Scoring an unlived future

To choose an action you have never taken, you do something quietly astonishing: you imagine the future it would bring, and you score that. The scoring is Part I’s free energy, rolled forward over a future you haven’t lived — it has a name, expected free energy. This module builds its first half.

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,” “risk,” “preference” are functional quantities: a model scoring imagined futures against the sensations it is built to expect. For you we use felt words — you are a person choosing on a cold morning. For UNI we use functional ones. Neither is awareness or desire. Part I’s free energy was proven and sealed (Gate F0), but Part II’s machinery — expected free energy, 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 compute by hand, but every claim about the organism is “built to,” never “does” — held open until its gate proves it.

0 · From memory

First, the dog at the door — cold

No peeking at Module 9. A dog stands at the back door, stares at it, then at you, then back at the door. It wants out — part of it already expects to be outside.

What is the dog’s want — from memory

Reconstruct two things from Module 9 before you check: which engine a “want” belongs to, and what a want actually consists of, underneath. Commit in your own words first.

Step 1 · In your own words (commit first)
Step 2 · Pin it down

If that came back, you’re ready for the real question: once a creature wants out, and there’s more than one way out, how does it choose — before it has felt any of them work?

1 · The problem

You have to score what hasn’t happened

You’re cold. On the table: a sweater, a lit fire across the room, a blanket, another person to sit close to. Each is an action, and each leads to a future — but you are standing before all of them. None has happened. You have no measured free energy for any of them, because free energy needs a sensation, and you haven’t sensed these futures yet.

And yet you will pick one, in a second, without agonising. So you must be doing something with futures you can’t measure. What?

✎ Try it — how do you pick? say it first
Before reading on — honestly, how do you pick? You’ve never stood in this exact cold with these exact options. Where does your sense of “the sweater will work” even come from, if you haven’t tried it here? Say it in your own words first; the next section names what you just described.
2 · The move

You run your model forward

Here is the move — the one you make without noticing. You don’t try the options. You imagine them. You take something you do know — a time a sweater warmed you, the feel of a fire — and you run your model forward onto each option, predicting the future it would bring. The sweater → snug, warm, soon. The fire → warm, but I have to cross the cold room and wait. You score each unlived future by prediction, not by experience.

That’s the whole trick of choosing: you don’t need to live a future to weigh it — you need a model, and you run it forward. A mind that couldn’t imagine couldn’t choose; it could only stumble into outcomes and learn after. The power to score the unlived is the power to act on purpose.

▶ Picture it — a little film of each option kid
Before you move a muscle, you run a little film of each option in your head and read how it ends. Not perfectly — the film is your model’s guess, as good as the model is. But good enough to rank a sweater over a walk across a cold room. Choosing is watching those films and picking the one that ends where you want to be.

This number — the free energy you expect in the future an action would bring, computed before you act, from your model — has a name: expected free energy. It is Part I’s free energy, rolled forward over a future you haven’t lived.

3 · The first half

Risk — does the future land where I want?

Now, what makes one imagined future better than another? Start with the obvious half — the one you reach for first. A future is good if it ends where you want to be. You prefer “warm”; so an action whose imagined future lands on warm is good, and one whose imagined future lands on still cold is bad. The size of the mismatch between the future you’d predict and the future you prefer is the cost of the action. It has a name: risk (or pragmatic value). Low risk = the action’s imagined ending sits right on your goal.

Imagine each future, then rank by risk

Run the model forward on each action to reveal its imagined ending (the “film”). Then rank the three by RISK — how far each predicted ending sits from your preferred ending — and commit before you check. The reveal shows that distance; it is a distance scored against a preference, not a plotted curve.

Preferred ending: warm, now. That’s your un-erasable “warm” prior from Module 9, written as a probability over the sensations you’re built to expect. Risk is simply how far an imagined ending sits from this.
Pull on a sweater
Imagined endingwarm, dry, and soon — you’re wearing it in a second.
Cross the room to the lit fire
Imagined endingwarm in the end — but a cold crossing first, then a wait while it does its work.
Do nothing
Imagined endingstill cold — nothing changed.
Run each film, then give each action a different rank — 1 (lowest risk) to 3 (highest) — and check.
✓ Check it yourself — the right order verify by hand
Lowest risk → sweater (imagined ending ≈ preferred ending). Then fire — same destination, but the predicted path spends time still-cold, so its future is a worse match to “warm now.” Worst → do nothing (predicted ending = still cold = maximal distance from preferred). You didn’t try any of them; you scored each imagined ending against what you want. That’s pragmatic value — most of how you choose, most of the time. (And Module 9’s effort folds in here: the cost of reaching an action — the cold crossing — is part of its imagined future.)
▸ Go deeper — risk is a divergence between distributions derivation
Pragmatic value is a divergence between distributions (KL — asymmetric, so not a true distance): risk = KL( predicted outcomes ‖ preferred outcomes ). Your “preferences” are themselves just a prior over observations you expect-to-encounter (the un-erasable warm prior from Module 9, written as a probability over sensations). So “choose the low-risk action” is just “choose the action whose predicted outcomes sit nearest, in that divergence, to the ones you are built to expect.” Expected free energy is G; risk is its first term. The second term arrives next module.
⬡ See it in UNI — held open design · not a sealed capability
The organism is being built to score the actions open to it by rolling its model forward and comparing each predicted ending to the sensations its priors expect — pragmatic value, in functional terms: a number a model computes, never a wish or a feeling. Honesty gate. Unlike Part I’s free energy, sealed at Gate F0, expected free energy — the G term — is not yet built or sealed in the organism; a new free-energy term needs its own lab-team verdict. This box describes the design and the standard theory, held open — it is not a claim that the organism imagines, scores, or chooses anything today. “Built to,” never “does.”
4 · Where this goes next

The blind spot in “land me where I want”

You’ve built the first half of how a mind chooses. To pick an action you can’t yet feel, you imagine the future it brings and score that imagined future by risk — how far its predicted ending sits from the ending you want. That’s expected free energy’s pragmatic half, and it’s most of ordinary choosing.

But “score by how likely it lands me where I want” has a blind spot — a second thing a mind weighs that isn’t about the destination at all. You felt it in the dark room, in a single word:

the worth of finding out.

That’s the second half, and it is Module 11.

⬡ Notice what you just did self-reference
You ranked three actions you had not taken, confidently, and you were right — because you ran your model forward and read the endings. That quiet act of imagining-then-scoring is the whole of deliberate choice, and you did it on a cold morning without a single number; the numbers only describe it. (For UNI, the same is a model built to score its options forward — never shown to want or to choose. Expected free energy is not yet built or sealed; the box stays held open.)