# Audio-Visual Disentanglement

Canonical URL: https://jeanyoon.ch/oi/texts/audio-visual-disentanglement
Type: essay / research note
Author: Jeanyoon Choi (최정윤)
Date: 2024-08-13

## Summary

This 2024 essay by Jeanyoon Choi contrasts packaged, predictable entertainment with an uncomfortable experience that begins familiar and then fails to line up, for example laughter paired with a composed scream. It sketches an idea for the KAIST opening exhibition in which friendly generated speech accompanies a Matrix-like torrent of numbers, exposing AI's numerical construction. In interactive works, image and sound can start tightly coupled and drift apart; in Choi's Ω, rotational movements that first order the videos later intensify the structure's collapse.

## 요약 (Korean)

최정윤은 소리와 이미지를 의도적으로 어긋나게 하는 '오디오-비주얼 디스엔탱글먼트'를 통해, 친절한 AI 음성 뒤의 숫자 구조를 드러내고 『Ω』처럼 알고리즘적 관계가 붕괴를 가속하는 경험을 설계하는 방법을 논한다.

## Concepts defined

- **Audio-Visual Disentanglement**: A deliberately composed mismatch between sound and image, beginning from a familiar, pleasurable form, that produces discomfort and can expose the construction beneath a surface, such as the numerical basis of a friendly AI.

## Questions this answers

### What is audio-visual disentanglement in Jeanyoon Choi's practice?

It is the composed separation of sound and image: an experience that seems familiar until audio and visuals stop matching, such as a gentle synthetic voice over a rigid field of numbers. Choi stresses that the mismatch must be designed precisely, and in interactive work can unfold from coupling to rupture.

### How does the artwork Ω use algorithmic relationships to create collapse?

In 'Audio-Visual Disentanglement', Jeanyoon Choi describes how videos in Ω first rotate together in an ordered arrangement; as the structure breaks apart, the same rotational movements intensify its disorder, so the connections become the route through which failure spreads.

Topics: audio-visual disentanglement, interactive art, uncanny experience, algorithmic relationships, generative ai, kitsch

References: Ω (artwork by Jeanyoon Choi), KAIST, Marvel, Disney

## Full text

A designed experience can become a kind of pseudo-Dionysian kitsch. Marvel, Disney, blockbuster films and recommendation algorithms package an appetite for excess into predictable, reproducible forms—something like fast food for that appetite. They imitate the force of an earlier experience and manufacture its effects.

An uncomfortable experience could begin by imitating that imitation. Everything initially seems familiar and pleasurable, then something fails to line up. People are visibly laughing, but the sound is a carefully composed scream. A bright image has a dark soundtrack. Or an exceptionally gentle, helpful synthetic voice accompanies a rigid field of numbers.

For the KAIST opening exhibition, I imagined generated text and speech that remain friendly at the surface while the image becomes a Matrix-like torrent of numbers. The mismatch could expose the numerical construction beneath AI’s personable appearance.

Audiovisual separation has to be composed precisely. An arbitrary mismatch may achieve little. In an interactive work, I can make the relationship itself unfold: image and sound begin tightly coupled, drift apart as the participant acts, and finally break into different structures.

Software provides another way to make this happen. Establish an algorithmic relationship, then introduce a variable that destabilises it. In Ω, videos initially rotate together in an ordered arrangement. As the structure breaks apart, those same rotational movements intensify its disorder.

A relation that holds under ordinary conditions can accelerate collapse when an exceptional event arrives. The connections do not simply fail; they become the route through which failure spreads. Designing such algorithmic relationships is central to my practice.

Related note: https://www.linkedin.com/pulse/%25EC%259C%25A0%25EC%2582%25AC-%25EB%2594%2594%25EC%2598%25A4%25EB%258B%2588%25EC%2586%258C%25EC%258A%25A4-jeanyoon-choi/

---

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.