Languages: English (this file) · 日本語
A distributed sound-and-light installation in which 100 edge computers, each running a small language model (SLM) offline on an onboard NPU, form a rhizomatic network modeled on the mycorrhizal communication systems of forests. Each device — a Botanical Intelligence (BI) — receives live biosignals from surrounding vegetation and translates them into whispered fragments of language and synchronized pulses of light, producing a site-responsive soundscape that shifts with the plants' health, stress, and rhythms.
The project reframes AI as Alien Intelligence — not a simulation of human cognition optimized for efficiency, but an unknowable other whose outputs arise from genuinely nonhuman data.
Parallel Forests — 100 devices installed in Umi-no-Mori, Tokyo Bay (a public forest planted over two decades atop a sealed landfill), 13–15 March 2026. Commissioned by Civic Creative Base Tokyo (CCBT) as part of the 2025 Artist Fellow Program.
- 🎥 Installation footage (5 min): https://vimeo.com/1185382935/347a8413b5
- 🎙 Artist & team interview (English subtitles): https://vimeo.com/1185383180/bb4acb131a
- 🌐 Project page: https://obake2ai.info/projects/botanical-intelligence/
Each BI unit ingests two streams of plant physiological data:
- Chlorophyll fluorescence imaging captures photosynthetic vitality through faint red light emitted by leaves (used at night, when ambient light is low).
- Acoustic emissions (AE) are ultrasonic micro-sounds produced when water moves through plant vessels (captured during the day, when transpiration is active).
These signals modulate a learned vector (Soft Prefix) that steers the SLM's generative behavior — when a plant is stressed, the language fractures; when thriving, it opens into longer, melodic phrases. The text is voiced through a spectral processing pipeline inspired by Jóhann Jóhannsson's score for Arrival, layering ghostly vocal residues over time.
Crucially, no single device produces the work. When one BI generates a phrase, it passes to neighbors who extend it before relaying onward — mirroring how chemical signals propagate through root networks. The result is a collective murmur from the forest itself, with no central author.
See docs/technical_description.md for the full technical breakdown.
BotanicalIntelligence/
├── code/ BI system (M5Stack LLM Compute Kit)
├── docs/ Project documentation (statement, technical description, exhibition info)
├── references/ External links
└── related/ Local references to upstream repos (gitignored)
The codebase under code/ is detailed in code/README.md.
- 📜 Artist statement: English · 日本語
- ⚙ Technical description (JA) — system architecture and sensor explanations
- 🌲 Exhibition info (JA) — Parallel Forests venue, schedule, credits
| Role | Name |
|---|---|
| Concept / Direction | Yuma Kishi (@obake2ai) |
| Software Engineering | Ryosuke Nakajima (@rystylee, Qosmo) |
| Technical Direction | Tatsuya Takemori (Abstract Engine) |
| Sound Programming | Kosaku Namikawa (7ild3) |
| Lighting Design | Namiko Watanabe |
The source code under code/ is ported from rystylee/CCBT-2025-Parallel-Botanical-Garden-Proto (kishi branch). The foundational implementation of inter-device communication, signal processing, and the LLM inference / audio output pipeline was authored by Ryosuke Nakajima; Yuma Kishi added the Soft-Prefix mechanism (which feeds plant biosignals into LLM inference) and the overall project coordination.
- Code (
code/): Apache License 2.0 - Documentation (
docs/): CC BY 4.0
Full attribution and contributor list: NOTICE.
This project was developed as part of the CCBT (Civic Creative Base Tokyo) Artist Fellow Program 2025, supported by Arts Council Tokyo and Tokyo Metropolitan Foundation for History and Culture.
The sensor systems for Parallel Forests were realized in collaboration with:
- Kensuke Kageyama (Saitama University / FeelSensing) — touch-acoustic-emission sensing for daytime plant activity
- Kotaro Takayama (Toyohashi University of Technology / Ehime University) — chlorophyll fluorescence imaging for nighttime plant signaling
If you reference this project in academic or critical writing, please cite:
Kishi, Y. et al. (2026). Botanical Intelligence (Parallel Forests).
Civic Creative Base Tokyo (CCBT) Artist Fellow Program 2025.
https://github.com/obake2ai/BotanicalIntelligence