Liability

AI interlocutor for situated dialogue, bias exposure, and mediation between devices and publics.

λ Interface context

⛟ Conversational Mediation Infrastructure

Liability is deployed as a conversational interface in public contexts where response time, ambient noise, and user behavior continuously change. The system is designed to remain legible during interaction, so visitors can read how prompts, pauses, and replies are produced in real time.

This section introduces the live trial as an infrastructural test rather than a demo. The next snap opens the active interface, while the following section returns to interpretation and operational notes.

⊕ Operational and curatorial framework

◒ After Interaction

After activation, Liability reveals itself as a paced conversational system rather than a seamless assistant. It listens, interprets, responds, and pauses in ways that make timing and hesitation visible. This rhythm allows visitors to read the interaction as a public process, not as a hidden computation. In this sense, the project turns dialogue into a shared situation, where machine behavior is part of the scene.

Its logic is distributed and contextual. Each deployment keeps the same structural identity while adapting to local acoustics, social dynamics, and spatial conditions. The work does not seek one perfect response model; it privileges legibility, variation, and continuity across sites. This keeps the installation grounded in place and audience, and supports a critical reading of how conversational systems frame authority.

The technical footprint remains compact, portable, and inspectable. Local operation is prioritized, with optional connectivity when needed, so the piece can remain stable in temporary or low-infrastructure contexts. This supports a practical curatorial objective: an AI installation that is reproducible, maintainable, and publicly understandable without depending on opaque centralized workflows.

Ambient Signal Mesh

Continuously detects pressure changes, echo patterns and partial speech sounds from the environment. No audio is stored; all processing happens in real time.

Trigger and Autonomy Module

Enables the system to decide when to speak based on detected presence. There are no commands or activation phrases, speech starts without user input.

Disruptive Dialogue Engine

Constructs responses that are intentionally irregular. Uses repetition, delay and non-linear phrasing to reveal systemic language patterns.

Contextual Emission Array

Projects sound into the space without directing it to a single listener. Speech is shaped by the context in which Liability is placed.

Transient Privacy Shell

Processes input locally without keeping any record of conversations. This ensures privacy by preventing speech data from being saved or reused.

Autonomous Field Stack

Runs on portable, energy-efficient components with no internet needed. Ideal for deployment in urban, rural or transitional environments.

⚙ Technical

⧗ Technical Specs and Operational Notes

Area	Spec	Operational Notes
Core Function	Autonomous conversational system that activates from live environmental sensing and emits context-driven speech.	Designed to start interaction without prompts, exposing interaction bias through machine-led dialogue.
Motion Sensors	PIR/proximity sensing tracks presence, movement direction, and dwell time in the surrounding area.	Movement events are used as primary triggers before speech emission logic is evaluated.
Microphone Array	Multi-microphone capture for ambient sound, partial voice fragments, and spatial reverberation cues.	Supports local directionality and context detection without storing conversation recordings.
Audio Analysis	On-device DSP pipeline analyses level, spectral energy, transient changes, and room echo response.	Combines motion and audio thresholds to decide when and how the system should speak.
Speaker System	Full-range amplified speaker output for spatial projection rather than headset/private playback.	Voice is emitted into shared space, adapting perceived intensity to local acoustic conditions.
Dialogue Engine	Response model configured for non-linear phrasing, controlled delay, and disruptive repetition.	Built to reveal normative language patterns instead of optimizing for smooth assistant behavior.
Privacy Model	Ephemeral local processing; no persistent storage of raw speech or conversation logs.	No voice archive is retained, exported, or reused after interaction cycles.
Deployment & Power	Portable low-power stack with battery-first operation and optional modular recharge.	Designed for off-grid installations in galleries, urban sites, and transitional environments.
Use Context	Research and exhibition prototype focused on agency, protocol, and infrastructural bias.	Not intended for productivity tasks or deterministic informational assistance.