2021 Interactive Generative Installation

Wunderkammer

Interactive Generative Installation

Video

System activation — Network begins learning cycle

After 1 hour — High adaptive responsiveness

After 30 hours — Declining interpolation capacity

After 70 hours — Network exhaustion

Concept

Renaissance cabinet of curiosities transformed into intelligent sonic organism. Commissioned for Spazio Materia, Prato—a 15th-century pavilion-vaulted building. The installation responds to visitor movements and sounds through neural network-based sound generation. Each acoustic event modifies the network's interpolation space, creating unique sound objects as audible traces of human presence.

Finite life cycle: approximately 72 hours, mirroring biological organisms. Born with excellent adaptive capabilities, matures through visitor interaction, gradually exhausts capacity for novelty. At the beginning, visitors exert tangible control. As hours progress, the system becomes less reactive, settling into homogeneous patterns. By 70 hours, the work achieves senescence—visitors witness an autonomous system playing out its final algorithmic gestures.

Questions the nature of AI systems, agency, and the relationship between human action and machine response. Unlike traditional interactive installations promising infinite responsiveness, Wunderkammer acknowledges entropy, limitation, and temporal boundaries of computational processes.

Specifications

Premiere Spazio Materia, Prato — May 27–30, 2021
Duration ~72 hours (continuous operation)
Playback 4.1 audio channels + neon light installation
Tools Ableton Live, Faust, Max/MSP, Python
Architecture 15th-century pavilion vault structure

System Architecture

Wunderkammer control interface

Synthesis parameter manipulation and preset creation

Wunderkammer neural network visualization

Neural network parameter space with preset distribution

Physical acoustic modeling, spectro-morphological analysis, neural network parameter interpolation. Visitor sounds analyzed for timbre, spectral content, amplitude, temporal evolution. High-dimensional feature vectors navigate the network's parameter space. Regression-based neural network reads parameter space, enabling smooth interpolation between synthesis states. As the installation runs, novelty diminishes—exhaustion phase emerges.

Credits

Artistic Direction & Audio Programming
Alessandro Anatrini
Light Design
Jacopo Buono
Production
Mosè Risaliti, Massimiliano Fortunati