The Limits of Running It Forward

Everything in this book points one direction: small rules, repeated, build large forms, and finding the rule is how you find the leverage. Read quickly, that sounds like a promise about prediction — get the rule, and the form follows from it. It is not one. And the difference is worth having explicitly, because a reader expecting prediction and not getting it will conclude the framework failed, rather than that it was never offering that in the first place.

Two different things make a Global Form unpredictable, and they ask for different responses, which is why they are worth keeping apart.

The first is sensitive dependence. Run a deterministic system of the sensitive kind forward from two starting states differing by an amount too small to measure, and the two runs do not stay near each other. They diverge, at a rate that compounds, until they have nothing to do with one another. Nothing random was introduced. The rule was the same rule both times. The only difference was in a decimal place nobody could have read.

This is why weather forecasting has a horizon rather than an accuracy problem. Better instruments push the horizon further out; they do not remove it, because the precision required to predict further ahead grows faster than any instrument can supply. Which gives a blunt practical rule: where the limit is sensitive dependence, measuring harder buys a little and never enough. It is not a resourcing problem that a better budget would solve. It is the shape of the thing.

The second is that you never had the whole rule set — and this one the framework already owns, because it falls directly out of the Nest.

A Global Form at one scale is a Local Rule at the next. So any boundary you draw around a system is a convenience rather than a fact: the thing you drew around is being entered, continuously, by rules running at scales you did not include. You did not fail to finish the list. There is no list.

Mathematics has an exact version of this, and it is worth walking because the exact version says something the intuition misses. Measure a coastline with a ruler a mile long and you get a number. Measure the same coastline with a ruler a foot long and the number is larger — not slightly, substantially, because a foot-long ruler follows inlets that a mile-long ruler stepped straight across. Go down to an inch and it grows again. The measurement does not converge on a true length. It keeps increasing across every scale anyone can measure, and without bound in the mathematical curve, because structure is present at every resolution you bring to it.

There is a number for how fast that growth happens, and it is the honest form of “how much structure is packed in here.” A line has dimension one. A filled plane has dimension two. A coastline sits between them — the west coast of Britain near 1.25 — because it is more than a line and less than a surface. The dimension is not a whole number, and that is the entire content of the idea: it measures how thoroughly a pattern fills the space it is embedded in.

Turned from a coastline toward a situation, this says something sharper than “it’s complicated.” Looking closer does not reduce your uncertainty by finally revealing the last few rules. Looking closer reveals more rule. Enumeration does not fail because you ran out of patience. It fails because there is no bottom to enumerate down to.

So the response here is different from the first case: where the limit is unenumerable rules, the move is to notice that the boundary you drew was never real, and to stop treating whatever crosses it as interference. The thing you called an outside factor was always part of the system. You just did not draw it in.

And the limit runs the other way too, which matters just as much for practice. From a form you can work backward to a rule that would produce it — but rarely to only one. A sorted neighborhood can come from mild preferences on every street, or from a rule somebody wrote about who was allowed to borrow, and the finished pattern looks the same from above. So a rule you have named is a hypothesis. Where several rules fit, the work is to look for evidence that chooses among them: change the one you suspect, and see whether the form moves. A name that has never been tested that way is a guess that happens to fit.

Which brings up the charge this chapter exists to answer.

A reader this far in may reasonably suspect the book of determinism, and to a point the suspicion is correct. Fractalism does claim that things have causes, that causes are findable, and that a form nobody chose was nonetheless produced by rules that can be named. That is a commitment to mechanism, and it is a real one, held on purpose.

But mechanism is a smaller claim than determinism, and nothing here needs the larger one. Determinism says the future is fixed by the present. Mechanism says only that what happens has findable causes — which can hold whether or not the world is deterministic underneath, and which is the only part any practice in this book rests on. If the universe is deterministic, your Local Rule is one of the things doing the determining. If it is not, your Local Rule is one of the things shaping the distribution. Either way you are an input; either way the practice is identical. The metaphysics does not reach the practice, and that is the honest reason this book takes no position on it rather than a diplomatic one.

What the book does have is an answer to the thing people are usually worried about underneath the word, which is fatalism rather than determinism.

And fatalism here is not false. It is unavailable, which is the stronger result.

“It was always going to happen this way, whatever you did” is a reading of the whole, taken from outside the whole. Part Five’s last figure holds that there is no such position: the recursion chain terminates, there is no scale above the terminal one, and therefore no vantage from which the totality could be read. Not a vantage that is hard to reach. One that does not exist.

That makes fatalism a claim which can never be checked, never be operated on, and never make contact with anything you might do differently. This book has a name for a claim held that way, as the frame instead of a finding. It is a paradigm — and the top of the leverage ladder says precisely what to do with one. Not stand on it. Pick it up, and look at whether it reaches anything.

There is a structural version of the same point, and it is the reason to hold the vantage as absent rather than merely out of reach. A system trying to hold a complete, running description of itself does not get to finish it from inside — the description would have to include itself describing, and then that, and it does not close. This is a regress rather than a proof, and it is worth holding as one: nothing here has been demonstrated, only followed far enough to see that it does not terminate. Douglas Hofstadter’s strange loop is the near relative — a climb through levels that lands back where it started — and it is what any self-referential thing runs into when it tries to get outside its own account. You can add another level, and another, and the outside never arrives, because you carried the inside up with you. A mind modelling the universe is inside the universe, and whatever it produces is part of the thing it is a model of. That is Perspective, from Part One, followed to its end: every view is taken from inside.

What this leaves you, which is more than it sounds like.

Every rung of the leverage ladder tells you where to push. Not one of them tells you what comes out: you change a number, a flow of information, a rule, a goal, the paradigm — and then you find out. The Global Form is not available as a direct object of action. It is available as a direction — steering by a star, not aiming a gun — and the entire argument for working the rule instead of the outcome is that the outcome was never a thing you could get your hands on.

The same split holds when patterns combine. How two loops are joined can be read: which part sits inside the most patterns and so moves the most when it moves, what is connected that nobody mentioned, what else stops if one part stops. Which way the combined pattern runs cannot be read off its parts, because it behaves like whichever loop is currently stronger, and that shifts. A reinforcing loop meeting a balancing one goes with the stronger; two balancing loops settle somewhere between their goals; two reinforcing loops compound unless they brake each other. You can reason your way to where to push. For which way a larger pattern is pointing, you have to look.

Which makes unpredictability an argument for the practice rather than against it. If forms were forecastable, the sensible method would be to forecast and then act on the forecast, and most of this book would be unnecessary machinery bolted onto a weather report. They are not, so the available move is the other one: change what you are repeatedly exposed to, and let the distribution of what reaches you shift. You do not need to know which specific thing is coming. You need to be standing where what arrives is survivable, and occasionally very good.

And this is what a named pattern actually buys, since it is not prediction. A Standing Third does not tell you what happens next week. It tells you what kind of thing keeps happening — which is the difference between weather and climate, and the two get collapsed constantly. Nobody can tell you the temperature in a given city on a Tuesday fourteen months out. Almost anywhere, anybody can tell you which of July and January will be hotter. The first is a trajectory and it is not available at any price. The second is the shape the system keeps returning to, and it is available, reliable, and entirely enough to plan a life around.

Try this: Take one thing you have been trying to work out the outcome of, and ask which you have been attempting — a trajectory or a climate. If it is a trajectory, notice how much of your effort has gone into getting more information, and ask honestly whether more information has ever narrowed it. Then ask the other question instead: what kind of thing keeps happening here, and what would I change if I stopped waiting to find out which particular one comes next?