The paper
One click from any big claim to the primary source.
When we say something, we want you to be able to check it. So here is the work itself: the record, its honest status, what it claims in plain terms, and the one piece of math we added on top, named out loud. Read it. Try to break it.
- Title
- An Organic Operator and AI Operator Collaborative Review of Active Inference Free Energy Minimization
- Authors
- Polzin et al. (2026)
- Status
- Unrefereed preprint. Layer 2 expert review pending.
- DOI
- 10.5281/zenodo.19785799
A preprint is the work, posted in the open before formal review. It is not peer reviewed yet. We say that plainly so you read it for what it is: a careful argument, out where anyone can check it.
What it is
What it is not
In plain terms
What the paper is about.
Active inference is one way of describing how a mind acts under uncertainty: it builds a model of the world, then moves to make the world match what it expects. One quantity, called precision, sets how much the mind trusts what it senses. The paper works through that math and shows, in a small simulated maze, how turning precision up or down changes the behavior you see: from confident and decisive to lost and wandering.
The point underneath it is simple. When the present feels less frightening, the future feels safe enough to explore. That is the line UNI is built on, and the paper is the careful version of it.
I did the math. Scarcity is an accounting, and what is manufactured can be dismantled.
Disclosed in full
The one piece of math we added.
We will not hide our changes inside someone else's result. The verified engine puts the agent's preferences over what it observes. We added one thing: a preference over the hidden states themselves, a goal prior. In plain terms, it gives the agent a reason to head toward the goal instead of stalling a few tiles short of it.
A goal prior over hidden states:
P_pref(s) ∝ exp(−γ · distance‑to‑goal)
included in expected free energy as expected surprise, −E_q[ln P_pref(s)], exactly parallel to the observation‑preference term.
This is a standard active-inference construct. It changes the agent's generative model, and nothing else: the precision-weighting math is unchanged, and the maze worlds are pure geometry. We name it here so you can hold the result and the change side by side.
Read it, then test it
Go to the source. Try to break it.
The Zenodo record is the primary source, with a citable DOI. The labs let you move the dials yourself and watch the behavior change. If a claim does not hold, we want to know. That is the honest version of confidence.
Preprint. Expert review pending. Not clinical evidence.