Jen Liu's The Land at the Bottom of the Sea, Part 3, 2023. Everyone lives in the ocean now. The ocean is a vast warehouse of genetic resources for our emergent technoscientific industries.

Excerpt from five-part work. Single channel 4K video with stereo sound, 27:43. Generative algorithm using machine learning (GPT-2, Diffusion Model), Fine-tuning (NVIDIA T4 GPUs), Contour generation (OpenCV) and Steganography (Python, Linux).

Jen Liu's The Land at the Bottom of the Sea, Part 4, 2023. Welcome! Let us take your supply chain to the next level.

Artist Statement

What happens if you are liquidated – financially, socially, politically, subject to climate-induced flooding? Under what conditions is it possible to survive?

The Land at the Bottom of the Sea is a live action and 3D animation video, as well as related works (painting, online media, biomaterial and sculptural objects) that consider this issue, as the last chapter of Pink Slime Caesar Shift, a multi-year body of work that proposed to (mis)use developments in food biotech to create alternative networks for labor activism in and around South China. This last chapter of the body of work considers several failures: the failure of Chinese labor NGO’s to survive in the face of overwhelming governmental force, the failure of techno-optimism to provide solutions to concrete environmental and social problems, and the failure of a body of art to accurately assess future possibilities. They all must come to an end, they’ve all drowned in the deeps.

I worked with Soul Choi to build multiple NLP models based on selected bodies of text – from beauty marketing to social media of Guangdong factory workers, the models were successful but also built to fail: none can speak to a political fact that has been wiped from social memory: what happens on the day after you are politically/socially liquidated? And after that? What is the political afterlife of a dead movement? From there, Soul also researched and designed a steganographic encoder, such that thousands of images of disappeared activists I culled from the internet, can be preserved into perpetuity, no matter what is erased, censored, embedded with broken links and obsolete technology in the coming years.

Collaborator Statement

In collaboration with Jen’s work, as an AI technologist, I was uniquely positioned to share the narratives of labor workers, activists, and feminists who faced unethical working environments, resulting in detrimental physical and mental health consequences, or who fearlessly voiced their concerns, only to be silenced by political or social liquidation. When Jen first handed me her years of research, I was shocked by my own ignorance and the vast scope of the project.

In retrospect, the summer of 2022, when we started our Backslash project, was quite peculiar. Everyone was "generating" a profile picture or something with AI, with the rise of Diffusion Models and tools like midjourney and Dall-E 2. As we grapple with the question of what AI’s generations mean for human creative works, their responsibilities, authenticity, worth, etc, we started researching text and video generation.

Video Generation with AI

Our exploration began with Diffusion Models as well, aiming to generate video segments that AI would later distort and expand. I remember explaining the process to Jen: how the model starts with a page of Gaussian noise and gradually refines the image based on the prompt, layer by layer, until a clear picture emerges during the generation process. Jen pointed out the parallels between this process and how we are trying to build something from the "liquidated" things. For me, the murky underwater footage of flooded areas visually echoed the initial noise state where these models begin. So murky, where do we go from there? Although these initial experiments didn't make it to the final work due to limitations in control and cultural inclusivity, they lead us to deeper exploration of AI's potential in storytelling. I used Latent Diffusion Model to generate each frame of the video as an image and applied FILM: Frame Interpolation for Large Motion (research from Washington University) to interpolate between those generated frames for smoother transitions. Works archived in this github repo.

Text Generation with AI

Concurrently, we kept ideating how to deliver their stories with AI. What about text? Important to note that this was the pre-ChatGPT era. I fine-tuned models like Recurrent Neural Networks (RNNs), particularly Long Short-Term Memory (LSTM), BERT, and GPT-2 on diverse texts, from social media posts to Chinese literature and corporate announcements. LSTMs struggled with the complex relationships in the diverse corpus, resulting in inconsistent texts. BERT's bidirectional context awareness improved consistency, but coherence remained elusive. GPT-2 (it was open-sourced and fine-tunable on our machines unlike its successor GPT-3.5 and forth) was the most promising. Pre-trained models indeed performed better in natural generation, indicating that the more it had "studied" the "unliquidated" content online, the better it got. I’m still pondering upon AI's reliance on existing, mainstream narratives and its potential to perpetuate the silencing of marginalized voices. The architecture of these archaic text generation methods used in our process, illustrated below ("An archaic structure...").

Can AIs tell a story that we dare to tell?

Throughout the project, I was driven by the desire to explore AI's potential in telling the stories we dared not tell after the "liquidations". And I had to confront the harsh reality that the stories we wanted to tell were already erased from the AI's training data. Out-of-the box generative models were more proof of the liquidation than a protector. This realization hit hard when Diffusion models consistently generated blond, double-eyelid features with red Chinese patterns in the background, instead of accurately representing East Asian womens’ features. The same limitations emerged when we tried to incorporate writing styles from Asian literature, highlighting the glaring absence of these voices in the AI's knowledge base. (In the grand scheme of things, I am part of a history that faces the risk of liquidation.)

Steganography

To protect these stories, we are archiving the raw content—videos, texts, and pictures—collected throughout the *Pink Slime Caesar Shift* via steganography. Each pixel of the video contains information about the real stories, which can be decoded and reconstructed by the viewer following these directions. Unlike cryptography, steganography conceals the very fact that something is hidden. Viewers must acknowledge the liquidated stories and access them with a specific algorithm. Through this collaboration with the audience, we attempt to protect the liquidated stories that AI currently cannot tell properly at this time.