A word choice is not decoration. When this program uses the phrase General Natural Intelligence, the word natural is doing real work, and the word artificial is being kept out on purpose. This piece explains why.
The short version: our program is a working hypothesis on an attainable path toward General Natural Intelligence, natural not artificial. That sentence is the frame. Every claim on our sites is offered as calibrated signal, not verdict, and we do not use the language of the AI industry to describe our own work.
What the word "natural" is doing
Active inference, the science underneath this program, describes how a living system reduces the gap between what it expects and what it senses, by acting on the world and by updating its internal model. That is the technical spine, published in Parr, Pezzulo and Friston, Active Inference (MIT Press, 2022). Class E It is a theory of how nature already learns. It was not invented to make software cleverer. It was written down to explain how brains, bodies, and cells manage uncertainty in the wild.
When we say natural intelligence, we are pointing at that mechanism. When we say General Natural Intelligence, we are pointing at the direction that mechanism generalizes, across scales, across species, and across the systems people build inside institutions and classrooms. The word path matters. We are describing a research direction with visible mile markers, not a finished product with a ribbon on it.
What the word "artificial" would smuggle in
The phrase artificial intelligence carries baggage this program does not want to inherit. It suggests something built to imitate a mind from the outside, trained on scraped human output, and marketed as arriving. It also carries a cultural expectation that the vendor knows more than they can prove, and that the buyer should accept the claim on trust.
That is not what we are doing, and it is not the posture we want. The honest posture is that this is a build program with published gates, an evidence ledger, and a falsifier for every non-trivial claim. Class C Calling that build artificial intelligence would smuggle in the wrong social contract. Calling it natural, with the word path attached, keeps the contract honest: growing evidence, science in the open, and room to be wrong.
What we do not claim about our own work
The frame gets its strength from what it refuses to say. The list below is not a legal boilerplate. It is the actual red-line list this family operates by.
- We do not claim to have general intelligence, and we do not use the phrase artificial general intelligence for our work. Anyone doing so is speaking past the evidence.
- We do not claim endorsement by third parties who have not endorsed us. Where a scholar's public writing informs the field, we link the writing and describe it in our own words, and we do not put their name on our roadmap.
- We do not claim to heal, treat, or lower trauma, and we do not claim to rewire brains. The pedagogical work here is trauma-informed and non-clinical. Trauma care belongs to licensed clinicians.
- We do not claim the math is easy or that anyone can do it without effort. The math is real and it takes time. The plain-language on-ramp is the door, not the destination.
- We do not claim final results. Where a claim is uncertain, we mark it as unverified. Class U Where a claim has a specific test that would break it, we publish the test with it.
Why the frame matters right now
The field is moving. Vendors are labeling more and more of their output as intelligence, and the definitions are stretching. That is exactly the moment for a research program to pick its words with care. If everyone on the block calls their software artificial intelligence, the phrase stops meaning anything useful, and the honest builds get lumped in with the marketing.
Choosing natural is a way of stepping out of that lump. It says: the mechanism we care about is the one nature already uses, we are writing it down carefully, and we are willing to be measured against it. It also says: we are not selling the illusion that a machine is a person. The systems this program builds are tools with visible seams. Human-in-the-loop is not a warning label bolted on later. It is how the loop is designed from the start.
How to read anything else on this site
If a page on our sites talks about intelligence without qualifying it, read it as shorthand for the frame in this post, not as a claim to have arrived. Every non-trivial claim carries an evidence class: A for empirical-in-session, B for code or inspection, C for configuration or integration, E for expert citation, F for a falsifier present, U for unverified. If a claim on our sites does not carry one of those tags and it sounds strong, that is a bug. Bring it, and we will fix it.
The full method and taxonomy live on the transparency page. The mechanism itself, prediction and update and felt safety, is introduced in the prediction primer. The place where humans and teams practice the loop together is the workshop. Those three doors together are the whole invitation.
Transparency and evidence classes
How we label every non-trivial claim, and how to spot one that has slipped past the tag.
The mechanismHow prediction shapes experience
The plain-language primer on prediction, error, and felt safety that sits under the frame.
The workshopWork with us directly
Two days of practicing the loop with humans in the room. Applications and details.
The mathUNI, the science front
The technical side of the same program, with citations, benchmarks, and open gates.
Evidence classes used in this post: Class C configuration and integration, Class E expert citation, Class U unverified. Full method and taxonomy at the transparency page.