A mind doesn’t only chase what it wants. It pays to resolve what it doesn’t know — because information is an investment that lowers every future bill. This is the second half of how a mind chooses, and the half that completes the engine of behaviour: expected free energy = risk + ambiguity.
Five doors; the plain line runs straight through — take whichever fit: ▶ Picture it, ✎ Try it (commit before the reveal), ✓ Check it yourself, ▸ Go deeper, ⬡ See it in UNI. Runs locally — no account, no network — adapts to light and dark; honours reduced motion.
The one fence that never moves — and in Part II it is stricter. “Expected free energy,” “ambiguity,” “information gain,” “boredom” are functional quantities: a model’s expected reduction in its own uncertainty, and the cost when that reduction is starved. For you we use felt words — curiosity, the itch to find out. For UNI we use functional ones. Neither is curiosity or desire. Part I’s free energy was proven and sealed (Gate F0), but Part II’s machinery — expected free energy, risk plus ambiguity, the G term — is not yet built or sealed in the organism; it needs its own lab-team verdict. So the math here is standard active inference you can reason through by hand, but every claim about the organism is “built to,” never “does” — held open until its gate proves it.
No peeking at Module 10. A cook tastes a sauce that’s not quite right and reaches — without trying either yet — for one of two jars.
Reconstruct two things from Module 10 before you check: how the cook ranks the jars without tasting, and the name of the thing being scored. Commit in your own words first.
If that came back — imagine, then score by risk — here is the half it leaves out.
Picture two options that are dead equal on everything Module 10 can see — your model says they’re equally likely to warm you, and equally little effort to reach. By the rule “score by risk (and cost),” they’re a tie. You’d flip a coin.
You’re cold in an unfamiliar cabin. Two options, truly tied on risk and effort: the blanket you can already see, and the knob on a strange stove. Same expected warmth, same reach. But they are not the same, and you feel it. Turn the stove-knob and — warm or not — you find something out: does this cabin have working heat? The blanket tells you nothing new. The stove resolves an uncertainty the blanket leaves sitting there.
Blanket vs a strange stove-knob, pinned EQUAL on risk and effort — so a mind scoring by outcome alone must flip a coin. Pick one, then watch what separates them, and what it’s worth over many cold nights. A reveal of a mechanism (information lowers future cost), not a plotted curve.
Turning the knob costs energy now, and its immediate payoff is uncertain. But if it resolves “does this cabin have heat?”, then every future cold moment gets cheaper — you’ll never again have to guess. You spend a little now to save a lot later: an investment, borrowing energy from the future against the returns the prediction yields across many cycles.
That is epistemic value — the worth of finding out — and it is not a metaphor. Information is worth exactly the future free energy it saves you. A mind pays to learn because learning lowers the bill forever after. So the full score of an action has two halves, and you have now built both:
Expected free energy G = risk + ambiguity.
— risk (Module 10): will the future land where I want? — the pragmatic half.
— ambiguity (this module): will it resolve what I don’t know? — the epistemic half.
A mind minimises G — it picks the action whose imagined future best gets what it wants and tells it what it needs to know.
Two “equally warming” options usually aren’t equal once you look further out. A blanket warms — but it tethers you, pins your mobility; and sit under it long enough and you’d get bored. So its future violates other things you’re built to want — to move, to be stimulated — even while it satisfies “warm.” You don’t score an action against one preference. You score it against your whole stack of priors, rolled over the whole future: warm, mobile, fed, stimulated, safe — all of it, weighed together.
And notice what boredom turns out to be: the felt alarm of an unmet epistemic prior — understimulation is its own kind of surprise. The drive to find out isn’t a luxury a mind adds on top; it’s woven into the same stack that keeps it warm and whole.
The engine of choice is complete: a mind scores each action by the expected free energy of its imagined future — risk (get what I want) plus ambiguity (find out what I don’t) — against its whole stack of priors, and picks the lowest. Explore and exploit, want and wonder, all in one number.
And a question has been quietly answered that we never stopped to notice: you have been computing free energy twice over — once in Part I, over beliefs about now, and now over actions into the future. Are those two different things, or the same thing wearing two faces? That’s Module 12, and the answer unifies the whole book.