An MIT study we can't stop thinking about, and what it means for plant music.
A few weeks ago we read a paper we have not stopped talking about. The headline finding, in one sentence: a research team at MIT trained a neural network to classify a person's emotional state using nothing but the electrical activity of a plant sitting near them, and it worked.
Before we go any further, let us say what this is and what this is not. It is not proof that plants are conscious. It is not proof that they feel human emotions in any sense that a human would recognize. It is not a magic trick.
What it is is one of the cleanest measurements we have ever seen of something many PlantWave listeners have noticed for years and could not quite name. Plants are continuously responsive to the people near them, in ways measurable enough to train a machine learning model on. That responsiveness is the same underlying phenomenon biosonification translates into sound.
What Did the MIT Plant Emotions Study Actually Measure?
The study was led by Peter Gloor, a research scientist at MIT, working with collaborators at the University of Cologne. The paper, Plant Bioelectrical Signals for Environmental and Emotional State Classification, was published in Biosensors in 2024.
The setup is incredibly simple. A small houseplant (typically a Kalanchoe or a Purple Heart, Tradescantia pallida) gets a pair of electrodes attached to one of its leaves and a ground wire pushed into the soil. A person sits in the room with the plant. Sometimes the person is feeling happy. Sometimes sad. A camera quietly tracks the person's face to verify what state they are actually in.
The researchers record the plant's voltage signal across many sessions. Then they take that signal alone, no camera data, no human bio-data, just the plant, and train a neural network on it. The network learns to classify whether the human in the room was happy or sad.
In the published 2024 paper, it did so with 73% accuracy.
That 73% figure was the headline number from the published paper. In the two years since, the research program has expanded the experiment. More plants. More people. A wider range of emotional states. They have moved from a binary happy-or-sad classifier to a four-state model that includes angry, happy, sad, and neutral.
The latest validated accuracy, on that four-state model, is 97%.
That number is on the public research dashboard at biolingo.org. A 97% classification accuracy on four emotional states is not the kind of result a single study produces by accident. It is the kind of result that suggests the underlying signal is real.
How Can a Plant Detect a Person's Emotional State?
The honest answer is that the field is still figuring this out. The current best-supported hypothesis is the electric field of the human heart.
The human heart produces a small but measurable electric field that extends a few feet beyond the body. It is what an ECG machine is reading when a cardiologist hooks you up at a checkup. The MIT-affiliated researchers use the same kind of sensor (a hobbyist-grade chip called an AD8232) on the plant, and a Polar chest strap on the human, and record both signals at the same time.
Then they run causality analysis on the two streams. Changes in the human's cardiac signal precede correlated changes in the plant's voltage signal, with a delay of anywhere from a few seconds to about half a minute.
Other signals correlate too. Changes in CO2 from breathing. Volatile organic compounds in the air. Small thermal shifts. The heart-field signal is the cleanest one so far, but plant electrical signaling appears to integrate several inputs at once.
When you sit next to a plant, your heart is broadcasting, very gently, on a frequency the plant can pick up.
Is Plant Emotion Detection Real Science?
It is, and the institutional record is what makes that clear.
Peter Gloor has been a research scientist at MIT for thirty years and is an honorary professor at the University of Cologne. His Cologne research seminar has run continuously for two decades, producing roughly thirty student projects a semester on plants, animals, and machine learning.
The plant emotion detection work runs through a five-year research consortium called BioLingo, spanning nine institutions across the United States, Switzerland, Germany, Spain, and Poland, with more than twenty researchers and twelve peer-reviewed papers published to date.
When we say "an MIT study," we are pointing at one paper from a larger, peer-reviewed program. That distinction matters, because the science behind PlantWave is not validated by single anecdotal studies. It is validated by replicated, peer-reviewed research, conducted by credentialed institutions, accumulating over years.
The BioLingo program's hypothesis for why plants can detect human emotional state in the first place is evolutionary. Plants evolved electrical signaling as an early-warning system for approaching herbivores. A sheep approaching a meadow is bad news for the meadow. A dog approaching is less bad but still worth noticing. A human approaching is generally fine. Plants distinguish these in roughly the order of danger the animals historically posed.
Plants, in this telling, have been listening to animal hearts for hundreds of millions of years. We are just one more animal who walked into the room. Gloor describes his own research as "always on the bleeding edge, and on the leading edge." Coming from him, that lands as understatement.
Plant Bioelectrical Signals vs. PlantWave's Biosonification
Here is where the MIT research connects to the science PlantWave is built on, and where the two approaches differ.
The BioLingo group measures plants the way a doctor measures a heart: with a voltage-sensitive ECG-style sensor that picks up the actual electrical pulses moving across the plant's tissue.
PlantWave measures plants a different way. PlantWave's hardware reads the slow rise and fall of electrical conductivity on the leaf surface, an impedance-based measurement of the same underlying living tissue. This is the foundation of biosonification, the field that translates biological signals into sound.
We are reading a related electrical signal from the same plant tissue. Not the same signal. Both are legitimate biosonification approaches. They are two different windows into the same continuously responsive organism.
What the MIT bioelectrical signals research validates, for our purposes, is the bigger claim sitting underneath everything PlantWave does. Plants are not passive. They are continuously responsive to the people and creatures and conditions around them, in ways measurable enough to train a neural network on. That responsiveness is what PlantWave translates into music. Different sensor, same underlying truth.
PlantWave is cited in BioLingo's published research. The relationship is active.
How to Listen to Your Plant After Reading This
You do not have to wait for the science to catch up.
The next time you sit with your plant and PlantWave, try this listening exercise. Spend five minutes in stillness. Notice what the music is doing. Then bring to mind something that makes you genuinely happy: a memory, a person, a place. Notice if anything shifts. Then bring to mind something that makes you genuinely sad. Notice again.
We are not promising you will hear a difference every time. The signal is small and the variables are many. But the next time someone asks whether the plant is actually doing anything, or whether what you are hearing is just random, you will have an answer.
A research team at MIT and a network of European labs trained a model on the electrical activity of a Tradescantia plant and used it to read whether the human in the room was happy or sad. The model now hits 97% on four emotional states. The plant did not know any of this was being recorded. It was simply doing what it has been doing for hundreds of millions of years.
The plant in front of you is not blank. It has been in the room with you the whole time. We make a device that lets you hear what that sounds like.
Key Takeaways
- An MIT-led research program trained a neural network on plant bioelectrical signals alone to classify human emotional states.
- The published 2024 study reported 73% accuracy on a binary happy/sad classification. Subsequent validated work has reached 97% accuracy on a four-state model (angry, happy, sad, neutral).
- The best-current-guess mechanism is the electric field of the human heart, with CO2 and volatile organic compounds as secondary signals.
- The work is part of BioLingo, a five-year, nine-institution, twenty-researcher consortium with twelve peer-reviewed papers to date. PlantWave is cited in their work.
- PlantWave reads a related but distinct electrical signal from the same plant tissue (impedance / conductivity-based, not voltage), making both PlantWave and the BioLingo research legitimate approaches to biosonification.
Frequently Asked Questions
Can plants really detect human emotions?
Peer-reviewed research from MIT and the BioLingo consortium has demonstrated that a neural network trained on plant bioelectrical signals alone can classify human emotional states with up to 97% accuracy on a four-state model. The leading hypothesis for how plants do this is the electric field of the human heart. The plant is not "feeling" emotions in a human sense; it is responding to measurable signals the human body emits, which happen to vary with emotional state.
What are plant bioelectrical signals?
Plant bioelectrical signals are the small, continuous changes in voltage and conductivity across living plant tissue. Plants use these signals internally to respond to light, touch, temperature, water, herbivore damage, and the presence of nearby animals. Modern sensors can read these signals non-invasively from the leaf surface. PlantWave reads conductivity changes; the MIT-affiliated research reads voltage changes. Both are different windows into the same underlying plant electrical activity.
How accurate is the MIT plant emotion study?
The published 2024 paper in Biosensors reported 73% accuracy on a binary happy-versus-sad classification. The latest validated work, as published on the BioLingo research dashboard, reaches 97% accuracy on a four-state model (angry, happy, sad, neutral).
Is plant emotion detection the same thing as plant music?
No, though they share a foundation. Plant emotion detection uses a machine learning classifier to read a discrete state from a plant's bioelectrical signal. Plant music uses a real-time sonification engine to translate the same family of signals into continuous, expressive sound. The two are siblings under biosonification: one classifies, one translates.
What is BioLingo?
BioLingo is a five-year international research consortium led by Peter Gloor at MIT, with collaborating institutions in the United States, Switzerland, Germany, Spain, and Poland. The program has published twelve peer-reviewed papers on plant bioelectrical signaling, animal-presence detection, and emotional-state classification. PlantWave is cited in the published research.
How does PlantWave compare to the MIT research?
PlantWave uses an impedance-based sensor that reads electrical conductivity changes on the leaf surface, optimized for real-time generative music. The MIT-affiliated research uses a voltage-based ECG-style sensor optimized for classification tasks. Different sensors, related signals, complementary purposes.
PlantWave translates plants' bioelectrical signals into real-time generative music. Learn more →
Studies and sources referenced: Gloor, P. A. (2024). "Plant Bioelectrical Signals for Environmental and Emotional State Classification." Biosensors, 15, 744. BioLingo research program: biolingo.org, twelve peer-reviewed papers across nine institutions, latest validated emotion-classification accuracy 97% on a four-state model.
