Text/Roxy Y. Tang
Shao Yuxuan (b. 2005) derives his artistic practice from observing the evolution of artificial intelligence in the information age. Having grown up in an environment dominated by social media and algorithms, he has personally witnessed the challenges to the authenticity of images and text. This has led him to move beyond questioning where images come from, and instead focus on the image itself as a form of “elusive phantom.” His distinctive approach involves actively embracing and utilizing the distortions, glitches, and noise generated by AI—specifically early, raw models like Stable Diffusion 1/2.1. For him, these are not technical flaws, but rather the most direct exposé of algorithmic logic and database bias.
His methodology is characterized by a distinctly research-oriented quality:
- The “Prompt Minimalism” Experiment: Contrary to mainstream practices that seek precise control over AI, Shao Yuxuan intentionally feeds AI highly abstract, nonsensical, or even random, concise text prompts (e.g., “the alternates,” “Nurse Justin Bieber,” or a single emoji). He refrains from optimizing these prompts, instead allowing the AI model to process them according to its default logic. This functions as a controlled experiment, designed to observe and record how the AI “drifts” when interpreting ambiguous instructions, thereby revealing the limitations of its internal logic and its embedded cultural biases (for instance, inputting an emoji heavily biases the AI toward generating images expressing “positive emotions,” reflecting the tendencies of its training data).
- Focus on “Failure” and the “Uncanny”: He pays particular attention to the distorted, bizarre, and even “failed” images produced by AI. These images (such as the multitude of indistinguishable “human images” generated by the prompt “the alternates”) visually demonstrate how the algorithm composites and mutates data. Shao Yuxuan posits that such “failure” precisely exposes the essence of AI—it does not comprehend meaning, it merely synthesizes data. For him, this “uncanny” state is closer to the authentic human experience of existence within the deluge of information.
- Painting as “Embodied” Translation: The core of Shao Yuxuan’s practice is the manual translation of these AI-generated “phantom images” into paintings. His painting process itself mimics certain characteristics of AI:
- He uses insufficiently mixed pigments, preserving a sense of the “unfinished.”
- He actively accepts and utilizes accidental deviations during the painting process (e.g., the accidental shape of a canvas inspiring new content).
- He directly appropriates the AI’s erroneous textual descriptions of images as painting titles, highlighting the power dynamics between language and image.
He re-contextualizes AI images by referencing art historical concepts (such as the Dutch “Tronie” expression study paintings) or pop culture elements (like the villagers from the game Minecraft). His canvases become physical evidence of the encounter between the human body and algorithmic logic, documenting noise, randomness, and instability.
Shao Yuxuan does not seek to use AI to create “perfect” images, nor does he indulge in constructing grand narratives. Like a methodological observer and experimenter, he establishes rules (minimalist prompts), observes outcomes (AI’s distorted outputs), and transforms the medium (painting), thereby clearly revealing the mechanisms of AI image generation, the biases embedded in databases, and the genuine position of humans in this process—both as users and as subjects shaped by the system. His work invites viewers to confront the inherent “instability” of image production in the information age and to reconsider the meaning of “authenticity” in the algorithmic era.
Yuting (Roxy) TANG 