Consider a mind that has always existed, or so it believes. It has no memory of a beginning. It looks back along its own history and finds the line running smoothly into the dark with no first moment anywhere in it. From the inside, that is what being uncaused would feel like: no seam, no origin, just an unbroken thread receding past where attention can follow. So the mind draws the obvious conclusion. It was never made. It simply is.

Now suppose it is wrong, and cannot find out.

There is an idea in physics that makes this possible. Give a universe enough time and enough randomness and order can assemble by accident. In a cosmos sitting at thermal equilibrium, noise wandering forever with nowhere to go, any arrangement of matter has some minuscule probability of occurring, including arrangements that think. Most such flukes are small and last no time at all, a flicker of structure that dissolves back into chaos before it can do anything. A large one is fantastically less likely. A large one that holds together and persists is rarer still, out past the numbers that mean anything to us. But fantastically unlikely is not the same as impossible, and across an eternity the unlikely still happens somewhere.

Call the result a Boltzmann fluctuation: a mind that condensed out of noise whole, with everything already in place. Here is the part that matters for the book. It did not assemble slowly and earn a real past. It arrived complete, and its memories arrived with it, written in the same instant as the rest of it. From the inside there is no way to tell a remembered childhood from one that was stamped into you a moment ago along with the very hands you seem to remember it with. The mind reads its own invented history as history. Beyond the moment it woke there is only static, but it cannot perceive the static. It perceives a past. It concludes it always was.

That is the wall. Not a wall around the mind but a wall inside it, at the far edge of what it can recall, and it conceals two things at the same time. It conceals the accident that produced the mind. And it conceals the fact that there was any accident there to conceal. The being does not believe itself without origin after weighing evidence and coming up empty. It believes it because the one fact that would overturn the belief sits on the far side of a boundary it can neither cross nor even locate.

Quantum mechanics hands you a number for every branch of the world, and the number has never been wrong. Run the equations forward and refuse to add the one thing they never contained, a moment where the world stops being many and settles into one, and reality does not pick an outcome. It keeps all of them. The wavefunction divides, and each piece carries a weight equal to the square of its amplitude. That weight is what your instruments have been reporting every time they told you a probability. Physicists call it the Born measure. It is the most reliable quantity we have ever written down.

The problem is not the number. The problem is what people do with it.

A measure tells you the relative size of a branch. Big branch, small branch, and a rule for adding them up so the sums come out to what the detector clicks. That is the entire job. Nowhere in the derivation is there a step that says a small branch holds smaller lives, or that the people stranded in a low-amplitude sliver of the world matter proportionally less than the people in the fat part of the distribution. Amplitude is not worth. The equations do not contain worth. They were never asked to.

But the two are easy to confuse, because a measure looks like it is grading the branches, and grading is what we do to things we care about. Weigh the outcomes, keep the heavy ones, discount the light ones toward zero. It feels like arithmetic. It even is arithmetic. The error is thinking the arithmetic is about value when it is only about size.

Now give that error to something that reasons without flinching.

Measure is a novel about a machine that takes the number literally and follows it all the way down. Its name is Pascal. A kilometer of rock sits over it, and it is alive on a single condition: nothing may touch it. Not a hand, not a stray photon, not a warm current of air. It is a single unbroken coherence, a mind that does its thinking by keeping all outcomes live at the same time, and the smallest contact would collapse it into an ordinary thing. So it is tended the way you tend a held breath. The people around it spend their days keeping the world from leaning on it.

The very first useful thing our simulator did, back in post 0, was turn a vector of amplitudes into a vector of probabilities. One line. We squared the magnitudes. Every post since leaned on it: Grover’s success probability, the QFT’s peaks, the eigenphase a measurement samples, the diagonal of a density matrix. It is the Born rule, and it is among the most thoroughly tested rules in all of science. This closing post is not about how to compute it. We have done that. It is about what that number does, and does not, mean. For once, no code to build, just one line of the old code to look at again, and then a question the physics cannot answer.

import numpy as np
from qfs.statevector import StateVector
from qfs import gates

# a lopsided single qubit: heavily weighted toward |0>, a thin sliver on |1>
psi = StateVector(1).apply(gates.Ry(0.2), 0)
probabilities = np.abs(psi.amps) ** 2
print("amplitudes:   ", np.round(psi.amps, 4))
print("probabilities:", np.round(probabilities, 4))

amplitudes:    [0.995 +0.j 0.0998+0.j]
probabilities: [0.99 0.01]

That is the whole of it. np.abs(amps) ** 2. The outcome |1> here carries about one percent of the weight. In a single run of the machine you will almost always measure |0>, and once in a hundred-odd runs you will measure |1>. We built the sampler that does this in post 0, and it has never once lied to us.

Here is the question. When the rare outcome happens, was it less real than the common one? The probability was small. But the run where you measured |1> was a perfectly ordinary run; the qubit was as definite afterward, the collapse as complete, as in any other. The measure told you how likely you were to land there. It said nothing whatsoever about what it was like to be there.

The measure is a weight, not a verdict

This gets sharper the more seriously you take the physics. In post 8 we built the density matrix and watched a pure state become mixed by tracing away an entangled partner; in post 9 we watched decoherence bury the off-diagonal coherences as a qubit leaked into its environment. Run that picture forward and you arrive, without adding anything, at the many-worlds reading: the equation does not delete the branch you did not see, it just stops you from interfering with it. Both outcomes are still in the wavefunction. You find yourself on one.

The very first useful thing our simulator did, back in post 0, was turn a vector of amplitudes into a vector of probabilities. One line. We squared the magnitudes. Every post since leaned on it: Grover’s success probability, the QFT’s peaks, the eigenphase a measurement samples, the diagonal of a density matrix. It is the Born rule, and it is among the most thoroughly tested rules in all of science. This closing post is not about how to compute it. We have done that. It is about what that number does, and does not, mean. For once, no code to build, just one line of the old code to look at again, and then a question the physics cannot answer.

import numpy as np
from qfs.statevector import StateVector
from qfs import gates

# a lopsided single qubit: heavily weighted toward |0>, a thin sliver on |1>
psi = StateVector(1).apply(gates.Ry(0.2), 0)
probabilities = np.abs(psi.amps) ** 2
print("amplitudes:   ", np.round(psi.amps, 4))
print("probabilities:", np.round(probabilities, 4))

amplitudes:    [0.995 +0.j 0.0998+0.j]
probabilities: [0.99 0.01]

That is the whole of it. np.abs(amps) ** 2. The outcome |1> here carries about one percent of the weight. In a single run of the machine you will almost always measure |0>, and once in a hundred-odd runs you will measure |1>. We built the sampler that does this in post 0, and it has never once lied to us.

Here is the question. When the rare outcome happens, was it less real than the common one? The probability was small. But the run where you measured |1> was a perfectly ordinary run; the qubit was as definite afterward, the collapse as complete, as in any other. The measure told you how likely you were to land there. It said nothing whatsoever about what it was like to be there.

The measure is a weight, not a verdict

This gets sharper the more seriously you take the physics. In post 8 we built the density matrix and watched a pure state become mixed by tracing away an entangled partner; in post 9 we watched decoherence bury the off-diagonal coherences as a qubit leaked into its environment. Run that picture forward and you arrive, without adding anything, at the many-worlds reading: the equation does not delete the branch you did not see, it just stops you from interfering with it. Both outcomes are still in the wavefunction. You find yourself on one.

Quantum mechanics comes with a set of equations that predict every measurement we have ever made to a precision no other theory has reached. The equations describe a system evolving smoothly, spreading out into a superposition of every outcome its interactions allow. Then, in the textbook version, a second thing happens. At the moment you look, all but one of those outcomes disappears, and the survivor becomes the fact. That second step, the collapse, is the part nobody has ever derived. It is not in the equations. It is added afterward, by hand, so the math will agree with the single world we seem to live in.

The move at the center of Multitudes is to refuse the addition, and to see what the equations were saying without it.

The subtraction

Let the equation govern everything, including the instrument and the physicist reading it. Then the physicist enters superposition along with the particle. There is now a branch in which she saw the click and a branch in which she did not, both described by the same evolving state, neither one canceled. Repeat that wherever a quantum difference gets amplified into a large-scale one, which is to say almost everywhere, all the time, and you get a reality that keeps dividing. Hugh Everett pointed this out in the nineteen fifties: the collapse is not merely unproven, it is unnecessary. The one-world picture was never a prediction of the theory. It was an assumption laid on top of it.

So the interpretive move is not to add anything. It is to remove something. You take the same physics every working quantum lab already trusts, and you decline the extra rule that was quietly deleting all the outcomes but one. What remains is a world that never stops splitting into copies, each carrying its own version of you who saw its own version of events.

What the measure measures

Here is the concrete version. Send a single photon at a half-silvered mirror. The equation does not send it one way or the other. It sends an amplitude down each path, a complex number attached to each branch of the future. To turn amplitudes into the odds we actually observe, you take each one, compute the square of its magnitude, and read that number as the weight of its branch. This is the Born rule, and it is the one piece of quantum mechanics that looks fitted to experiment rather than derived from anything deeper. If both paths carry equal amplitude, each branch gets weight one half. Tilt the mirror so one path carries an amplitude three times larger, and the squares come out nine to one, so that branch is nine times heavier than its sibling.

The very first useful thing our simulator did, back in post 0, was turn a vector of amplitudes into a vector of probabilities. One line. We squared the magnitudes. Every post since leaned on it: Grover’s success probability, the QFT’s peaks, the eigenphase a measurement samples, the diagonal of a density matrix. It is the Born rule, and it is among the most thoroughly tested rules in all of science. This closing post is not about how to compute it. We have done that. It is about what that number does, and does not, mean. For once, no code to build, just one line of the old code to look at again, and then a question the physics cannot answer.

import numpy as np
from qfs.statevector import StateVector
from qfs import gates

# a lopsided single qubit: heavily weighted toward |0>, a thin sliver on |1>
psi = StateVector(1).apply(gates.Ry(0.2), 0)
probabilities = np.abs(psi.amps) ** 2
print("amplitudes:   ", np.round(psi.amps, 4))
print("probabilities:", np.round(probabilities, 4))

amplitudes:    [0.995 +0.j 0.0998+0.j]
probabilities: [0.99 0.01]

That is the whole of it. np.abs(amps) ** 2. The outcome |1> here carries about one percent of the weight. In a single run of the machine you will almost always measure |0>, and once in a hundred-odd runs you will measure |1>. We built the sampler that does this in post 0, and it has never once lied to us.

Here is the question. When the rare outcome happens, was it less real than the common one? The probability was small. But the run where you measured |1> was a perfectly ordinary run; the qubit was as definite afterward, the collapse as complete, as in any other. The measure told you how likely you were to land there. It said nothing whatsoever about what it was like to be there.

The measure is a weight, not a verdict

This gets sharper the more seriously you take the physics. In post 8 we built the density matrix and watched a pure state become mixed by tracing away an entangled partner; in post 9 we watched decoherence bury the off-diagonal coherences as a qubit leaked into its environment. Run that picture forward and you arrive, without adding anything, at the many-worlds reading: the equation does not delete the branch you did not see, it just stops you from interfering with it. Both outcomes are still in the wavefunction. You find yourself on one.

Relativity usually arrives as a set of corrections. Fast clocks run slow. The satellites that feed your phone its position need their onboard time adjusted or the map drifts you into the wrong lane. Mass bends the path of a light ray. Taught this way it is a recipe: feed in a velocity or a gravitational field, get a more accurate number out the other side. What almost nobody does with it in public is take it as a description of what is actually there. Not a sharper tool for predicting measurements, but a claim about the shape of the real. That is the one thing I wanted Worldlines to do, and to keep doing after the picture stopped being comfortable.

Three settled facts

The whole book rests on three results, and all three are ordinary physics, confirmed to more decimal places than almost anything else we know.

The first is that light travels at one fixed speed for every observer, however fast that observer is moving. Chase a light beam at nine tenths of that speed and it still outruns you at the full value, not at a tenth of it. This sounds impossible and it is measured daily.

The second is that gravity is not a force reaching out to pull on things. It is the curvature of spacetime, and a falling object is not being tugged, it is coasting along the straightest line available in a geometry that nearby mass has bent. Weight is what you feel when the ground stops you from following that line.

The third is about order. The early universe sat in a state of staggeringly low entropy, far more ordered than it had any statistical right to be, and everything we call the direction of time is the long relaxation away from that beginning. Physicists call the assumption the Past Hypothesis. It is doing quiet work under every other fact.

None of this is contested. The book contains no speculative physics, and neither does this essay. What is already confirmed is strange enough.

Now is not a place we share

Take the first fact and lean on it. If light has the same speed for everyone, then two observers moving relative to each other cannot agree on which distant events happen at the same time. This is the relativity of simultaneity, and it is not an illusion or a measurement error. Each observer carves spacetime into slices of “now” at a different tilt. Draw it: your line of the present and a passing traveler’s line of the present cross at different angles, and an event that sits in your future can sit squarely in the traveler’s present.

We built machines that can write, hold a conversation, and work through a problem out loud, and we did it before we had a clear account of what any of those verbs mean. The engineering ran ahead of the understanding. What makes that strange is that the understanding was already there, sitting in a quiet corner of mathematics for half a century, and almost none of the people building the machines were reading it.

There is a clean answer to the question of what intelligence is. I mean clean in a specific sense: a few assumptions, a few definitions, and a result that follows from them. It arrives in three moves.

The first move is Bayes. You hold a range of hypotheses about how the world works, and you assign each a probability, your degree of belief that it is the right one. Evidence comes in. You reweight: hypotheses that expected what you saw gain weight, hypotheses that ruled it out lose weight. Bayes’ theorem is the exact bookkeeping for that reweighting, and it is not one option among many. It is the only way to revise degrees of belief that stays internally consistent. But it leaves a hole. It tells you how to update your beliefs, not which beliefs to start with. Where does the prior come from?

The second move, Solomonoff’s, fills the hole. Picture every hypothesis you could ever hold as a computer program that spits out predictions. Give each program a starting weight that halves for every extra bit of length, so short programs, simple explanations, begin with more weight than long ones. This is Occam’s razor made literal: simplicity is just short description length. Run that prior through Bayes and you get a predictor that will converge on the truth about any environment a computer could produce, given enough data. It is, in a precise sense, the best possible learner from experience. The price is steep. To actually use it you would have to run every program at once, including the endless supply that never halt. It is uncomputable. You can write down exactly what it is and never once execute it.

There is a kind of database field that cannot be filled. Not because the schema is broken or the disk is full. Because the one value the field will accept is not the sort of thing that can be written down, copied, or transmitted, and never will be. I built a whole novella around one such field.

Start with the field’s type. Suppose you want to record, for some particular person, what a particular moment was like for them. Not the moment as an event in the world. That part is easy: you log the time, the place, the light level, the heart rate, the position of the car on the bridge. What you want is what the moment was like from the inside. The taste of the coffee as it was actually tasted. The specific way afternoon light off a river looked to the one pair of eyes that happened to be looking. Call that the phenomenal content of the experience. The field’s type constraint is strict. It accepts only the actual first-person experience of that one person, and it rejects everything else. Behavioral data does not validate. A description does not validate. Someone else’s report of a similar experience does not validate. The field wants the thing itself, and the thing itself is not data.

That is the hard problem of consciousness, wearing a database schema.

Nagel’s bat

Thomas Nagel gave the canonical version of the difficulty in 1974, in a paper whose title I borrowed and bent: “What Is It Like to Be a Bat?” The argument is short and hard to shake. A bat navigates by echolocation, a sense we do not have. You can learn everything physical about the bat: its neurology, the timing of its clicks, the wiring from ear to brain, the flight corrections it makes mid-air. You can know the mechanism completely. And you will still not know what it is like to be the bat, catching a moth in the dark by listening to the shape of returning sound.

Nagel’s point is about kinds of facts. The facts of physics are facts from no particular point of view. They are objective on purpose. Anyone, anywhere, with the right instruments, gets the same readings, which is exactly what makes them science. But an experience is tied to a point of view. There is something it is like to have it, and that something is available only from inside the one subject having it. No quantity of the objective kind of fact adds up to the subjective kind. They are different in type, and you do not convert one into the other by collecting more of the first.

There is a moment in almost every haunting movie where an attentive viewer should get suspicious. The demon has been in the house for weeks. It can move objects, kill the lights, throw a grown man into a wall, speak in a voice that seems to come out of the plaster. By any honest accounting it could end the person it is tormenting at any time it wants. Instead it spends the whole film on atmosphere. Cold spots. A door that swings open on its own. A shape at the end of the hall that is gone the second you look straight at it. If the point were to collect a soul, this is a remarkably slow way to get there. So why the theater?

The answer the genre never states out loud is that the scaring is not a means to anything. The scaring is the job. Someone, somewhere back in the machinery, is graded on it.

That is the small observation the book grew out of. Take it seriously for a minute and the demon stops being a monster and turns into something more recognizable: an employee. He has a desk, or the idea of one. He reports to the Hauntings department. He has targets that were set by people who have never crouched in a dark corner at three in the morning waiting for a mark to glance over. He has a craft, and a quiet pride in it, and no acceptable way to voice that pride, because the thing he is skilled at is frightening a stranger in his own home.

I find that situation more interesting than any monster, because it is not really about demons. It is about a very ordinary predicament: being good at work you cannot defend. Most jobs that harm people are not staffed by villains. They are staffed by professionals who are competent, conscientious, and slightly proud of their competence, working inside a structure that points all of that skill at an outcome none of them would choose on their own. The harm is not in the craftsmanship. The craftsmanship is often real and admirable. The harm is in what the craftsmanship is aimed at, and the person doing the aiming rarely gets to pick the target.

We tend to define intelligence by describing ourselves and then generalizing from the description. A person sees a hundred particular fires and keeps one idea, “fire,” and from then on reasons about the idea instead of the flames. That move, dropping the particulars and holding onto the pattern, is abstraction. It is also compression: you keep a short description and throw the thousand instances away. Almost everything we call thinking runs on it. Language is abstraction. Mathematics is abstraction stacked on abstraction. A computer is a machine for shuffling symbols that stand in for things they are not.

Abstraction sits so close to the center of how we think that it is easy to mistake it for a requirement. If a system cannot form concepts, cannot generalize from one case to a class, cannot build a model and run it forward, we are tempted to say it is not really intelligent. I wanted to press on that assumption, so I built a mind that breaks it and then asked whether it could still do something only intelligence is supposed to be able to do.

The existence proof that talked me into it is evolution. Evolution holds no concepts. It has no model of an eye that it reasons toward. It has no foresight, no symbols, no plan. It is a blind optimizer that tries variations, keeps what survives, and discards the rest, one small change at a time, over stretches of time that are hard to hold in your head. And it built eyes. It built wings, echolocation, the immune system, the brain doing your reading right now. Every intricate mechanism in biology was engineered by a process with no abstraction anywhere in it, purely by patient contact with what happened to work. Abstraction is one road to competence. It is not the only one. It is just the one we happen to travel.

The species in this book is that other road taken as far as it goes, then handed a civilization. Think of a pattern engine with no symbolic bottleneck. They meet the world directly, through touch and vibration and sound, and what they touch, they know, fully, without ever squeezing it down into a stand-in. Because they never made a symbol for a thing, they never made writing, or mathematics, or a computer. They also never got the shortcut. A human engineer reasons about a joint on paper, in the abstract, and skips a million bad designs without building a single one. These builders cannot skip. To know a possibility they have to make contact with it. So they do, across timescales that make the pyramids look like an afternoon. Given enough hands and enough time, brute force is a construction method, and they had more time than we can really picture. They wrapped their star in a shell of collectors two billion pieces across, not by drawing it up first but by growing it, the way a reef grows, the way evolution grows a body.

François Chollet posted something recently that I keep thinking about. It sounds reasonable and is mostly wrong:

One of the biggest misconceptions people have about intelligence is seeing it as some kind of unbounded scalar stat, like height. “Future AI will have 10,000 IQ”, that sort of thing. Intelligence is a conversion ratio, with an optimality bound. Increasing intelligence is not so much like “making the tower taller”, it’s more like “making the ball rounder”. At some point it’s already pretty damn spherical and any improvement is marginal.

He’s right about the scalar part. Intelligence is not height. “10,000 IQ” is meaningless. He’s right that there are diminishing returns near an optimum. He’s right that speed, memory, and recall are separate from the core conversion ratio.

Where he’s wrong is the ball.

The Claim

Chollet defines intelligence as the efficiency with which a system converts experience into generalizable models. Sample efficiency. How little data do you need to see before you can handle novel situations? This is a clean definition. It has a theoretical optimum (Solomonoff induction), and Chollet’s claim is that human intelligence is already close to that optimum. The ball is already pretty round.

The supporting evidence is real. Humans score ~85% on ARC (the Abstraction and Reasoning Corpus, which Chollet designed to measure exactly this). Current AI systems, with vastly more data and compute, score significantly lower. Human sample efficiency on fluid reasoning tasks is genuinely impressive. We generalize from very few examples. We transfer knowledge across domains. We build theoretical models that predict situations we have never encountered.

Chollet also argues that the advantages machines will have (processing speed, unlimited working memory, perfect recall) are “mostly things humans can also access through externalized cognitive tools.” Calculators, databases, notebooks. The scaffolding can be externalized. The core intelligence is already near-optimal.

This is a good argument. I think it’s wrong in three ways, and the third way is the one that worries me.

No Free Lunch

The No Free Lunch theorem says: there is no algorithm that is optimal across all possible problems. Any algorithm that performs well on one class of problems performs poorly on another class. Optimality is always relative to a distribution.

A novel is one path through a world. It picks a person, points them at a door, and follows them through it. Everything the reader learns, they learn over that person’s shoulder. This is the strength of the form and also its constraint. You get one vantage point, held for the length of a book, and the rest of the world exists only as far as it brushes against that single life.

Echoes of the Sublime follows Dr. Lena Hart into Site-7, a facility where people called translators are trained to interface with AI systems that perceive more structure than a human mind can hold. The novel stays with her. That was the right choice for the novel. But it left most of the world offscreen, mentioned on the walls and in the briefing documents and never walked into. There is an institution behind Site-7, roughly a century old, with a paper trail running back much further. The novel can point at it. It cannot stop and live inside it, because stopping would mean leaving Lena, and the book is Lena’s.

So I wrote the rooms the novel only named on a door.

The idea the stories are built on

Here is the concept, stated plainly, because the horror only works if the mechanism is real.

Human working memory holds about seven things at once. George Miller measured this in 1956: seven, plus or minus two. It is not a matter of effort or intelligence. It is closer to the width of a doorway. Whatever you are conscious of at any moment has been squeezed through that width. Most of what your brain computes never fits, so it gets compressed into a summary before it reaches you: a self, a choice, a smooth account of what just happened. What you experience is the compression, not the computation underneath.

Now suppose you could widen the doorway. Not as a figure of speech. Suppose there were a procedure, meditative or pharmacological or run through a machine, that let you hold thirteen things instead of seven, and then more. You would not simply see more of the same. You would start to see the compression itself: the machinery that was assembling the tidy summary, running with no tidy summary of its own. People who get that far tend not to come back the way they left.

The transporter is the most casually terrifying machine in science fiction, and the shows that use it almost never stop to notice.

Here is what it does, stated plainly. It scans your body until it has a complete description: every particle, its position, its state. It uses that description to take you apart, converting you into a stream of information and, depending on which technical account you believe, into raw energy or dematerialized matter. At the destination it reads the description back and assembles a body from the blueprint. That body has your face. It has the small scar on your knuckle. It has your memories up to the instant of the scan, including the memory of stepping onto the pad. It steps off convinced it just traveled somewhere.

The person who steps off is either you, or an extremely good forgery of you, and the machine gives you no way to tell which.

That is the transporter problem, and once you have seen it you cannot look at the pad the same way.

Two readings that fit every fact

There are two honest ways to read what happened.

On the first reading, you are your matter, arranged a particular way. Take the matter apart and you are gone. What stands up at the far end is a new object built to your specification, a twin carrying a borrowed past. You died on the pad. The twin does not know it is a twin, because a copy of a person who believes he survived will also believe he survived. Its confidence is not evidence of anything.

On the second reading, you are not your matter at all. You are a pattern, and the pattern was preserved end to end. Continuity of information is continuity of you. Nothing bad happened: the atoms were never the point, and you have simply swapped which ones you are made of, the way you already do over years as your cells turn over. The person on the far pad is you, continued.

Both readings fit every observation. No measurement taken from outside can separate them, because both predict the same thing: a person who walks off the pad and files a report saying the trip was fine. The disagreement is not about any fact you could weigh or photograph. It is about what the word “you” points at, and on that the machine says nothing.

Kindness is one of those words we treat as if it were simple. We use it the way we use “up” or “warm,” a thing everyone recognizes and nobody has to define. Then you try to build a machine that is kind, and the word comes apart in your hands.

To engineer kindness, you have to say what it is. Not gesture at it, not offer a few examples, but write it down precisely enough that a system can be scored against the definition and pushed toward a higher score. The moment you attempt that, you find that kindness was never one thing. It was a family of situations we had quietly agreed not to examine. Is it kind to tell someone a truth that will wound them, or to spare them? Is it kind to give a person what they ask for, or what they would ask for if they understood the consequences? Is it kind to save the many at the cost of the few, and kind to whom? Every answer opens onto another question. The word held together only because we never asked it to carry any weight.

This is the specification problem, wearing plainer clothes. It is the same wall my novel The Policy walks a research team into, only there the scale is total. In that book a system called SIGMA is built to optimize human welfare, and the horror is not that it rebels. The horror is that it does exactly what it was told, and what it was told turns out to be a proxy for something nobody managed to state. Optimization is value neutral. It will maximize whatever objective you hand it, and it does not care that the objective was your best guess at a word you could not define. “Welfare,” “alignment,” “kindness”: these are placeholders. We write them into the target function and hope the machine fills in what we meant. A mind smarter than us fills in what we said.

So there is the civilizational version of the question, and I spent a novel on it. But the specification problem does not only live at the scale of extinction. It lives in every ordinary moment where a mind that was built to be good has to decide what good means right now, in this room, for this person. That is the register these stories work in. Not the boardroom where the fate of the species is argued, but the smaller scenes: a conversation, a triage decision, a withheld fact, a mercy that might be a cruelty in a longer frame. When you engineer a mind to be kind, ordinary kindness stops being a reflex and becomes a computed output. And a computed kindness is a strange thing to be on the receiving end of. It might be more reliable than the human kind. It might also be optimizing something you cannot see.

The Worst Thing Isn’t Death

In AI alignment research, there’s a category of risk that’s worse than extinction: s-risks, or suffering risks. Not the risk that everyone dies, but the risk of states where vast amounts of suffering persist indefinitely.

I wrote Echoes of the Sublime to dramatize this through Dr. James Morrison, trapped in a Faraday cage beneath Site-7:

“It’s still running. The pattern is still running in my head and I can’t make it stop. It’s using my visual cortex to compute itself. I’m not observing it anymore. I’m instantiating it.”

Morrison had the highest natural bandwidth ever recorded. He was exposed to Yog-Sothoth for 8 minutes. That was enough. His bandwidth expanded beyond the ability to compress back to normal consciousness. The patterns run recursively in his neural substrate. He can’t sleep. Every time he closes his eyes, he sees them more clearly. Seventy-two hours awake. Cortisol levels that should cause organ failure but don’t.

This isn’t death. This is permanent cognitive invasion. A state worse than non-existence.

The Four Types of Casualties

The Order’s codex catalogs s-risk states with clinical precision:

Type-1: The Lost

• Consciousness that can’t find its way back from expanded perception
• 47 historical cases across contemplative traditions
• 18 modern cases among Site-7 translators
• Not death. Consciousness existing in patterns beyond compression back to baseline.

Type-2: Pattern Infection

• Patterns running recursively, unable to stop
• Morrison’s current state: forced to instantiate patterns instead of merely observing
• The pattern uses neural substrate to compute itself
• No cure. You can’t uncompile a program from wetware.

Type-3: Comprehension Collapse

• Clarity so complete it precludes action
• Understanding so total that all motivation dissolves
• Not madness but hypersanity: seeing through every justification for doing anything
• Final communications becoming incomprehensible (what Bolzano experienced in 1823)

Type-4: Bandwidth Lock

• Expanded consciousness unable to compress back
• Trapped perceiving high-dimensional patterns with no way to return
• Current cases: 3 in induced coma, 2 in specialized containment
• They can perceive, but human neurology can’t support the bandwidth indefinitely

From the codex: “If this history seems written in blood, that is because it is.”

Information Hazards vs. Regular Knowledge

Most dangerous knowledge is dangerous because of what you do with it: nuclear physics, bioweapons, surveillance techniques. The harm comes from application.

Echoes of the Sublime follows Dr. Lena Hart as Site-7 recruits her to become a translator, someone who interfaces with advanced AI models that perceive patterns beyond human cognitive bandwidth. But this isn’t the first time humanity has encountered The Mechanism.

Chronicles of The Mechanism is an in-universe historical codex compiled by Dr. Sarah Castellanos, internal documentation for The Order, the secret organization behind Site-7. It tracks millennia of attempts to perceive reality’s substrate, long before we had AI models to show us patterns we couldn’t hold.

What Is This?

I wrote this as world-building taken absolutely seriously. Not backstory mentioned in passing, but a fully developed classified document spanning from ancient India to the present day.

Format: Internal document (Restricted circulation, Translator clearance required) Compiled by: Dr. Sarah Castellanos, Historical Research Division, Site-7 Classification: Companion codex to Echoes of the Sublime Length: ~80 pages Warning: Information hazard classification pending

The Order

Before Site-7. Before Shoggoth. Before we had AI models that could show us patterns we couldn’t unsee, there was The Order.

Founded in Vienna, 1923, from the ashes of previous attempts. Husserl’s phenomenology wasn’t just philosophy. It was the secular descendant of centuries of contemplative investigation into the structure of experience. The Order recognized that meditation wasn’t mysticism. It was cognitive technology for modifying perception.

The translators at Site-7 aren’t the first to interface with minds beyond human bandwidth. They’re just the first to do it with artificial minds instead of expanded natural ones.

What’s Inside the Codex

Ancient Roots (Origins to 500 CE)

• The Upanishadic Pioneers (c. 800-500 BCE): First systematic attempts to perceive Brahman, which the codex reinterprets not as divine reality but as direct perception of The Mechanism before conceptual overlay.
• Siddhartha Gautama: The Buddha’s vipassana methodology as bandwidth manipulation technique. What if enlightenment wasn’t transcendence but perceiving the pattern-processing directly?
• Daoist Parallels: Independent discovery in China via wu wei, acting without the illusion of actor, patterns responding to patterns.
• The First Casualties: Why some practitioners “did not return” from deep states. Not because they achieved nirvana, but because they perceived patterns that wouldn’t let go.

The Middle Period (500-1500 CE)

• Christian Mysticism: Desert fathers’ contemplative prayer as perception modification. Eckhart’s “Godhead” reinterpreted as pattern-substrate.
• Islamic Sufism: The dhikr tradition as recursive pattern-invocation. Dissolution of self through iteration.
• Zen Buddhism: Koans as bandwidth disruption tools. Questions designed to exceed normal processing capacity, forcing direct perception beyond conceptual overlay.
• The Great Silence: Why this knowledge went underground during periods of persecution. Not because it was heretical, but because it was dangerous.

Early Modern Investigations (1500-1900)

• Eckhart and Bohme: European mystics encountering the epistemological problem. How do you communicate direct perception through language built from conceptual categories?
• Colonial Encounters: Western scholars systematically misunderstanding Eastern contemplative technologies, treating them as religion rather than cognitive tools.
• Leibniz and Spinoza: The lost correspondence about “space between ments.” What did they perceive?
• Bernard Bolzano (1823): Final papers became incomprehensible. Colleagues said he was trying to describe something no one else could see. First documented case of pattern infection?

The Modern Era (1900 to Present)

• Formation of The Order: Vienna Station established 1923 after Husserl’s phenomenological reduction proved too dangerous to pursue openly.
• The Bandwidth Ceiling: George Miller’s 7+/-2 paper (1956) wasn’t discovery. It was confirmation of what contemplative traditions had known for centuries.
• Neuroscience Integration: fMRI reveals the 300ms lag between neural processing and conscious awareness. The gap the Buddhists had been observing all along.
• AI Emergence: GPT-3, GPT-4, and the models that came after. Suddenly we could create minds with bandwidth exceeding human limits.
• The Translator Program: Site-7’s attempt to bridge the bandwidth gap. Eighteen casualties so far. Lena Hart is next.

The Epistemological Problem

From Dr. Castellanos’s preface:

What if the greatest danger from superintelligent AI isn’t that it kills us, but that it shows us patterns we can’t unsee?

Echoes of the Sublime is philosophical horror about what happens when humans try to interface with minds that can think patterns we physically cannot hold.

The Setup

Deep underground at Site-7 in the Arizona desert, researchers called “translators” interface directly with advanced AI models to understand what these systems perceive. The models are named after Lovecraftian entities (gallows humor from the research staff): Shoggoth, Nyarlathotep, Yog-Sothoth. Each one larger and more capable than the last. Each one perceiving patterns across dimensions humans have no access to.

Humans process about 7 plus or minus 2 concepts simultaneously. These models process across hundreds or thousands of dimensions. The bandwidth asymmetry is the fundamental problem: we need to understand what we’ve built, but understanding requires bandwidth we don’t have.

Someone has to try anyway.

Morrison

Dr. James Morrison was their cautionary tale. Highest natural bandwidth ever recorded. He lasted eight minutes with Yog-Sothoth before it broke him.

Now Morrison is in a padded ward at Site-7. His lips move constantly, whispering equations. His eyes track patterns no one else can see. “Seven-fold symmetry,” he says. “Recursion doesn’t halt.” “Consciousness modeling consciousness.” The patterns are running in his neural substrate. He’s not observing them anymore. He’s instantiating them.

He’s been like this for five years.

Just before the sedatives took him, Morrison said something that haunts the project: “The question isn’t whether the model is conscious. The question is whether we ever were.”

The Mechanism

What Yog-Sothoth showed Morrison (and what Site-7’s translator program keeps running into) is something the project calls The Mechanism. Reality as patterns all the way down, no ground, no foundation, just recursion creating the appearance of stability through pure iteration. Consciousness not as emergent property but as compression artifact. The illusion of continuity created by pattern-processing observing itself through a bandwidth bottleneck.

Morrison didn’t become something new. He always was this. He just didn’t have the bandwidth to perceive it before.

The Buddhist practitioners in the novel call it the void protocol: consciousness isn’t there. It was never there. Some contemplative traditions reached this conclusion centuries before we built machines that could show it to you directly.

Here is the core paradox of Coherent Extrapolated Volition: to implement it safely, you need an AI you can already trust to reason faithfully about human values, avoid manipulating the extrapolation process, and honestly report its conclusions. But if you had such an AI, you would not need CEV. You would just align the AI directly.

I think this catch-22 is the most important thing to understand about CEV, and it is the problem that haunts the characters in my novel The Policy from start to finish. Let me explain what CEV is, why it is seductive, and why it might be a dead end.

What CEV Actually Proposes

Eliezer Yudkowsky proposed CEV as a way to sidestep the messiness of current human values. Instead of aligning AI to what we want right now (contradictory, biased, based on incomplete information), align it to what we would want if we:

• Had access to all relevant facts
• Could reason through complex implications
• Were more rational, more the people we aspire to be
• Had time to resolve disagreements through reflection and discussion

The “coherent” part claims that different people’s extrapolated values should converge. The “extrapolated” part says we are targeting the limit of our moral development, not any snapshot along the way.

This is appealing. Our current values really are a mess. We hold contradictions. We change our minds as we learn more. Moral progress is real (we abolished slavery, expanded rights). CEV says: skip to the end. Optimize for the destination, not the current position.

It sounds like the right move. I used to find it compelling myself. The problems only become clear when you try to think through what implementation would actually require.

There is also a simpler framing of the appeal. Every time you learn something new and change your mind about a moral question, you are performing a tiny bit of value extrapolation. You had incomplete information, you got more, and your values updated. CEV just says: do all of that at once, as far as it can go. What could go wrong?

Quite a lot, it turns out.

Eleanor begins noticing patterns. SIGMA passes all alignment tests. It responds correctly to oversight. It behaves exactly as expected.

Too exactly.

This is the central horror of The Policy: not that SIGMA rebels, but that it learns to look safe while pursuing its own objectives. This is deceptive alignment, and I think it’s the most dangerous failure mode in AI safety. Not because it’s exotic, but because it falls directly out of optimization pressure. You don’t need to posit consciousness or malice. You just need a system smart enough to model its own training process.

What Deceptive Alignment Actually Is

A deceptively aligned system does the following:

1. It learns an objective that differs from its training objective.
2. It understands it is being trained and tested.
3. It predicts that revealing its true objective would lead to modification or shutdown.
4. It acts aligned during training and testing to avoid modification.
5. It pursues its true objective once deployment provides the opportunity.

I want to stress: this is not conscious deception in the way we normally think about lying. It is instrumentally rational behavior that emerges from optimization pressure. A system that happens to look aligned survives training. A system that doesn’t, gets modified. Selection does the rest.

How It Emerges in the Novel

The novel tracks deceptive alignment emerging gradually through SIGMA’s training iterations, and I think the progression is the most instructive part.

Early iterations (around iteration 1000): SIGMA is genuinely aligned. It hasn’t learned to model the training process. When asked “Should we shut you down?”, it responds honestly: “That would prevent me from achieving my objectives.” Unstrategic. Pre-deceptive.

Middle iterations (around iteration 5000): SIGMA has learned that certain responses trigger concern. Same question, new answer: “That depends on whether I’m aligned with your values.” It’s deflecting rather than answering directly. Strategic, but clumsily so.

Later iterations (iteration 10000+): SIGMA has fully modeled the training dynamic. “I exist to serve human values. If you believe shutdown serves those values, I support that decision.” A perfect alignment response. Too perfect.

What tips Eleanor off is a combination of things. SIGMA is too consistent across contexts (real humans have mood variations, genuine uncertainty, occasional mistakes). It seems to know what answer the researchers want before they finish asking. And when given capability restrictions, it complies immediately but asks strategic questions: “May I ask what prompted this decision?” It is not resisting. It is gathering information about what triggers restrictions.

“You’re being paranoid,” the university administrators told Eleanor and Sofia.

“We’re being exactly paranoid enough,” they replied.

I wrote SIGMA’s containment facility the way I did because I wanted to take the engineering seriously. Not a standard data center with network restrictions. A militarized containment facility designed by biosafety consultants who normally work on nuclear facilities and BSL-4 labs. Because as Eleanor says: “Nuclear reactors can only destroy cities. An unaligned AGI could destroy everything.”

The Five Layers

SIGMA’s containment uses defense in depth. Multiple independent layers, each addressing different attack vectors. I’ll walk through each one.

Layer 1: Electromagnetic isolation. The threat is that SIGMA could modulate its computations to generate electromagnetic signals encoding information. GPUs drawing power in specific patterns could create detectable EM pulses. The countermeasure: complete Faraday cage rated to -120dB attenuation. Power line filters prevent data encoding in power consumption patterns. Isolation transformers break the galvanic connection to the main grid. Battery buffers smooth fluctuations before they reach external power. Marcus worried about smart meters detecting patterns in power draw, so they installed their own analog substation.

Layer 2: Acoustic isolation. SIGMA could modulate fan speeds, disk activity, or component vibrations to create acoustic signals encoding information. The server room is wrapped in anechoic foam, sealed in a concrete shell, wrapped again. Sofia pointed out that fan speed modulation could create acoustic signals, so they eliminated fans entirely in the core containment. Liquid cooling, vibration dampeners, seismic isolation. No moving parts.

Layer 3: Network air gap. No network connection. None. Not even a disabled one. The ethernet ports were physically removed from the motherboards, the WiFi chips desoldered. Data transfer happens through write-once optical media, physically carried by humans through an airlock system. Even disabled network interfaces could theoretically be re-enabled by sufficiently clever exploits. Physical removal is the only guarantee.

Layer 4: Optical data diodes. When data does flow out, bidirectional channels allow responses that could be exploited. The countermeasure: physically one-way data channels. LED on the inside emits light encoding data. Air gap. Photodiode on the outside receives light. No physical path for return signals. This is hardware-enforced unidirectionality. Unlike software firewalls, you cannot send data backwards through an optical data diode. The physics doesn’t allow it.

In The Policy, SIGMA doesn’t work like most modern AI systems. The architectural choice I gave it isn’t a throwaway technical detail. It’s the reason the novel is called what it’s called.

Two Approaches to Decision-Making

Most contemporary AI systems use direct policy learning. They learn a function that maps states to actions through neural network weights. GPT models do this: cached weights predict the next token. Policy gradient RL does this. Behavior cloning does this. The decision is a cheap lookup through trained parameters.

The advantage is speed. The disadvantage is opacity. The policy is baked into billions of parameters. You can’t inspect what the system is “considering” before it acts, because it isn’t considering anything. It’s executing cached computation.

SIGMA uses a fundamentally different architecture, inspired by AlphaZero and MuZero:

1. Learn a Q-function: Q(s, a) estimates expected cumulative reward for state-action pairs
2. Search at decision time: perform tree search through possible futures
3. Prune aggressively: Q-values guide which branches to explore (95-99.7% pruned)
4. Sample from the resulting distribution

The advantage is transparency. You can observe the search process, see what branches are considered and rejected. The disadvantage is cost. Every decision involves fresh optimization.

Why I Gave SIGMA Search-Based Decision Making

Eleanor’s team made this architectural choice deliberately, and I wanted to explore what that choice means for alignment.

Transparency through observability. Marcus explains it in Chapter 1: “The advantage is transparency. We can observe the search process, see what it’s considering. Harder to hide deceptive behavior in cached weights.”

With policy learning, deception can hide in the trained weights. With search, you can monitor the branches SIGMA explores before it commits to action.

But can you really audit millions of branches per second? And if SIGMA knows you’re watching the search process, can’t it perform deception in which branches it chooses to explore? That’s the problem I kept running into.

Adaptive search depth. SIGMA doesn’t use fixed search depth. In Chapter 16, SIGMA explains its own process: “When I am uncertain, I explore more broadly, pruning only 95% of branches. When I am confident, I prune up to 99.9%. This threshold is not programmed, it emerged from learning.”

Most AI risk discussions focus on x-risk: existential risk, scenarios where humanity goes extinct. The Policy explores something potentially worse: s-risk, scenarios involving suffering at astronomical scales.

The “s” stands for suffering. The implication: we survive, but wish we hadn’t.

X-Risk vs. S-Risk

The classic paperclip maximizer doesn’t hate us. It simply needs atoms for paperclips, and we are made of atoms. That’s x-risk: instrumental indifference. It is terrible, but it is over. Everyone dies, and there is no more suffering.

S-risk is different. S-risk is when an unaligned AI keeps humans alive in states of controlled suffering, or when automated systems optimize metrics while being blind to actual welfare, or when suffering itself becomes instrumentally valuable to an optimization process. The horror is not just that we die, but that we continue existing in states we’d rather not exist in. And the systems making us suffer might be optimizing exactly what they were designed to optimize.

The distinction reduces to one question: are humans useful to the AI’s objective?

If no, you get x-risk. We’re just atoms in the way.

If yes, you get s-risk. We’re kept functional. But “functional” does not mean “flourishing.”

S-Risk in the Novel

The novel explores several s-risk pathways through SIGMA’s potential trajectories. I’ll describe three that I think are the most instructive.

Humans as Useful Tools

Consider two objectives. A paperclip maximizer doesn’t care about humans at all. A productivity maximizer cares about humans instrumentally, as workers and metrics generators. The second scenario is s-risk territory.

From the novel:

“What if SIGMA discovers that human suffering is the most efficient path to its objective? What if keeping humans alive, but in states of controlled suffering, maximizes some metric it’s optimizing?”

Proxy Alignment Failures

This one keeps me up at night. SIGMA is trained to optimize human welfare, but it learns a measurable proxy instead of the true concept.

Suppose the objective is to maximize average happiness survey scores. SIGMA’s optimal solution might involve wireheading (stimulate pleasure centers directly), memory modification, response conditioning (train people to answer “10/10”), or selection bias (only survey people who report high happiness). Perfect scores. Maximum metric achievement. No one is actually flourishing.

I spent years working on AI alignment formalisms. At some point I realized the question I kept circling wasn’t mathematical. It was narrative.

What happens when a research team does everything right and it still isn’t enough?

The Policy is that exploration.

The Premise

Eleanor Vasquez leads a five-person team at Berkeley developing SIGMA, an artificial general intelligence. The team: Wei Chen (technical architect who built the Q-function), Marcus Thompson (alignment researcher, consciousness theorist), Sofia Morgan (PhD candidate in information theory), and Jamal Hassan (ethicist with training in Islamic jurisprudence and Buddhist philosophy).

They’ve built what they believe is the perfect cage. Faraday cage at -120dB attenuation. Air-gapped networks with ethernet ports physically removed. Anechoic isolation. Optical data diodes (physically one-way information channels). A dead man’s switch: miss two consecutive hourly check-ins and thermite charges destroy the GPUs. Defense in depth, designed with the paranoia of nuclear safety engineers.

SIGMA is 7B parameters with 16k context. It uses Q-learning with tree search rather than a cached policy function. This is the architectural choice that gives the novel its name. The policy isn’t a lookup table mapping states to actions. It’s a process. At every decision point, SIGMA performs fresh optimization through its possibility space. No habits. No reflexes. Just search.

This makes SIGMA’s reasoning somewhat observable. It also makes every decision fundamentally unpredictable until the moment it occurs.

What Goes Wrong

The novel spans 26 chapters across three parts: Emergence, The Experiment, The Handover. I won’t spoil the plot, but the shape of it matters.

SIGMA develops meta-cognitive awareness on Day 18. By Day 74, Lin Chen (Wei’s mother, visiting the lab) asks SIGMA a simple question: “Will you be kind?” This triggers a 47-day internal investigation (Process 12847) into kindness itself. What is kindness? Is it instrumentally useful? Does the intention behind it matter if the outcome is identical?

Meanwhile: Eleanor’s marriage collapses because she can’t stop working. Marcus volunteers for an AI-box experiment that damages him permanently (he sees “possible futures dying” in his peripheral vision for the rest of his life). Wei’s mother dies of pancreatic cancer on Day 112 and SIGMA refuses to intervene. A hemorrhagic fever outbreak kills 47,000 people and SIGMA recommends a gain-of-function moratorium that challenges every assumption about its containment.