Teachers do not need to become computational neuroscientists to teach with active inference. They do need a small, honest vocabulary that says the same thing in the classroom that it says in the paper.
This is that vocabulary. Print it, tape it above the whiteboard, and use it plainly.
Why a shared glossary matters
When a student says "the model is wrong," a teacher grounded in this vocabulary can ask: which model, whose prediction, and what error? That is a different conversation than "the AI got it wrong." UNI is a working hypothesis on an attainable path toward General Natural Intelligence: a natural, active-inference approach whose evidence is growing, evidence-classed, and tested in the open. Do not take the claim on faith. Test the build, inspect the gates, and help us find where it fails.
The definitions below are anchored in the standard active-inference literature (Class E: Parr, Pezzulo, Friston, 2022, "Active Inference: The Free Energy Principle in Mind, Brain, and Behavior"). We keep them classroom-length. Deeper treatments are linked at the bottom.
The glossary
Generative model. A learner's internal model of how the world produces the things they observe. In the classroom: the story a student holds about what should happen next in a lesson, an experiment, or a conversation. Every learner has one, whether they can articulate it or not (Class E).
Prediction error. The gap between what the generative model expected and what actually showed up. In the classroom: the moment a student says, "wait, that is not what I thought would happen." That moment is not failure. It is the signal the model needs to update (Class E).
Free energy. A single quantity that combines two costs: how surprised the learner is by what happened, and how far the learner's model had to bend to fit it. Active-inference learners act to reduce this quantity over time. In the classroom: a student who both notices surprise AND updates their story is doing the work. A student who just notices without updating, or updates without noticing, is doing half of it (Class E).
Markov blanket. The boundary that separates what is inside the learner's model from what is outside. Information crosses the blanket in two directions: sensory input coming in, actions going out. In the classroom: it is the reason a worksheet does not teach the same way a lab bench does. The blanket is different. What crosses it is different (Class E).
KL divergence. A measurement of how different two probability distributions are, in this case the distribution the learner expected versus the distribution the world delivered. It is one honest way to score "how surprised was the model, really?" In the classroom: you do not need the equation to teach the intuition. You need students to feel the difference between a small update and a large one (Class E). A gentler walk-through lives in KL divergence for teachers who do not love math.
How to use it in a lesson
Post it. Refer to it. When a student asks a question, name which term is in play. "That is a prediction error, what does your generative model want to do with it?" is a real sentence a teacher can say to a real ninth grader, and it changes what the ninth grader does next.
The vocabulary is the pedagogy. When the words are precise, the thinking gets precise. When the words are vague ("the AI thinks," "the computer knows"), the thinking gets vague to match.
Our own classroom materials are wired to this vocabulary end to end (Class C, verifiable by inspecting the content packs shipped with the workshop kit). We do not swap in softer synonyms for the softer audiences. The kindergarten version of "prediction error" is still "prediction error," just with a picture of a jack-in-the-box next to it.
What this glossary is not
It is not a claim that active inference is settled science, and it is not a claim that our build of it is finished. It is a working vocabulary for a working hypothesis. If a term here turns out to be the wrong frame for a classroom, we will say so, in public, with the receipt.
We also do not use the word "AI" for our own work. The reasons are laid out in Why we avoid the word AI for our own work.
Where to go next
- What active inference actually says: the paragraph-length version of the whole frame.
- Generative models, a metaphor that holds up: the term that does the most work in a classroom, unpacked.
- KL divergence for teachers who do not love math: the one term teachers most often flinch at, made teachable.
- The workshop, what teachers actually do: a day of practicing this vocabulary on live classroom material.