What can a plant tell us about the place it lives?
A plant responds to light, water and air continuously, and a change leaves a trace that outlasts it. This is a proposal to find out whether those traces carry information we can use: three instrumented plants on a bench for three months, a known pattern of light, and a simple rule that tries to read the plant. If the rule learns nothing that ordinary sensors did not already know, that is the answer. If it does, there is a much larger experiment waiting behind it.
| Status: | Concept; not yet funded or sited |
| First test: | 3 rigs, 3 months |
| Then: | an array of 8, a year |
| What we change: | The light on each plant, on a schedule we choose: a recorded pattern in a fixed window each day. Nothing else, in the first test. |
| What we measure: | Tiny electrical changes on the stem and changes in leaf thickness, continuously, next to ordinary instruments for light, temperature, humidity and soil moisture. Once a week an independent instrument checks what the leaf is actually doing. |
| How we check: | A simple rule, a weighted sum of the recorded signals, tries to predict what the plant will do next and to recognize which light pattern just happened. Every answer is compared with the same rule run on ordinary sensors alone, and on a rig with no plant at all, on days the rule never saw. |
| What it decides: | Four yes-or-no questions: is the recording reliable, is the response repeatable, does the plant add anything beyond the instruments, and is it still true two weeks later? Four yeses build the array. Four noes get published too. |
In one trial we change the light according to a recorded pattern. Before a later trial, the rule predicts the plant's response, and we compare the forecast with what the instruments record. In a second test we hide the light pattern and ask whether the plant's signals reveal which pattern occurred. Whether either works is unknown.
The method is called reservoir computing. It uses a physical system's changing responses to transform information over time, then trains a simple readout to turn measurements of that system into an answer. Here the physical system is a plant. The claim under test is narrow: that the plant's measured dynamics hold useful memory of recent inputs, so that a deliberately simple readout can do something ordinary sensors and a comparable conventional model cannot do as well. [shown in earlier research] In 2022 a group in Belgium showed something like this on strawberry plants over eight days; in 2025 a group in Germany recorded ivy outdoors for five months. Neither ran a test built to roll into something larger.
| Rig | What it is | Why it is there |
|---|---|---|
| A | A strawberry plant with leaf clips and stem electrodes. | The species the 2022 study used, so the first result can be compared with theirs. |
| B | An English ivy in a pot with the same sensors. | The species that continues into the larger array; evergreen, and the one with a published multi-month outdoor record. |
| C | No plant: the same sensors on a passive stand-in, the same light, the same box. | Anything this rig appears to learn comes from the instruments. |
Small on purpose. The first test uses the same sensors, the same code and the same rules the larger array would, so nothing learned here has to be relearned at scale.
A plant senses its surroundings for as long as it lives. A leaf integrates light, water, temperature, chemistry and touch as part of being alive, and it does so on sunlight, repairing itself, for years. If that integration can be read, a living organism becomes an environmental sentinel: something whose measured response helps reveal the conditions around it, in a greenhouse, a building, or one day a sealed habitat with no biosphere to borrow from. That is the near reason.
The far reason is a question about computing. Silicon's energy per operation has stopped falling fast, and the unconventional-computing programs are funding every physical system that might carry part of the load: memristors, photonic loops, analog chips. A plant is the cheapest such system and the hardest to read: it drifts, grows and goes dormant in winter. Methods that can read it would apply to every physical system that drifts, whether or not the plant itself is useful. [possible later experiment] The versions of that idea that reach further, canopies paired with computers, kelp arrays that read the ocean while taking up carbon, a habitat whose garden is also its first inference layer, each need their own evidence, and the full brief says what evidence.
If the first test passes, this is what it rolls into: eight vines on ten-foot arbors around a table, a greenhouse that is also a place to sit, and a full year of seasons recorded through the plants. The drawing shows the proposal; no site is chosen.