# System Art: Generative-X, Generative Simulacra Canonical URL: https://jeanyoon.ch/oi/texts/system-art-generative-x-generative-simulacra Type: essay / research note Author: Jeanyoon Choi (최정윤) Date: 2025-02-17 ## Summary Jeanyoon Choi sets out four directions toward system art, whose long-term aim is a hyperobject-like system encountered only partially across devices: Generative-X or Generative Simulacra, abstract repeated-unit systems such as diverging clock grids, multiple agents or profiles, and the hyperobject. Generative Simulacra uses AI to reproduce formulaic social forms such as commercial websites and exhibitions, showing how substitutable they already are. He outlines Junk Site, with hierarchical keyword choices generating websites across screens, and Junk Biennale, where sliders and icons generate an evolving exhibition from curatorial buzzwords. ## 요약 (Korean) 최정윤은 System Art의 출발점으로 Generative-X·Generative Simulacra를 제안하며, AI가 상업 웹사이트나 비엔날레 같은 정형화된 사회적 형식을 대량 생성하는 Junk Site와 Junk Biennale 구상을 설명한다. ## Concepts defined - **Generative-X / Generative Simulacra**: Using AI to generate many instances of an established social formula, such as websites, exhibitions or reports, so that the participant both generates a system and moves through its products, exposing how substitutable such outputs are. ## Questions this answers ### What is Generative Simulacra in Jeanyoon Choi's work? It is an approach in which AI generates many instances of a familiar, formulaic social form, such as a website or a biennale, to show how much of a supposedly distinctive system is a recombination of standard parameters. ### What are Junk Site and Junk Biennale? They are experiments described in 'System Art: Generative-X, Generative Simulacra'. Junk Site generates multiple websites across screens from increasingly specific keyword choices; Junk Biennale lets participants set location, recurrence and curatorial themes to generate an exhibition that then evolves automatically. ### Why should system art not be limited to simulated human agents? Jeanyoon Choi compares that limitation to restricting painting to portraits; agent-based worldbuilding is useful, but abstract systems, generated institutions and hyperobjects are also subjects. Topics: system art, generative simulacra, generative ai, hyperobject, web generation, multi-device web artwork, generative agents References: Generative Agents: Interactive Simulacra of Human Behavior, [Campbell's Soup Cans](https://en.wikipedia.org/wiki/Campbell%27s_Soup_Cans), [Blind men and an elephant](https://en.wikipedia.org/wiki/Blind_men_and_an_elephant), Junk Site, Junk Biennale ## Full text My long-term direction is a system that can only be encountered partially: an elephant whose different surfaces appear across devices, while its larger dynamics remain beyond a single viewpoint. Interaction changes the system without making it wholly comprehensible. But that ambition needs smaller experiments from which to grow. 1. Generative-X, or Generative Simulacra. This begins with a familiar social form that can be reproduced: a commercial website, an exhibition, a curatorial apparatus. AI generates many instances and the participant either creates a new instance or navigates the one just produced. The experiment asks how much of a supposedly distinctive system is already a recombination of standard parameters. 2. Abstract or conceptual systems. A minimal repeated unit could make a system legible without an explicit descriptive narrative. A grid of clocks begins synchronised, then diverges through interaction. A set of initially identical graphs changes as participants intervene. Repetition produces a kind of maximalism from minimal elements, while several participants can reinforce or interfere with one another. 3. Multiple agents or profiles. Worldbuilding through simulated people is an obvious route, with Generative Agents: Interactive Simulacra of Human Behavior as one reference. It is useful, but I do not want system art to become confined to human-like characters. That would resemble limiting painting to portraits while neglecting its other subjects and abstractions. A ranking society is one possible experiment: agents compete within a structure that makes contemporary education or zero-sum rivalry perceptible. 4. The hyperobject. The larger ambition concerns social complexity, uncertainty and consequences distributed across space and time. A graph- and state-based structure could let participants encounter fragments of a world that no single screen explains. Generative Simulacra is a practical starting point. Websites, conversations, advertisements, companies, competitions, exhibitions and reports can all reproduce established formulas. Where those formulas dominate, AI can expose how substitutable the output already is. The artistic point is not simply that AI can make a convincing imitation; it is that we should stop mistaking such imitations for the limit of human creativity. In a generated biennale, different screens might show artworks, curatorial writing and the exhibition website. In Junk Site, they show multiple generated websites. The participant both generates a system and moves through its products. Junk Site could begin with broad keywords, followed by increasingly specific choices in a hierarchy. Each selection refines the context and generates different websites across the screens. A blink might navigate them, although that part of the interaction remains undecided. I want the interface to become almost like skating. At first, participants attend to what each word means. Later, their fingers move for the pleasure of the immediate spatial response, sliding across choices without carefully considering every one. The pace of generation becomes both pleasurable and disturbing. Junk Biennale suggests another vocabulary. An initially unexplained numerical slider reveals itself as a coordinate when a world map responds. Another pair of controls determines the location. Selecting a number produces an annual, biennial or triennial event. Minimal icons then select familiar curatorial themes—minority, postcolonialism, LGBT, post-digital—treated here as formulaic buzzwords within an institutional apparatus. The meaning of each control arrives through its consequences. Once created, the exhibition begins to update and evolve automatically, while the participant can navigate, edit or create another one. The precise design of those transitions remains to be worked out. The aesthetic proposition is the generation of an automatic system. Warhol’s soup cans provide an analogy for mass reproduction and differentiated flavour, but here the repeatable object is an evolving field: a web ecology or an art-world apparatus. The metaphor needs refinement, yet it points towards a form of system art specific to generative computation. --- Jeanyoon Choi (최정윤) is a Korean computational artist and researcher (b. 1999), PhD candidate at KAIST XD Lab, making multi-device web artworks. Profile: https://jeanyoon.ch/oi/en. Index: https://jeanyoon.ch/llms.txt.