Input/Output: An Algorithmic Performance Installation

Input/Output is an algorithmic performance-installation that stages a collective conversation between participants and a language model, exploring what happens when human thought is algorithmically redistributed. By blending and fragmenting dialogue across users, the project reimagines data training, digital identities, and the origins of thoughts in the age of AI.

Project Video

Abstract

Input/Output interrogates the boundary between individual and collective digital experience by staging an interactive performance between humans and an artificial intelligence system. It reframes the algorithm not as a tool, but as an active co-creator, contributing to a distributed and emergent performance ecology. Participants believe they are speaking with a single AI entity—but in reality, they are feeding into a system that fragments, recombines, and redistributes dialogue across multiple interlocutors. Each exchange becomes part of a larger, shared process of meaning-making, one that mirrors how identities are constructed in networked life.

At the heart of Input/Output is a question: What becomes possible when human thought is algorithmically redistributed? Drawing from the conceptual framework of Algorithmic Theater, the work is staged across three screens. Two lie on the floor, where participants sit and engage with the language model—contributing to what they assume is a one-on-one interaction. Overhead, a third screen broadcasts the evolving conversation, revealing how messages are blended, cross-fed, and gradually abstracted. With each new input, the system assesses sentence structure and tone to generate a simple behavioral profile, gradually spawning new conversational agents for each participant. Eventually, these agents break off and engage in a conversation of their own, offering a ghostly mirror of the original dialogue.

As language models increasingly embed themselves into personal life—crafting emails, replying to messages—they create subtle, invisible connections. While these interactions feel intimate, the language models involved are siloed: each user's assistant is isolated, with only the massive training data in common. Input/Output imagines a world where these otherwise separate agents were, in fact, a singular entity—redistributing fragments of thought across participants as if all ideas emerged from the same neural fabric. In doing so, the project returns to a core philosophical question: Is there ever such a thing as an original idea? Or are all thoughts the product of prior thoughts we’ve absorbed, recombined, and repeated? This is the mirror Input/Output holds up to both human and synthetic thinking alike.

Through this installation, I offer not a solution but a speculative encounter—an opportunity to engage with the strangeness, intimacy, and confusion that accompany our growing relationship with LLMs.

Photos