You do not need a programming background to teach with UNI. You do need a willingness to hold two ideas in your head at once: a picture of what a generative model is doing, and a picture of what your students are doing when they wrestle with it.
This pillar maps the readiness path for teachers who have never opened a terminal. It names what you will need to learn, what you can safely leave to the workbench, and what the honest limits are today. No math-made-easy pitch. No universal-access promise. A working path that some teachers have walked, and some teachers have found is not for them yet.
What UNI is, in one paragraph a teacher can use
Universal Natural Intelligence (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. "Natural, not artificial" is not marketing. It is the design constraint. The system builds a generative model of its situation, updates that model against surprise, and acts to reduce expected surprise. That loop is what your students will learn to see.
The four concepts you will be teaching
A teacher does not need to derive the free-energy functional. A teacher does need a working grasp of four ideas, in language that survives contact with a classroom.
1. Generative model, in plain terms
A generative model is a compressed picture of "how the world tends to go" from a particular point of view. Not a database of facts. A set of expectations that can generate predictions about what comes next. When a student reads a sentence and anticipates the next word, that is a generative model at work (Class E, see Parr, Pezzulo, Friston, 2022, "Active Inference").
The teacher's job is to make the model visible. Not the math of it. The behavior of it: what it predicts, what surprises it, how it changes.
2. Bayesian updating, without the equation
Bayesian updating is the disciplined form of "I had a prior belief, I saw new evidence, now I have a new belief that accounts for both." The equation exists. Teachers do not need to compute it. Teachers do need to model the move honestly in front of students: name the prior, name the evidence, show the update, and name the residual uncertainty. That last step is the one most curricula skip (Class U, based on classroom observation, not a controlled study).
3. Markov blanket, as a boundary metaphor
A Markov blanket is the statistical boundary that separates what is "inside" a system from what is "outside," while letting them influence each other through a defined interface. For teaching purposes, a Markov blanket is a discipline about what a learner has direct access to, what they infer, and what they can only act on. It is the antidote to the fantasy that a student "just gets it" by osmosis (Class E, following Parr and colleagues; Class U as a pedagogical framing).
4. Free-energy minimization, as a direction of travel
You do not need to teach the free-energy principle as physics. You do need to teach the direction: a good learner acts to reduce the gap between what they predict and what they perceive, and updates their model when acting is not enough. Curiosity is not a personality trait in this frame. It is a policy for reducing expected surprise (Class E).
What you do not need to learn
- You do not need to write Elixir, Python, or any other language to teach with UNI.
- You do not need to derive KL divergence. You do need to know it measures how far one distribution sits from another, and that "surprise" in this literature has a precise sense.
- You do not need to run the workbench yourself on day one. You need to know what it shows, and how to read what it shows with a class.
- You do not need to claim the science is settled. It is not. Your calm about that is a teaching asset.
What you do need to learn
- Enough vocabulary to read a workbench trace without freezing: prior, posterior, prediction error, policy, precision (Class B, from workbench inspection).
- Enough Bayes to model an honest update in front of a class, on any topic, in under three minutes.
- Enough about the Markov blanket idea to design an activity where a student's inference is separated from their observation, and both are separated from their action.
- Enough about evidence classes to grade your own claims. We use A empirical-in-session, B code and inspection, C configuration and integration, E expert citation, F falsifier present, U unverified. Ask your students to grade yours (Class C, the class taxonomy is a project convention, not a field standard).
The workbench, and why a non-programmer can use it
The teacher's workbench is a running instance of the UNI stack with the internals made legible. It exposes the generative model's current beliefs, its prediction errors, and the policies it is considering, at a level a teacher can read and narrate. It is not a black box with a friendly wrapper. It is the actual system, with the gates and the traces visible (Class B, from direct inspection; Class C, from the way the workbench is wired to the live services).
You will not be writing code. You will be reading state, asking the workbench to run a scenario, and pausing it at the moments a curriculum needs to open up.
For a longer walk-through, see the teacher's workbench tour. For what an early week of lessons around it looks like, see a teacher's first week with active inference.
Honest limits, stated up front
Some things this path will not do:
- It will not make you a researcher in a term. Reading Parr, Pezzulo, and Friston (2022) is a multi-month practice.
- It will not give you a script for every subject. The workbench is an instrument. Your curriculum is still yours to design.
- It will not remove the discomfort of teaching under uncertainty. It gives you a defensible way to be honest about that uncertainty in front of a class.
We do not claim UNI heals anyone. We do not claim it treats a condition. We claim it is a working hypothesis with growing, evidence-classed evidence, and we invite you to inspect the receipts.
Where to send a math-hungry learner
Some teachers, and some students, want the equations. That is a good instinct, and it is not our lane on this site. Themesis publishes a compact primer on the statistical mechanics vocabulary that sits under active inference: T3, Top Ten Terms in Statistical Mechanics for AI. We recommend it as prep for the UNI Workshop for math-hungry learners; it is complementary, not a substitute.
If a learner wants a hands-on coding path in a different stack than our Elixir workbench, Themesis also offers Building Active Inference in Python; a complementary hands-on Python course with a different stack than our Elixir workbench.
We link these because they are honest work in the same field. We do not paraphrase them. If you use them, you should be able to say what you learned in your own words, and where you disagree.
What comes next
If you have read this far, you are already past the hardest gate: the willingness to sit with a hypothesis and treat evidence classes as a discipline, not a decoration. From here, three moves are worth making, in order.
- Read recommended reading order for teachers so you spend your reading time on the right texts, in the right sequence.
- Take a slow pass through curriculum gates that teach thinking to see how the gate concept translates into unit design.
- When you are ready for a live instrument, come to the workshop. It is where the workbench, the reading, and your own classroom questions meet in the same room.
The path is not fast. It is walkable.