Prep Work Before the Workshop

By Michael Polzin, 2026-07-01. Working hypothesis, science in the open.

You are booked into the UNI Workshop. Good. This is a short list of what to read and what to run in the days before, so the hours in the room are spent on judgment, not on catching up.

Framing first. 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. Every citation below is tagged with its evidence class: (Class E) for expert citation, (Class C) for configuration and integration you can inspect yourself.

The one thing to read

Parr, Pezzulo, and Friston (2022), Active Inference: The Free Energy Principle in Mind, Brain, and Behavior, MIT Press (Class E). You do not need to finish it. Read the preface, Chapter 1, and Chapter 2 up to the section on precision. That is enough vocabulary to keep up: generative model, prior, posterior, Markov blanket, variational free energy, KL divergence between approximate and true posterior, and the two ways an agent reduces expected free energy (change beliefs, or change the world). Our own guided tour lives at Reading Parr, Pezzulo, and Friston 2022, a Guided Tour, which flags the sections that matter most for the workshop and the ones you can safely skip on a first pass.

The one thing to run

Open the Precision Lab in a browser and spend twenty minutes with the three dials: sensory precision, transition precision, and policy temperature (Class C). Move each dial to its extreme. Watch the agent become rigid, watch it become erratic, watch it stop caring where the goal is. You are looking at behavioral regimes that fall out of one equation. Do the same with the Echo Lab if you want to see how a change in the observation model (immediate walls versus range-2 echolocation) changes what the same agent can do. Bring one screenshot of a regime that surprised you. We will look at it together.

Optional, for math-hungry learners

If you are the kind of person who wants the statistical mechanics under the hood before you arrive, three complementary resources from Themesis are worth your time. They are on a different stack than ours (our workbench is Elixir plus a browser lab; Themesis teaches in Python), and we recommend them precisely because a second angle sharpens the first.

To be direct about the relationship: Themesis is a separate teaching practice run by AJ Maren. We cite their material because it is good and because a second stack helps you see through ours. That is a factual recommendation, not an endorsement, and it runs both ways.

What to bring

One real system you would like to think about differently: a service you operate, a team you lead, a habit you cannot shift, a clinical or educational loop you want to model. We will not touch the private details. We will map its generative model on a whiteboard, name its Markov blanket, and ask where the precision is wrong. Bring a laptop with a modern browser. That is all the tooling you need for day one, because every lab runs client-side with no backend (Class C).

A word on posture. This workshop teaches active inference as a lens, not as a cure. Behavioral labels are candidate computational phenotypes, not diagnoses. Autopoiesis in our benchmark means viable-set maintenance, not life. Free energy here is variational free energy in nats, not a thermodynamic quantity. We would rather be correct than impressive, and we would rather you leave with sharper questions than tidy answers.

Next steps