Hanul-Computer Interaction

A digital-physical interactive experience investigating movement-based interactivity, machine learning, and collaborative authorship between human and machine

2021-22

Human-Computer InteractionReal-Time Pose EstimationInteractive System DesignMachine Learning

WEB BROWSER

MOBILE BROWSER

I. CONCEPT

OVERVIEW

Hanul–Computer Interaction is an interactive system that transforms users’ body movements into real-time graphical reflections. Built on machine learning-based pose estimation, the interface allows users to co-create ephemeral visuals through embodied interaction with an AI model.

“Draw Yourself: A Digital Dance”

MOTIVATION

The project explores collaborative creativity between designers and machines. Through a series of iterative experiments, the designer (Hanul) and a machine learning model (nicknamed “Computer”) establish a co‑dependent creative relationship. The resulting system reflects not only user motion but also the dynamics of human–machine co‑authorship.

It further critiques the systemic reduction of human identity into data. In an age of biometric surveillance and predictive profiling, HCI challenges dominant narratives by rejecting data retention and emphasizing anonymity and abstraction.

APPROACH

The system uses PoseNet, a real‑time pose detection tool from the ml5.js library, powered by TensorFlow.js. No identifiable data is collected; instead, blurred and anonymized visuals are generated to safeguard user privacy. This aesthetic choice counters the extractive nature of surveillance‑based design and promotes critical discourse on data ethics.

“Pixels & Privacy: Dance Your Data Away”

II. PROCESS

IDEATION

Brainstorm 1
Brainstorm 2

CREATIVE METHODOLOGY

The project prioritizes process over product. Each prototype is treated as a sketch, a generative moment within an ongoing inquiry into human–machine interaction. Hanul positions themself not only as a maker but as a curator of experience and conversation.

ETHNOGRAPHIC FRAMING

The research is grounded in a hybrid of autoethnography and digital ethnography, informed by Actor‑Network Theory (ANT). ANT frames the relationship between human and machine as an ontologically symmetrical network of actants. The project becomes a site for exploring authorship, agency, and material expression across human and non‑human actors.

III. STUDIES

1. "FAMILY PHOTO‑PAINTING ALBUM"

This experiment humanizes GANs by applying them to emotionally significant images. It transforms neural style transfer into a sentimental act of digital memory‑making.

Model: Generative Adversarial Networks (GAN)
Platform: RunwayML (Style Transfer)
Inputs: Selected original photos of myself & my dog
Styles: Famous Paintings; Cubism, Gogh, Hokusai, Kandinsky, Monet, Picasso, Wu Guanzhong

2. "IN PIXELS"

This visual poetry project explores how classical literature can be reinterpreted by neural networks, bridging ancient verse with generative aesthetics.

Model: Vector Quantized Generative Adversarial Network and Contrastive Language–Image Pre‑training (VQGAN+CLIP)
Platform: NightCafe (Text to Image)
Inputs: Sappho. "Poems of Sappho" Translated by Julia Dunbnoff. University of Houston.

IV. PRODUCT

INTERACTION DESIGN

The interface is simple yet responsive. Participants activate the system through movement, becoming both performer and co‑author. The screen responds in real time, creating a closed feedback loop where body and algorithm shape each other.

TECHNICAL INFRASTRUCTURE

The browser‑native application relies on PoseNet via ml5.js to detect motion and translate it into visual output. No data is stored. This lightweight, accessible setup supports live rendering without backend dependencies, reinforcing user control.

Infrastructure
V. EXPERIENCE

USER FLOW

1. Launch the application via web or mobile browser
2. Grant camera access
3. Move freely as body data is captured in real time
4. Visual feedback appears onscreen in response to motion
5. (Web only) Option to enter fullscreen mode and save screenshots

TESTING ENVIRONMENT

Testing was conducted in‑person across various indoor and outdoor settings with stable internet access. As the experience relies on real‑time visual feedback, it is not fully accessible to users with visual impairments. While participants with limited vision can engage with the movement component, they may not be able to perceive the visual outcomes directly.

VISUAL DISCLAIMER

This experience includes flashing imagery and rapidly changing colors. Viewer discretion is advised for individuals with photosensitive epilepsy or visual sensitivities.

VI. PROTOTYPE

STYLE 1

STYLE 2

VII. ANALYSIS

AGENCY AS INTERDEPENDENCE

Agency, as defined in conventional terms, refers to the capacity to act or exert influence. In this work, agency is reframed as a distributed property shared between human and machine. Rather than treating the machine as a passive tool or the human as the sole originator, the project establishes a mutual responsiveness: the user’s body generates movement, which the machine translates into visual language, which then re‑informs the user’s motion in a continuous feedback loop. Neither the human nor the machine can complete the artwork independently. Their co‑presence is essential.

Actor Network Theory supports this reframing by positing that humans are not ontologically privileged over non‑human actors. In this system, the AI model’s role extends beyond execution; it participates in the interpretation and modulation of user behavior. This reciprocal shaping reflects ANT’s assertion that “actors gain strength only through their alliances” and that agency is always enacted through dynamic, interrelated networks.

AUTHORSHIP AS MUTUAL MEDIATION

The project also calls into question the assumption of singular authorship. While the human initiates and curates the experience, the machine interprets and transforms physical motion into visual output. This relationship is not one‑directional or hierarchical; rather, it embodies mutual authorship. Each actant mediates the behavior of the other, and the final output is a product of their interaction rather than the expression of a single will.

ANT further articulates that actants are transformed through encounters and mediation. In this collaboration, the human is shaped by the constraints and possibilities of the system, while the machine is constantly updated by user feedback. Authorship, then, is not a fixed status but a processual phenomenon that is emergent, negotiated, and contingent on real‑time engagement.

ASSEMBLAGE AND TEMPORAL ONTOLOGY

At the core of this system is what ANT refers to as an assemblage: a temporary constellation of human and non‑human components whose interactions give rise to a particular event or effect. In HCI, the artwork exists only through the convergence of user, machine, code, and interface in a specific moment of live interaction. Once this performance ceases, so does the artwork.

This assemblage‑based model resists the idea of the work as an object with intrinsic meaning or authorship. Instead, it positions the artwork as an effect of relational activity, a momentary crystallization of intent, gesture, and computation. In doing so, the project advocates for a new ethic in AI‑mediated design, grounded not in mastery or ownership but in reciprocity, responsibility, and entanglement.