[ Foundation ]
Mapping human emotions
Three years ago, another person we loved started to disappear. A name forgotten here, a story repeated there.
The diagnosis had a clinical name, but the things it took didn’t. There’s no scan that captures the weight someone gives to loyalty, their biggest pet peeve, or the precise texture of their stubbornness.
When we disappear, dementia or not, those things are gone. Living, ever distorted, in the ones who remember.
But… does it have to be this way?
If we can’t stop the loss, maybe we can capture what is being lost. Not the memories or the voice, but the very layer underneath governing it all. Making you, you.
We believed it was not only possible, but a necessity. Everyone seems focused on memory, language, advanced reasoning, treating the emotional layer like a given, something already settled, hardly worth a second glance. We couldn’t disagree more. What makes us human must endure.
We built a foundation and named it from the Greek Metempsychosis: the transmigration of the soul. We’re not mystics, they were just asking the question three millennia too early.
Three years of R&D later, we have a working model. Here’s what we built:
Emotional Architecture (EMA)
A system that learns how you feel about things, and can predict how you’d feel about things you’ve never seen.
When you hear that a friend lied to protect someone, something happens before you form an opinion. You assign a weight, drawing from your history, your values, your relationship to honesty and loyalty. It happens fast, it’s consistent, and it’s yours. Someone else hearing the same thing assigns a different weight simply because their internal wiring is different.
Psychology calls some of these moments self-defining memories: the load-bearing walls of your identity. Our model captures this wiring, not the memory itself. With enough honest responses, it maps how you weigh the world. Feed it a completely new scenario, it predicts your response.
Two people, same scenario, two different signatures. The difference is their Emotional Architecture.
Results
We tested EMA using language models across 9k+ scenarios covering hundreds of emotional traits across dozens of life contexts.
The choice of LLMs was strategic for two reasons.
// 01
Their responses are a diluted residue of human emotional behavior, compressed into statistical patterns with no individual bias and no performance. Finding a replicable structure in that collective noise would validate the approach on the hardest possible test.
// 02
If different models produced different emotional maps, it would mean emotional architecture is not an artifact of one system. Different training data and different designs produce genuinely distinct emotional signatures, the same way different humans do.
Both predictions held. The maps were structured, clustering along axes corresponding to autonomy, compassion, justice, courage, without being programmed to look for them. And different models produced different maps.
The next step: apply this to humans, at scale.
How we collect
We show you a situation, an intention, and an action performed. You tell us if it made you feel good or bad. That’s it.
This is intentionally simple to avoid bias or over-analysis. The scenarios are designed to probe different regions of emotional space. You don’t know which region is being probed, this is also intentional to get your raw reaction before the belief systems kick in.
Each response is tied to an anonymous hash on your device. No account, no email, no name. If you want your answer removed, contact us with your hash.
Our mission
Build the missing emotional layer that current AI systems lack.
Map humanity’s emotional architecture at a scale never attempted before.
Make individual emotional signatures possible to capture, once the collective map allows it.
Safeguard what we collect for the long term, so this work outlives us.
Our plan
Map the models.
Apply our architecture to language models. Validate the approach.
Map humanity.
Collect human responses at scale. Same scenarios, same architecture, real people. And we confront them with EMA’s predictions, measuring its performance and nudging it accordingly.
Map an individual.
Once humanity’s map has enough resolution, attempt to map a single person. This is the phase that matters most, and the phase we cannot rush. A map of one person is only meaningful against the map of everyone.
Who we are
We are two engineers, friends for over 20 years, who decided to explore ways to preserve what makes someone who they are, before it’s too late.
Edouard
[ One line on background, role, or what drew him to this specific problem ]
Alexandre
More than a decade in the industrial world taught me that complex systems fail not from bad engineering, but from human emotions no one can fully predict. METEM became my way of making sure my daughters could carry a part of me forward, and find answers to the questions I won’t be there to answer.
Metem Foundation is a registered 501(c)(3) nonprofit. No investors. Funded entirely by people who believe this work should exist.
Be part of this
You can be part of this journey in two ways:
Add your mark. Go test EMA, see how it performs, and nudge it if you feel it needs to. Every response adds resolution to the map. And when enough resolution exists, we’ll know whether we can map a single mind. This map is incomplete without you.
Donate if you can. No donation is too small. This accelerates the research and unlocks long-term storage safeguards.