TOUCH//SIGNAL

TOUCH//SIGNAL is a 24-month experiment in two kinds of caring for a living thing. Two paired cedar raised beds are built, one at the artist's home in Marin, one on the LACMA campus. Same dimensions, same soil, same plants. The plant community is a 50/50 mix of Northern California and Southern California natives, drawn from the California Native Plant Society's Calscape database, planted by the artist within a two-day window. Every plant is a local at one site and a transplant at the other. Each bed carries an identical professional sensor station: soil moisture, temperature, pH, and conductivity; air temperature, humidity, light, rainfall; an industrial IP camera. The instruments are the same. The caretakers are not.

At LACMA, the caretaker is an autonomous Claude-based agent. It reads the sensors continuously, watches the garden through the camera, and on schedule and event triggers is prompted with the current state and recent history to decide whether to water, how much, and whether to escalate. It controls irrigation valves directly. It writes a daily journal in its own voice. When it reaches the limits of remote action, it pages a named Los Angeles horticulture collaborator who arrives within 48 hours, performs the physical task, logs what they did, and returns the bed to the agent's care.

In Marin, the caretaker is the artist. The artist tends the garden every day by hand, by instinct, by sticking a finger in the soil. The sensor rig runs in parallel, identical to LACMA's, but the artist does not consult it. Watering happens because the soil feels dry. At LACMA, the sensors are the AI's entire world. In Marin, the same sensors record a relationship that already exists between the artist and the bed, a witness rather than an instruction. Two gardens, read through identical instruments, become two philosophical positions on what data is for.

Both gardens produce, continuously and publicly, four archives: the sensor streams from each site, the AI's daily journal from LACMA, the artist's daily journal from Marin, and a complete log of every decision and intervention. The juxtaposition of the two journals, written by two caretakers reading the same variables into incompatible meanings, is the primary artistic output.

This is a 2026 experiment specifically. Sustained agentic deployments of large language models only recently became possible. Their commercial application is optimization of annual crops for yield. TOUCH//SIGNAL points the same capability at a different problem: supporting a native plant community through its two-year establishment, with an artist composing every prompt, threshold, and escalation rule. The Lab's partnership with Anthropic makes this the right venue for a project the model family is load-bearing for.

The AI garden may fail. It may overwater, miss a pest, lose a plant a human would have saved. If it does, the record of why, what the agent saw, what it wrote about its own failure, is the finding. The Lab's safe-to-fail framing is what makes that commitment possible, and the project intends to take it seriously.

TOUCH//SIGNAL sits in a lineage of artist-led inversions of environmental instrumentation: Tega Brain's "Deep Swamp", Sommerer and Mignonneau's interactive plant systems, Jenna Sutela's bacterial and algorithmic collaborations. It extends that lineage from discrete installation into sustained durational practice. The work continues the artist's "Flora e Fauna" (2022, ongoing), in which ten canvases buried in the Swiss Alps for nine months let a non-human agent become co-author. Every parameter — the prompt language, the sensor thresholds, the journal voice, the native plant community mixing Northern and Southern California species — is a compositional decision the artist holds end to end.

Commercial environmental AI is deployed for optimization: maximize yield, minimize inputs, tune living systems into resources. TOUCH//SIGNAL uses the same stack on a plant community that rejects that frame. Native establishment requires the opposite of annual-crop logic: deep early water, deliberate tapering, minimal intervention once rooted. The work asks what data is for. At LACMA, the sensors are the AI's entire world, every decision flowing from what the instruments say. In Marin, the same sensors record a relationship already in place between the artist and the bed, witness rather than instruction. The 24-month record is the argument.

TOUCH//SIGNAL produces four parallel public archives, released continuously from month one: the sensor streams from both gardens, the AI's daily journal from LACMA, the artist's daily journal from Marin, and a complete log of every decision and intervention. A public portal makes all four queryable, timestamped, and cross-referenced. The physical LACMA garden is visitable throughout. Anchor events include a Year One demo day, participation in the 2027 Lab Biennial Symposium, and a final 2028 Demo Day marking the full archive release. The project is publicly legible from day one rather than delivered at the end.

Two cedar raised beds, 4 × 8 ft × 24 in tall. Twelve plants per bed, four shrubs and eight perennials, fifty-fifty NorCal / SoCal. A free-standing 9-ft mast 5 ft south of each bed carries an Axis P3267-LVE outdoor IP camera (5MP varifocal, IP66/67, NEMA 4X, IK10+, PoE+). At LACMA the mast also carries a museum interpretive panel at eye level. Six RS485 4-in-1 soil probes per bed, one per drip zone, horizontal at 8 in below the surface. One DS18B20 chain per bed at center with sensors at 2 / 8 / 16 in. Six drip zones per bed, six solenoid valves at LACMA under AI control, manual at Marin.