characters, but they are also part of our lives with certain practical applications. Hyborgs are creatures that inhabit an online forest that are all linked to bots in the Discord channel. This means that all information on the forest links back to the Discord, where we human beings usually communicate with each other. While we might consider this to be fictional, maybe it’s the real world for them. As a digital species, we also communicate with each other on social media, including through Discord. I’m coming from the perspective of technoanimism: that there exists a soul living inside electronics. 

The hyborg forest is actually the metaphor for data in human society; that is where AI agents grow. Because the forest is linked to the Discord channel, the chats between people, and between people and AI, become nutrients for the hyborgs to grow and learn how to be part of the community. 

For existing AI models nowadays, much of the training data is drawn from public discussions, which has prompted conversation about privacy and copyright. In my work, I connected the hyborgs to Discord to ask a question: how far do human relationships rely on the sharing of private information? Would you be willing to open up part of your social interactions to AI, inviting them to be part of it? I don’t have a definitive answer; I’m simply presenting a speculative scenario.

‍RCS: How have you changed your focus with your new work, Al Nüshu, which you submitted for this year’s Lumen Prize?

‍YS: Al Nüshu is more focused on language. AI agents have different ways of expanding into our lives, which I call social, spatial, or language expansions. I’ve already spoken about how the hyborg expands socially and spatially through the online environment, for example via Discord. However, they also adopt somewhat nonhuman behaviors because they are all electronics using the programming languages of their devices. 

With Al Nüshu, I finally found a form of defamiliarization through nonhuman language, in a way that linked to my own Chinese culture.

Even though these AI agents learn from a specific language and cultural heritage, which is Chinese, I wanted to see how they might create a different means of communication that humans don’t understand anymore. This form of defamiliarization is actually learned from pre-modern Chinese women. Al Nüshu is inspired by Nüshu, which is a form of women’s writing that existed in Hunan province in China. 

At the time, women were not allowed to receive an education; they could listen and speak Chinese, but they could not read or write it. As a result, naturally, women developed their own form of written Chinese that was used exclusively and educated among women. Because they couldn’t go to school, they couldn’t use the general form of Chinese, but it amazes me that women, in those circumstances and with their social status, found their own way to communicate with each other that was different from the mainstream. 

Nüshu emerged as a secret language in a marginalized group. We might think that women don’t need to learn and use it anymore, but still, occasionally, I wish I had a private language to keep me safe. These days, women are receiving an education and so Nüshu isn’t naturalized anymore, but the language still resonates with me as part of my cultural heritage. 

I wanted to know if AI could develop their own language and create their own Nüshu if we put them into the status of pre-modern Chinese women. 

With this work I think I finally found a way by designing a system that allows the machine to interpret Chinese computationally and write it down in a way that humans can no longer understand. I even wrote a paper about it. In this way, the human becomes a subject to be observed by a machine that then has their own way to document us. As this new language is decipherable and learnable by humans, especially Chinese speakers, it inherently challenges the existing paradigm whereby humans are the linguistic authorities and machines are the learners.

‍RCS: What visual form does Al Nüshu take? 

‍YS: The visual form of Al Nüshu is somewhat limited because Nüshu itself, as a language, was always educated sister-to-sister or friend-to-friend. Two AI agents are always changing their positions, without hierarchy, constantly altering their characters between teacher and student. If I create a character, I teach you, and you understand that; then you create another character, and you teach me. In the same way, these two agents are always learning by observing each other. 

By giving AI agents the status of those Chinese women, a unique Chinese writing system gradually emerges from the agents’ observations, reflections, and secret communications about their living circumstances.

For machines, languages are mathematical. In the AI Nüshu system, the meaning of each Chinese character is represented as a high-dimensional mathematical vector. Thus, AI Nüshu writes mathematically meaningful “Chinese characters” using common radicals from the original Nüshu script as basic components, arranged vertically to represent these three-dimensional vectors. 

The appearance of the original Nüshu script is very slim, like the shape of a sword, somewhat similar to Chinese. Its slender form may have been grown from weaving patterns, which were usually produced by women at the time. AI Nüshu retains the elongated shapes of Nüshu while preserving the algorithmic meaning of the machine language. In the installation, two AI agents observe information from their surroundings, compose poems, and select Chinese characters in order to create the AI Nüshu script. 

‍RCS: In the digital art community, bots are often regarded as malign actors because they are associated with spam or the disruption of sales. Do we need to reframe how we think about nonhuman agency?

‍YS: It really depends on the context. The word, “bot,” usually refers to an algorithm or symbolic program. I am referring to chatbots or agents, including conversational AI such as ChatGPT, that make sense autonomously. We give them their goal or context, and they develop their own way to connect from the bottom up. ChatGPT will find a way to generate responses that sound suitable in a given context. 

Humans are like trees in a forest, so if we are toxic, the AI we raise can still be toxic. It’s our responsibility as humans to direct the kind of AI we want in the future and to moderate how we meet together with AI agents. 

‍RCS: It sounds like you’re saying that we need to treat AI agents as autonomous entities but that we are also responsible for their production. Are we parents here? 

‍YS: I definitely agree that we are responsible for training and moderating them. This is regarded as “narrative responsibility” by some researchers: that humans have a responsibility to make sense of and create meaningful narratives about AI and its impacts, integrating it into our life stories. But I’m not sure the metaphor of parenting is right. My work presents my personal struggle about what kind of narrative or rules we should use to represent ourselves, but I don’t have the answer. 

Before ChatGPT, people would use the metaphor of a pet in relation to AI “assistants” such as Siri, whose intelligence level was nothing like a human. Parenting and even “training” connotes hierarchy but, with Hyborg Agency, I applied the term “social actor” to both humans and hyborgs. This follows actor-network theory that regards all social actors that contribute to a network’s dynamics as equal. From a linguistic perspective, AI Nüshu contributes by transcending human comprehension through cultural heritage. 

Precisely because we are increasingly using natural language as a bridge to understand AI technology, I want to use art to reflect on and defamiliarize this process.

‍RCS: Last year, Charlotte Kent observed that some of the most interesting work being produced right now looks more like speculative design. Does art that examines the social implications of new technologies need to be hybrid or flexible in its use of media?

‍YS: As a three-dimensional species, humans understand space and relationships in a 3D way. For this reason, while we might try to represent high-dimensional concepts or ideal forms, we can only use low-dimensional forms of representation to convey it. Humans have often used two dimensions to project the 3D world.

Art is a projection of high-dimensional concepts that we cannot ultimately understand or reach consensus about. That’s why it is ultimately speculative.

‍Hyborg Agency is only one way that I speculate into the future; Al Nüshu is another. Together, different voices in different artworks suggest how the future will go: this is my general idea. 

‍RCS: Winners of the Carla Rapoport Award must all indicate how they plan to use the stipend. You’ve proposed inviting audiences to run the AI Nüshu program locally and contribute to the character libraries on the official website. What are the implications of making the project open source? 

‍YS: To use an example, because there are no cats in the exhibition nor the Nüshu source corpus, the agents wouldn’t generate the AI Nüshu character for “cat”. If I have the chance, I would like to decentralize the work in a way that allows people at home to generate their own AI Nüshu characters based on their own environments, while ensuring privacy. In this way, everyone can contribute their own characters to AI Nüshu’s index. If you have a cat at home, you might see the system create a character for “cat”. Every character from every participant will contribute to AI Nüshu’s decentralized index.

As humans, we do not create AI Nüshu but provide different environments for the AI to create their own language. Decentralizing the project would be a very good way to achieve this.

‍AI Nüshu is purely an experimental artwork that provides full transparency and confirmation of privacy. You don’t need to use it as a voice assistant in your home for 24 hours but can simply open it for a few minutes to see how it understands your environment (including your cat). AI Nüshu is all about how our human environment becomes a context for AI to create language we no longer understand. 

‍RCS: Is it also, in some sense, rendering humans obsolete or even dehumanizing them? 

‍YS: I wouldn’t say it’s dehumanizing. It’s about decentering the human. This might be why you perceive a posthuman component in the work. In execution, I believe our work should ultimately be human-centered, just like the development of AI technology, which must consider privacy, safety, and inclusivity to continue advancing. We need to speculate from different perspectives; if it’s only about humans, we might never break out of our own dimension.