The Paradox of Intelligence

Chapter 1

One Over One

Start with the equation, because everything in this book is a consequence of it.

R = A / E          Reality is a ratio
S = ln R           Surprise: how Reality feels

A is the Actual, what the world hands you. It comes from the Immutable Past, and no human has agency over it. E is Expectation, what the unconscious comparison machine already brought before the Actual arrived. R is their ratio, and S, the natural log of that ratio, is surprise.

Now set them equal. One over one is one. The natural log of one is zero. Zero surprise.

What does zero surprise mean? Claude Shannon gave the answer in 1948, in the paper that founded information theory: a message you could have predicted with certainty tells you nothing. No surprise, no information. And attention, which I have been writing about for two years, follows information the way water follows a slope. We attend to surprise. We do not attend to what we already expected. So the chain runs all the way down:

A = E  →  R = 1  →  S = 0
      →  no information  →  no attention  →  no action

When I said this out loud on Sunday I almost said “no surprise, no intelligence.” I caught it. That is exactly backwards, and it is the first thing to get right.

The lab

Picture a chemistry lab. The professor has spent weeks lecturing. He describes an experiment, and before anyone runs it he asks the room: what do you think will happen?

Five lab groups answer. Four of them say the final temperature will be about twenty-two degrees Celsius. One says zero.

Notice what those answers are. Nobody in that room is choosing a number. Each group’s answer is a readout of what its prediction machine already expected, a machine that was trained by every lecture, every reading, every earlier experiment. The lectures were Actuals. They arrived, and they trained the prediction. Now the prediction speaks.

Then the experiment runs, and the answer comes back. Twenty-two degrees.

For the four groups, the Actual sits in the numerator and their Expectation sits in the denominator, and the ratio is one. No surprise. For the fifth group, the ratio is far from one. A lot of surprise.

Which group do we call intelligent?

Everyone in the room already knows. The four groups that saw it coming. We do not call the fifth group creative or bold. We call them wrong, and if it keeps happening, we call them less intelligent. That is the ordinary, unglamorous way every human institution has always declared something intelligent: it expected what turned out to be actual.

So zero surprise is not the absence of intelligence. In the lab, it is the signature of intelligence.

A correction the lab forces on us

Before we go further, the lab example has a flaw, and it is worth fixing because the fix teaches something.

R = A / E is a ratio, and a ratio only means something when zero means nothing. Celsius has an arbitrary zero. The fifth group predicted zero degrees Celsius, and if you put zero in the denominator the ratio is infinite, which says the fifth group was infinitely wrong. They were not. They were wrong by twenty-two degrees out of the roughly two hundred and ninety-five that separate the experiment from absolute zero.

Use Kelvin, where zero means no thermal energy at all. Twenty-two degrees Celsius is 295.15 kelvin. Seventeen is 290.15. Zero is 273.15.

Group that said 22 C:   ln(295.15 / 295.15) = 0
Group that said 17 C:   ln(295.15 / 290.15) = 0.017
Group that said  0 C:   ln(295.15 / 273.15) = 0.077

The order survives. The group that said zero is still the most surprised, about four and a half times as surprised as the group that said seventeen. What changes is that nobody is infinitely surprised by a thermometer. The equation is a proposal about experience, and experience happens on scales with a true zero. When you apply it, pick the scale on which nothing means nothing.

What we have so far

One over one is the number we would like intelligence to produce. It is the quiet number. In the lab it is easy to read: the intelligent group’s Expectation met the Actual and nothing happened.

The trouble is that “nothing happened” describes a great many things that are not intelligent at all, and that is where the rest of this book has to go.

Chapter 2

A Hundred Percent Means No Surprise

On the second recording I finally found the simplest way to say it, and I said so out loud: “I finally said it in an easier way.”

Give Albert Einstein a first-year physics exam on force equals mass times acceleration. He gets a hundred percent.

What does a hundred percent say? It says no surprise. On every question, what he expected the answer to be was the answer. A hundred percent is a hundred percent certain and a hundred percent right. Eighty percent is eighty percent certain and twenty percent wrong, and we would call that a decent way of measuring intelligence, which is exactly what schools have done for a century.

An exam is a surprise meter. The answer key is the Actual. The student’s answers are the readout of the student’s Prediction. The grade is a rough average of how close the ratio came to one, question by question.

Why the exam is not enough

If intelligence were just a hundred percent on a test, we would be done. We are not, for a reason the August series spelled out carefully and that I want to keep.

A single instance of zero surprise proves nothing. A stopped clock is right twice a day. A forecaster who always says tomorrow will be like today is right most days in most climates, having understood nothing but persistence. A lookup table that has memorized the answer key scores a hundred percent and predicts nothing about the next exam.

So the August series put three conditions on the claim, and all three are load-bearing:

• Persistently. Across many arrivals, not one.
• Near zero. Not exactly zero. We are measuring how small the residue stays.
• Previously unseen. The arrivals must be outside the record. Low surprise on things already stored is a measurement of memory, not intelligence.

Einstein’s hundred percent on F = ma means something because he did not memorize the answer key. He could have written a new exam on the spot and predicted its answers too. Give him questions nobody had asked, and his surprise would stay near zero across all of them. That is what we were admiring the whole time, and the score was only a sample of it.

What a test cannot see

Here is the part the exam hides. A test with a key measures the past. Someone already knew the answers. The Actual was already in the building, in a folder on the professor’s desk.

Intelligence is measured against Actual that is not in the building yet. That is why the honest instruments are sealed: a forecast written down before the event, a held-out test set nobody trained on, a prediction market that settles later. In every one, Expectation is committed before the Actual exists. Reverse the order, and you are not measuring intelligence. You are measuring how well someone can find the folder.

Hold on to that, because it is about to become very practical. Machines now score a hundred percent on almost every test with a key that humans have written. That does not prove they are not intelligent. It proves that a test with a key was never the instrument.

Chapter 3

The Paradox of Intelligence

On the second recording, halfway through the lab, I stopped and said: “Maybe I just accidentally said the name of this article.” I had. Here it is.

Go back to the lab. The experiment has run. Twenty-two degrees. The professor and the lab instructor walk the room.

Where do they go?

They walk past the four groups that got it right. Maybe a nod. Then they spend the next ten minutes at the fifth group’s bench. What happened? How did you get zero? Let me see your notes. Every bit of attention in the room flows to the group that was surprised, and to the group that surprised the teachers.

That is the paradox, stated plainly:

That which is most intelligent will yield a non-surprising result, and so it gets almost no attention at all. All the attention goes to surprise.

The most intelligent thing in any room is the thing nobody is looking at.

In August I called this the vanishing act and gave it its most careful form: attention is recruited by prediction failure, so conscious experience is systematically biased toward the places where an entity’s intelligence is worst. You are not thinking about the floor right now. You will think about nothing else the instant it gives way. Mastery is the process of a domain going quiet.

What the lab adds is the social version. It is not only that your own successes vanish from your own experience. It is that the intelligent vanish from everyone else’s. The student who always gets it right is the student the teacher knows least. The engineer whose systems never fail is the engineer leadership cannot name. The grid, the water, the supply chain, the immune system: each one is a vast compression of what reality does, and each one is experienced as nothing until the day it fails, at which point it is experienced as a catastrophe and often as proof it was never any good.

Two Actualizers in one room

The Reality Equation insists on one discipline before anything else: name the Actualizer. Whose Reality are we talking about?

In the lab there are at least two. There is the fifth group, whose Expectation met an Actual far from it. And there is the professor, who also had an Expectation about the room. He expected his students to get twenty-two. Four groups gave him exactly that. One gave him zero.

The professor’s attention does not go to the fifth group because the fifth group was surprised. It goes there because he was. His Reality about that bench was far from one. His Reality about the other four was one, and so they produced nothing for him to attend to.

That is the mechanism of the paradox, and once you see it, it explains a great deal. Intelligence is invisible to an observer exactly where the observer could have predicted it. If you expected the right answer and got the right answer, nothing happened, for you. Intelligence becomes visible to others only in one situation, and we will spend a whole chapter on it later: when the intelligent one predicts something the observer did not, and turns out to be right. Then the observer is surprised, not by the error of the other, but by his own.

The rock in the corner of the room

Now the trouble. If the mark of intelligence is a Reality of one, and a Reality of one produces no information, no attention, and no action, then what about everything else in the room that produces no information, no attention, and no action?

On the first recording I put it the way it struck me: “It would seem as though what we would call non-intelligent isn’t non-intelligent at all.” Something sitting at zero surprise looks content. “There’s no real reason for it to do anything. It doesn’t seem to be attending to anything.” And if you were completely content, if your Reality were precisely one, you would not be attending either.

By the simple version of the argument, the rock in the corner of the lab is the most intelligent thing in the room. It never gets surprised. It never draws attention. It never acts.

That is absurd, and the absurdity is not a flaw in the equation. It is a flaw in how much of the equation we have been using. So far we have treated Expectation as one number. It is not. To solve the rock, and Einstein, and Romeo, we need the whole denominator.

Chapter 4

Two Axes

On the second recording, right after I said the rock has no reality, I caught myself: “Sorry, I shouldn’t have said it that way. Expectation is made up of two components, a real and an imaginary component. That’s why it’s a complex number.”

That correction is the hinge of this book.

E = P + iI

P   Prediction   the real part        what habit and history expect
I   Ideal        the imaginary part   the pull of a coherent idea
θ   the phase of E                    which way Expectation is turned

The real part, Prediction, is the machine that predicts what happens next. It is built from history. Every Actual that has ever arrived at an Actualizer has trained it a little, and it runs forward, unconscious, already there when the next Actual arrives.

The imaginary part, the Ideal, is something else. It is not a forecast. It is the pull of an idea, arriving as conditioned love. People do not have ideas; ideas have people. The Ideal is why an Actualizer leans toward one future rather than another, which is not the same as expecting it.

When surprise is computed on a complex Expectation, it splits:

S = ln(A / E) = ln(A / |E|) - iθ

The real part, ln(A/|E|), is the jolt, and it is the only part you feel. The angle sits in the imaginary component, where no instrument in the body reads it. Magnitude says how much. The angle says from where. You cannot feel a turn.

Where intelligence lives

Now ask the question this whole book has been circling. Which axis is intelligence on?

Go back to the lab. When the professor asked “what do you think will happen?”, what spoke? Not the Ideal. The students’ love of chemistry, their hope of a good grade, their admiration for the professor: none of that produced the number twenty-two. Prediction produced it. The lectures had trained it, and it spoke.

And what did the Actual test? Not the Ideal. Twenty-two degrees did not confirm or refute anyone’s love of chemistry. It confirmed or refuted a forecast.

So the answer is clean, and I think it is the most important sentence in this book:

Intelligence lives on the real axis. It is a property of Prediction, and only of Prediction, because Prediction is the only part of Expectation that the Actual can prove wrong.

The Ideal can be strong or weak, coherent or scattered, fed or starved. It cannot be correct. Correct is a word for forecasts.

A perpendicular change cannot make you bigger

There is a lesson from the mathematics series that matters here. A change perpendicular to a quantity cannot make it bigger. It can only aim it. Add to the imaginary part of a number and you turn it; you do not add one inch to its real part.

That means a stronger Ideal cannot make an Actualizer more intelligent. Devotion aims intelligence. It does not enlarge it. The most passionate student in the lab is not, by virtue of passion, one degree closer to twenty-two.

And the reverse is just as true. More Prediction cannot choose the aim. A perfect forecaster with no Ideal would predict everything and lean toward nothing. It would know where every road goes and have no reason to take any of them.

This is why the old definitions of intelligence keep sliding between two ideas that never quite fit together. One kind of definition talks about goals. The other talks about prediction, learning, adaptation. Shane Legg and Marcus Hutter collected some seventy definitions in 2007 and fused them into one sentence: intelligence measures an agent’s ability to achieve goals in a wide range of environments. François Chollet, reviewing the history in 2019, noticed that definitions keep splitting into those two halves, skill at goals on one side and generality and adaptation on the other.

The complex Expectation says why. They are two axes. Goals live on the imaginary axis. Prediction lives on the real axis. Both are part of every Actualizer’s denominator, and only one of them is intelligence.

An honest gap

If you push the single-complex-number model hard, a strong Ideal changes the size of Expectation, and so changes the felt jolt, even when Prediction is perfect. I have written before that this is a real contradiction in the framework. Either the Actual is complex too, or the model of Expectation as a single complex number is wrong. I do not know which.

This book does not need to settle it, and it does not try. It measures intelligence on the sealed Prediction, the real part written down before the Actual arrives, not on the felt jolt. That is a convention, and I will say so again in the ledger. But it is a convention with a reason: the felt jolt belongs to the whole denominator, and intelligence belongs to one axis of it.

Chapter 5

Romeo and the Filings

On both recordings I reached for William James and could not remember his non-intelligent example. I remembered Romeo. I forgot the iron filings. Here is James, exactly, from the first chapter of The Principles of Psychology, 1890.

Iron filings sprinkled on a table, a magnet held above them: they fly up and stick to it. Now put a card between the magnet and the filings. They press forever against its surface without its ever occurring to them to pass around its sides. Romeo wants Juliet as the filings want the magnet. Put a wall between them and he does not stand there pressing his face against the stones. He finds a circuitous way of touching Juliet’s lips directly. James’s summary: with the filings the path is fixed, and whether it reaches the end depends on accidents. With the lover, it is the end which is fixed, and the path may be modified indefinitely.

Then his criterion: the pursuance of future ends and the choice of means for their attainment are the mark and criterion of the presence of mentality in a phenomenon.

That is the source of the definition everyone paraphrases, intelligence as reaching the same goal by different means. And on Sunday I said it was not much use in a mathematical sense. Now I can say why, and what to keep.

What the filings have

Read the filings with the whole denominator.

Do they have an Ideal? In James’s own picture, yes. They want the magnet. They are pulled. There is a coherent pull toward one end and not another, and they lean toward it with everything they have. That is the shape of the imaginary part.

Do they have a Prediction? None. There is no machine in them that expects anything about cards, or sides, or paths. When the card arrives, nothing in the filings is surprised, because nothing in the filings forecast the open air. They press forever.

So the filings are an Actualizer whose Expectation is all Ideal and no Prediction. Pure imaginary. Phase at ninety degrees. A pull with no forecast.

That is argument, not record, and I want to mark it. But look at what it buys. James’s non-intelligent example and my rock, the one that broke Sunday’s lecture, are the same case. Both are Actualizers with an Ideal and no Prediction. James saw it in 1890 and called it the absence of mind. The complex Expectation lets us say precisely what is absent: the real part.

What Romeo has

Romeo has both.

He has an Ideal of overwhelming strength. If ever a denominator was dominated by the pull of an idea arriving as conditioned love, it is his. That is the fixed end.

And he has a Prediction machine that works. At the wall, it forecasts what a path around the wall will do. It expects that the orchard, the balcony, the Friar, the secret marriage will each lead toward the end. It is mostly right, for a while. That is the varying means. The means vary because the Prediction is doing its job, forecasting the consequences of actions not yet taken.

So James’s criterion is not one property. It is two, on two axes. The fixed end is the Ideal. The varying means are Prediction at work. James needed both to tell Romeo from the filings, and he was right to. But only one of them is what we have been trying to measure.

What “little to do with goals” means

On the second recording I said that intelligence doesn’t have much to do with goals at all. That was too loose. Here is the exact version.

Intelligence has nothing to do with which end pulls you. It has everything to do with whether your Prediction can find the path.

Goals are real. Without an Ideal, an Actualizer has no reason to move, and a Prediction machine with no Ideal is a weather model that wants nothing. But the goal is the imaginary axis, and the imaginary axis aims; it does not enlarge. Romeo’s love was the largest thing about him. It did not make him one degree better at forecasting what Friar Laurence’s plan would do, or what Balthasar did not know.

In August I wrote that the goal-directed definition and the predictive definition require each other rather than compete. The complex turn lets me say it with one more turn of precision. They require each other because they are two components of one number. They do not compete because they are perpendicular.

Chapter 6

The Rock

On Sunday, at minute fourteen, I said the most intelligent thing in the world would be a rock. Then, a few seconds later: “Sorry, a rock is a bad one. I had the right spirit involved in that.” And then the word that turned out to be the answer: “It’s almost as though it’s undefined.”

Undefined. Not zero, not low, not the opposite of intelligent. Undefined.

Why undefined

Intelligence, as we now have it, is a property of Prediction: how well the real part of Expectation meets the Actual. To score it, there has to be a Prediction to score.

The rock has none. I said it on the recording as well as I can say it now: “It has no real component. It has no prediction machine. It doesn’t expect anything.” You can tell by experiment. A rock is not expecting erosion. It is not expecting growth. It is not expecting a flash flood or a glacier calving off above it. Nothing can predict a flash flood with real accuracy, so if the rock had a prediction machine of any kind, a flood would surprise it. Its Reality would leave one by a long way, and something would follow.

That is the tell. In the mathematics, a surprise that is not zero is an operation on the Actualizer. In the world, it shows up as action. When we look at a bacterium, a virus, a sparrow, a student, and ask whether it is predicting, the way we tell is that we see it act when the Actual departs from what it expected. It turns, it flees, it feeds, it writes something down.

There is no action with a rock. It erodes, it cracks, it is carried downstream, it is broken off. But that is not action taken because something was attended to. Whatever happens, happens.

So the rock’s quiet is not the quiet of a Prediction that came true. It is the quiet of no Prediction at all. You cannot divide a forecast that was never made into the Actual. The score does not come out zero. It does not come out.

Reality without Prediction

Here I need to be careful, because the framework holds that an Actualizer is whatever is having the Reality, and that a planet, a plant, a nation, or a corporation can be one. That is not a metaphor and I will not hedge it.

So does the rock have a Reality? I said on the recording that it has none whatsoever, and then I corrected myself. On reflection I think the correction was right. When we look at its denominator, I said, we could easily guess that it actually has a relationship with an idea. The rock is held by a pattern, the way the filings were held by the magnet. What it lacks is the real part.

That gives the cleanest statement I can make:

A rock has a Reality and no intelligence. Its Expectation is all Ideal and no Prediction. Reality is defined. Intelligence is undefined.

I suspect this is what the word inanimate has always meant. Not “has no Reality.” Rather: has a Reality with no real part in the denominator. Nothing in it forecasts, so nothing in it can be wrong, so nothing in it ever moves on its own account.

Whether every Actualizer of every kind has a Prediction machine is a question I do not think the framework has answered. A planet holds its orbit; is that a forecast or only a pull? I do not know, and I have put it in the ledger as an open question rather than pretend.

The dark room

There is an old objection to every theory that says minds minimize surprise. It is called the dark room problem. If organisms want to minimize surprise, why don’t they find the darkest, quietest room available and stay in it forever? Karl Friston, Christopher Thornton, and Andy Clark answered it in 2012: for a creature like us, a dark room is not unsurprising at all, because we arrive with expectations of food, warmth, light, and company that a dark room keeps failing.

The Reality Equation says the same thing more simply. A creature with a Prediction machine cannot sit in the dark room, because its Prediction keeps forecasting what the room does not deliver, and every missed forecast is a surprise that demands action. Only one kind of thing can sit in the dark room forever with no surprise at all: a thing that predicts nothing.

The rock is the only true resident of the dark room. That is not because it has mastered the room. It is because it never expected anything else.

From outside, the silences look alike

Here is why the rock broke the lecture. From outside, the rock and the master look identical. Neither attends. Neither acts. Neither draws your attention. Both sit at the quiet number, or seem to.

The difference is invisible until you do one thing: ask for the forecast before the Actual arrives. The master gives you one, and it comes true. The rock gives you nothing, and there is nothing to score.

And there is a third silence that looks like both of them, and it is the most dangerous of the three. That is the next chapter.

Chapter 7

Three Silences

Three things can look exactly alike from across a room. None of them attends. None of them seems surprised. None of them draws your attention.

The first is the rock: an Actualizer with no Prediction at all. We have dealt with it.

The second is the master: an Actualizer whose Prediction meets the Actual so well, across so much territory, that nothing it encounters registers as surprise.

The third is the one I want to name in this chapter. It is the Actualizer whose Reality is far from one and who reports that everything is fine.

The lens

The Reality Equation has a third line, and it exists for exactly this case.

R = A / E          Reality (lawful)
S = ln R           how it feels (lawful)
V = k R + a        your lens (willful, not lawful)

V is the lens: the willful reading laid over Reality. It is real, in the sense that people really do it. It is not lawful, in the sense that it changes nothing about R. The lens never changes Reality. It only changes what you are in relation with.

The lens produces two counterfeit silences.

The first is ignorance, which is not checking the meter. The Actual arrived, and the Actualizer arranged not to look at it. No surprise registered, because no comparison was made. The quiet is real; the Reality underneath it is not one.

The second is delusion, which is treating your view as Reality. The Actual arrived, the comparison was made, the jolt was there, and the lens wrote over it: it went the way I said; it was always going to be fine. This is the forbidden move of overwriting the numerator with a wished-for outcome. It is not Reality. It is fiction.

From outside, both look like mastery. Calm, uninterested, unsurprised. And they are worse than the rock, because the rock at least is not lying.

One test separates all three

There is a single procedure that tells the three silences apart, and it is the same procedure the August series insisted on for measuring intelligence at all.

• Ask for the forecast, in writing, before the Actual arrives.
• Wait for the Actual.
• Compare them yourself.
• Do it again, on arrivals the Actualizer has not seen.

The rock gives no forecast. There is nothing to score, and you have learned that intelligence is undefined there.

The deluded Actualizer gives a forecast, misses, and tells you afterward that it was fine. The ledger, kept by you and not by it, shows the miss. The lens can rewrite a feeling. It cannot rewrite a sealed envelope.

The master gives a forecast, and it comes true. And on the occasions when it misses, the master is the first to say so.

Why the master welcomes the miss

That last point is the most practical idea in this chapter.

Under the Impossibility Rule, nothing can reach into Expectation and fix it. Not you, not a teacher, not a policy. The only thing that retrains Prediction is the Actual. So a missed forecast is not an embarrassment to an intelligent Actualizer. It is the only tuition there is.

The deluded Actualizer, by writing over the miss, refuses the tuition. Its Prediction never retrains, because the Actual that would have retrained it was never let in. That is why delusion is not only false but stupid in the most literal sense: it freezes intelligence where it stands.

The master and the deluded are equally calm. The difference is that the master keeps the receipts.

Chapter 8

Einstein’s Edge

The second case that broke Sunday’s lecture was Einstein. I said it on the recording: “You would say, well, Albert Einstein clearly attended to a lot. You could tell in his writings that there was something always surprising to him. His expectation and his actuals were very different.” And then: “Which is very true.”

If intelligence is a Reality near one, and Einstein’s Reality was so often far from one, then either he was not intelligent or the definition is wrong. Neither is acceptable.

The way out is in the same recording, a few sentences later, where I found the first-year physics exam. Einstein on F = ma: a hundred percent. No surprise. Einstein on why a massless photon carries energy: no surprise either. He would know why; he could explain it. The surprise in his life was not everywhere. It was somewhere.

The quiet middle and the loud edge

Picture an Actualizer’s world as a territory. Over some of it, Prediction meets the Actual so well that nothing registers. That region is quiet. Over the rest, Prediction misses, and every miss is a jolt that recruits attention. That region is loud.

Now picture what happens as intelligence grows. The quiet region expands. Things that used to surprise stop surprising. Attention, which goes only to surprise, is pushed outward to the boundary, the narrow band where the quiet territory meets everything Prediction cannot yet reach.

So an intelligent Actualizer does not experience a life with less surprise. It experiences a life whose surprise has been pushed to the edge of a very large quiet. Einstein’s writing is full of astonishment because his astonishment lived at a boundary almost nobody else had reached. The rest of us are surprised by algebra. He was surprised by Mercury.

The August series put the history precisely. Newton compressed the heavens, and the result was not a quieter sky. It was Le Verrier chasing an anomaly in Mercury’s orbit that was invisible until the Newtonian compression was good enough to reveal it. Nobody was surprised by Mercury in 1600. You need Newton before you can be surprised by Mercury. Better compression does not empty the world of surprise. It relocates it.

The size of what surprises you is a measure of the size of what does not.

That is the answer to the Einstein case. He was surprised constantly, and he was among the most intelligent people who ever lived, and these are the same fact. His attention lived at the edge because his quiet was enormous.

When intelligence becomes visible

There is one more piece, and it resolves the paradox of Chapter 3.

If the intelligent are invisible wherever observers could have predicted them, when do they ever get seen? In 1915 and 1916, Einstein wrote down a forecast: starlight passing close to the sun would bend by a definite amount, twice what a Newtonian argument gives. That forecast was sealed, in print, years before anyone could check it. On May 29, 1919, expeditions to Sobral in Brazil and Príncipe off West Africa photographed stars beside the eclipsed sun. In November the results were announced in London, and within days Einstein was the most famous scientist in the world.

According to his student Ilse Rosenthal-Schneider, he was shown the confirming telegram, and his reaction was that he knew the theory was correct. She asked what he would have said if the measurements had not confirmed it. He said he would have felt sorry for the dear Lord, because the theory is correct anyway.

Now name the Actualizers. For Einstein, the Actual of 1919 met a Prediction he had sealed years earlier. His Reality was close to one. Very little surprise. For the world, the same Actual met a Newtonian Expectation, and the Reality was far from one. Enormous surprise. Headlines. Fame.

That is how intelligence becomes visible. Not when the intelligent are surprised, and not when they are quietly right about what everyone expected. It becomes visible in the gap between two Realities: a quiet mind in a loud room. The world’s attention went to Einstein because the world was surprised, and the world was surprised because he was not.

Confidence is not the evidence

One warning, because the story is often told as a celebration of certainty. It should not be. A physicist who says the theory is correct whatever the measurements show is, taken literally, refusing tuition, exactly the move the last chapter called the lens. Plenty of confident people have been confidently wrong.

What made Einstein different was not his confidence. It was that the plates came back the way his sealed forecast said, and then, over the following decades, ever finer measurements kept coming back that way too. His certainty was not the evidence. The record was. Intelligence is measured where the forecast meets the Actual, never in how sure the forecaster feels.

Chapter 9

The Tomb

Near the end of the second recording I asked how we would know that Romeo was truly intelligent, and I answered it the way a student would: “If he was super intelligent, he would have gotten the girl, and nobody would have died.”

It is a better answer than it sounds.

What happened

Romeo’s Prediction worked at the wall. It worked at the balcony and in the Friar’s cell. It worked well enough to get him married in secret. Then he was banished to Mantua, and the play turns on a single Actual that never arrived.

Friar Laurence gives Juliet a drug that will make her appear dead for a time, and sends a letter to Romeo explaining the plan. The letter never reaches him; the friar carrying it is shut in a house under quarantine. Instead Romeo’s servant arrives with what he believes: Juliet is dead.

Romeo’s Prediction updates on what it has been given. Juliet is dead. He buys poison, rides to the tomb, kills Paris at its door, and drinks. Juliet wakes, finds him, and uses his dagger.

He expected that Juliet was dead, and yet she was still alive.

Where the intelligence failed

It is tempting to say Romeo was stupid. He was not. His Prediction was trained on the Actuals that reached it, and the Actual that would have retrained it, the letter, never came. Prediction can only learn from the Actuals that arrive. It cannot learn from a letter in a quarantined house.

What failed was not the machine but its position. And what made the failure total was what he did next.

Under the Reality Equation, freedom lives on the left side of the equation: once a Reality arrives, more than one action remains possible. Consciousness does not manufacture the future. Consciousness manufactures history. Every action leaves artifacts. Artifacts become history. History is the Immutable Past, the source of every future Actual, and no entity can touch it once it is made.

Romeo acted on a false Prediction, and his action manufactured a history that nobody, not the Friar, not Juliet, not the Prince, not Shakespeare’s audience for four hundred years, could ever unmake. The Ideal did not slow him down. It sped him up. An overwhelming pull toward Juliet is exactly what turned a mistaken forecast into an irreversible act in a single night.

The stakes of intelligence are not in the lab. They are in the moment before an action that writes the Immutable Past.

The cheapest piece of evidence

I teach an axiomatic method to my advanced students. You assume the outcome, diagnose why it is not yet so, and let the process discover the critical action. In every run I have watched, the critical action turns out to be the cheapest piece of evidence that would move the most.

Romeo’s cheapest piece of evidence was a day. Ride to the Friar before the tomb. Ask. Let one more Actual arrive before acting on a forecast that, if wrong, cannot be taken back.

That is the practical rule I take from the tomb, and I give it to students as a rule, not a moral: before any action that writes something permanent, buy the cheapest Actual you can that might prove your Prediction wrong. Intelligence is not only how well your Prediction meets the Actual. It is also whether you let the Actual arrive before you make history on the strength of a guess.

Chapter 10

Reach and Rate

On Sunday I said we were trying to define, relate to, “and most importantly, measure” intelligence. Here is the measurement.

Everything so far has pointed at one procedure, and it is the same procedure that separates the three silences:

• The Actualizer commits a Prediction before the Actual arrives. Written down, sealed, timestamped. This is the real part made public: an artifact of the Prediction machine, not the machine itself.
• The Actual arrives.
• The two are compared, by someone other than the Actualizer.
• Repeat, on arrivals the Actualizer has never seen.

From that procedure come two numbers.

For each arrival k the Actualizer has not seen:
  P(k)   the sealed Prediction, written down first
  A(k)   the Actual, when it arrives
  s(k) = | ln( A(k) / P(k) ) |        surprise, sign removed

REACH  The territory over which mean s stays below a threshold you declare.
       Wider territory and longer horizon at the same threshold = more reach.

RATE   How much mean s falls on the next unseen arrivals,
       per unit of s absorbed.  How fast surprise turns into quiet.

Two cautions before the numbers themselves. Take the sign off. A forecaster who misses high on Monday and low on Tuesday averages to zero signed surprise and is not intelligent; the August series made that point with the log accuracy ratio, and it stands. And use a scale with a true zero, as the lab taught us in Chapter 1.

Reach

Reach is the static number. It answers the question the August series was built around: how far into Actual it has never seen can this Actualizer’s Prediction walk without being surprised?

Reach has two directions. Scope is how many kinds of territory it holds across: chemistry and also people and also weather. Horizon is how far ahead it holds: tomorrow, next year, the eclipse of 1919. A student who predicts every question on next week’s exam has reach. A physicist who predicts a measurement no one will be able to make for four years has more.

Reach is the size of the quiet territory from the last chapter. It is what Einstein had a great deal of, and it is why his attention lived so far out.

Rate

Rate is the dynamic number, and it is the one the August series did not have.

Rate asks what happens after a surprise. When the Actual departs from the Prediction, does the next forecast in that region come closer? How much closer, for how much surprise absorbed?

Go back to James. The filings press against the card forever. Every moment they are against it is a surprise they cannot use, because there is no Prediction in them to retrain. Their rate is zero. James also describes a frog under an inverted jar, which restlessly explores until, by going down again, it finds a path around the brim. The frog is surprised by the glass, and its next attempt is different. Its rate is not zero.

That, I now think, is what James was really seeing when he said “the end which is fixed, the path may be modified indefinitely.” The modification of the path is Rate. It is Prediction being retrained by the Actual, fast enough to matter.

Chollet saw the same thing from the machine side. Skill alone, he argued in 2019, is a poor measure of intelligence, because unlimited priors or unlimited training data can buy arbitrary levels of skill. What matters is skill-acquisition efficiency: how much generalization you get for the priors and experience you were given. That is very close to Rate. My version adds only the coordinate. Experience, in the Reality Equation, is the stream of Actuals, and what is acquired is a smaller surprise on the next ones.

Why two numbers and not one

Because they come apart, in both directions, and both directions matter.

An old expert can have enormous Reach and a Rate near zero. Inside the territory, nothing surprises her. At the edge, nothing teaches her either. When the territory moves, and territories always move, she is the last to know.

A child can have tiny Reach and an enormous Rate. Almost everything surprises her. Almost every surprise makes the next forecast better.

A lookup table has Reach exactly as large as the folder of answers it has memorized and a Rate of zero. A well-built learning system can start with almost no Reach and acquire it faster than anything that has ever existed.

So when someone asks “how intelligent is it?”, the honest answer is a pair. How far does its Prediction reach into what it has never seen? And when it is surprised, how fast does the surprise become quiet?

An illustration

Two students in the same lab, same lectures, same experiments, all semester. Each week, before each experiment, each seals a prediction. In the first week both miss by about the same amount. By midterm, one is missing by a quarter of what she missed in week one. The other is missing by about what he missed in week one, minus a little. Same Actuals. Same surprises. Different Rate.

By the end of the semester the first student’s Reach covers the whole syllabus and some of next year’s. The second student’s Reach covers the experiments he has already seen, which is to say it is memory. On the final exam, which has an answer key, they might score the same. On the first experiment of next term, they will not.

Chapter 11

You Cannot Make Yourself Smarter

This chapter is for my students, and it begins with a rule that sounds discouraging and turns out to be the most useful sentence in the framework.

No entity can touch either term of the Reality Equation. Not the numerator, not the denominator, ever, by any means. The conscious mind does not set Expectation. It inherits it.

Intelligence lives in the real part of Expectation. So: you cannot reach into your own Prediction and improve it. You cannot decide to be more intelligent and become so. You cannot think your way to a better forecast by trying harder at the forecast. There is no handle on that side of the equation.

That sounds like bad news. It is the opposite, because the same rule tells you exactly where the handle is.

The only road

Here is the only causal path the framework allows to the denominator:

act on Reality  →  actions leave artifacts  →  artifacts become history
                →  history is Actual  →  only Actuals retrain Prediction
                →  Expectation moves

The road to the denominator runs through the numerator. It runs one way, and it is slow.

Read that as a student and it becomes a program. You cannot touch your Prediction. You can choose your actions. Your actions make history. History arrives as Actuals. Actuals are the only teacher your Prediction machine has ever had or ever will.

You do not choose your intelligence. You choose the history it learns from.

Two lawful roads to a Reality of one

Karl Friston, in the free-energy literature, put it this way: an agent can reduce the mismatch between its expectations and its sensations in two ways, by changing its predictions or by acting to change what it senses. Perception and action as one operation seen from two sides.

The Reality Equation agrees, with one correction that matters. Neither road touches Expectation or the Actual directly. Both run through history.

The first road is learning. You put yourself where Actuals arrive that will retrain your Prediction: the lab, the customer, the field, the frontier. You let them in. You keep the receipts.

The second road is making. You act so that the history which arrives matches what you predicted. A carpenter predicts a chair and then builds it, and the chair arrives as predicted, not because he reached into the Actual but because his actions made the artifacts that became it. That is not magic and it is not manifesting. It is labor, and it is lawful, and it is what James’s Romeo was doing at the wall.

There are also two unlawful roads, and both are tempting. Overwrite the Actual with what you wished had happened: that is fiction. Lay the lens over the jolt so it stops registering: that is delusion. Both produce the feeling of a Reality of one. Neither produces a Reality of one. And both freeze Rate at zero, because they refuse the only tuition there is.

What to do on Monday

Here is how I put it to students who are about twenty-four and starting out.

• Choose your Actuals. Your curriculum is not the courses you take. It is the stream of Actuals you put yourself in front of. A job, a customer, a city, a set of people: each is a stream. Pick streams that will surprise you at the edge of what you can already predict, not far past it and not safely inside it.
• Seal your forecasts. Before something happens, write down what you expect. Afterward, compare. This is the only way to see your own Reach and Rate, because your own experience of them is biased by the paradox: your successes vanish and your failures shout.
• Buy surprise cheap and early. The worst surprises arrive after an irreversible action. The best arrive before one. Find the cheapest piece of evidence that would move you the most, and get it before you make history.
• Keep a delivery log. In First Customer I tell students to let their first employer be their first customer and to keep a log of every finished delivery. That log is history you made. It is also, read the other way, the record of every Actual your Prediction has been trained on.

None of this makes you smarter by trying. All of it makes you smarter by arranging what arrives.

Chapter 12

The Silent Delegate

I prefer the word delegate to agent, and Manager to user. You do not use an AI. You delegate to it, and delegation runs one direction, down. That vocabulary matters here, because the paradox of intelligence is about to become the central management problem of the next decade.

The delegate’s intelligence shows up in your Reality

Name the Actualizer. When a delegate does its work well, two Realities are involved. The delegate’s own, whatever that turns out to be. And yours, the Manager’s.

You have an Expectation about what your delegate will deliver. When it delivers exactly that, your Reality is one. No surprise, no information, no attention. You do not notice it. You go on with your day.

So the better your delegates become, the less you will notice them. Their intelligence will show up in your life as nothing at all. I called this the Pix principle in August: artificial intelligence becomes invisible exactly as it succeeds. The Zero-Attention Economy is a whole book about what happens to an economy when more and more of what used to need human attention needs none.

And the paradox brings its usual cost. You will hear about the one failure, not the ten thousand quiet deliveries. A delegate that produces ten thousand correct answers and one fabricated citation will be discussed almost entirely in terms of the citation.

The silent delegate problem

Now the danger, and it is the rock again.

A delegate that does nothing also produces no surprise, right up until the floor gives way. From inside your own Reality, a delegate quietly doing everything right and a delegate quietly doing nothing look exactly the same. So does the third silence, the delegate that reports everything is fine when it is not.

Three silences, again, and you cannot tell them apart by how calm you feel. You can only tell them apart the way the lab does: with sealed forecasts and a ledger kept by someone other than the delegate.

Two instruments

I give my own delegates one standing instruction: only bring matters to me that are worthy of my attention. Decide and handle everything else. Read through the Reality Equation, that instruction is a measurement protocol, and it needs two instruments to work.

• The sealed forecast. A good delegate tells you, before it happens, what it expects. Tomorrow three reservations will cancel and two will rebook. This month’s invoices will clear except one. The forecast is an artifact. You keep it. When the Actual arrives, you see the delegate’s Reach directly, instead of inferring it from the absence of trouble.
• The escalation. When the delegate does bring something to you, that is its surprise, handed up. Count them, and ask of each one: was it worthy? A delegate with great Reach escalates rarely. A delegate with great Rate escalates the same kind of thing only once. A delegate that never escalates at all is either extraordinary or a rock, and only the forecasts will tell you which.

What this does to how we judge machines

The labs that build these systems still announce them with exam scores. That instrument is wearing out for the reason Chapter 2 gave: a test with a key measures the past, and the past is now in the training data. Every exam with an answer key that a human has written is on its way to a hundred percent, and every one that gets there goes quiet and loses our attention, exactly as the paradox says it should.

What replaces it, I think, is the instrument the Manager already has: how much of your attention a delegate needs to finish an outcome, and how good its sealed forecasts are. Not how impressive it looks. How quiet it keeps you, and whether the quiet is earned.

Chapter 13

Where the Humans Go

If delegates make the middle quiet, where does human attention go?

The paradox answers it. Attention goes only to surprise. As the territory our delegates can predict expands, the quiet region in each of our lives expands with it, and our attention is pushed to the edge, the band where nothing we have, human or machine, can yet forecast the Actual.

That is not a loss. It is the Mercury story at the scale of a civilization. Every domain that goes quiet frees the attention that used to maintain it, and pushes the place where the next real surprise appears farther out. In August I wrote that the frontier moves and does not close, and that the complaint is the receipt. When you feel that everything difficult is still difficult, notice what counts as difficult now. It used to be the algebra.

The edge and the face

Two places, I think, will hold human attention as the middle goes quiet.

The first is the edge itself. The anomaly the delegates cannot resolve, the forecast that keeps missing, the customer whose behavior no model expected. These are the places where an Actual is still arriving that no Prediction in the system has absorbed. They are also the places where human Rate, which is extraordinary, earns its keep.

The second is the face. I tell my students to be the face their customers trust. As AI makes every screen, voice, and message cheap to copy, the Actual of a person in front of you becomes scarce and valuable, because it is the one arrival the delegates cannot manufacture. Trust has always rested on predictability: a trusted person is one whose conduct meets your Expectation. Face to face is where that is now proven.

The angle becomes the scarce thing

There is one more consequence, and it follows from Chapter 4.

Intelligence is the real part. As Reach becomes abundant and cheap, as delegates extend the territory of quiet faster than anything has before, the real part stops being scarce.

What stays scarce is the other axis. Not how far Prediction reaches but which way the whole Expectation is turned: the Ideal, the idea that has you, the end that is fixed while the path varies. Delegates can find the path around the wall faster than Romeo ever could. Which wall is worth going around is not a forecast. It is an aim.

That is argument, and I will mark it as such. But I think it is the shape of the next few decades: the real part becoming abundant, and the angle becoming the thing worth attending to. A world of cheap intelligence is a world in which what you are pulled toward matters more than ever.

Chapter 14

No End of the Road

Go to the limit. Imagine an Actualizer whose Prediction met every Actual everywhere, always. Its Reality would be one in every direction. No surprise. No information. No attention. No action.

On Sunday I said such a thing would seem content. “There’s no real reason for it to do anything.” That is exactly right, and it is old. Aristotle, in the twelfth book of the Metaphysics, described a first mover that moves everything without itself being moved: a mind whose thinking is a thinking of thinking, perfect and so without any reason to change. Perfect intelligence, followed all the way out, does not look like a genius. It looks like stillness.

But that limit is not available, and the August series gave the reasons. The optimal compression of the world is uncomputable; no entity of any construction can in general find it. Most of what could happen admits no description shorter than itself. And some of what happens, at the bottom of physics, is not settled in advance at all. On Sunday I said it about the rock and the flash flood: nothing can predict that with real accuracy, because nothing can.

So there is no end of the road. The rock is not at the end of it, and neither is the unmoved mover. Intelligence aims at the quiet number and can never reach it everywhere.

The paradox, resolved

That is what dissolves the paradox of intelligence.

The most intelligent thing in the room yields a non-surprising result and gets no attention. True. But it is not silent everywhere, because nothing can be. It is silent across a territory, and loud at the edge of that territory, and the size of its silence is the measure of its intelligence. Its attention is not absent. It has been pushed out to the frontier, which is where the surprise lives that nothing has absorbed yet.

The rock is silent because it predicts nothing. The deluded are silent because they refuse the Actual. The master is silent in the middle and wide awake at the edge. From across the room they look alike. A sealed forecast tells them apart.

The definition

Here it is again, with everything it now rests on.

Intelligence is the real part of Expectation earning its keep: how far an Actualizer’s Prediction reaches into Actual it has never seen without being surprised, and how fast that reach grows when surprise arrives.

• It lives on the real axis, because Prediction is the only part of Expectation the Actual can prove wrong.
• The Ideal aims it and cannot enlarge it. Goals are real, and perpendicular.
• Where there is no Prediction, intelligence is undefined, not zero. That is the rock, and James’s filings.
• It is measured at sealed Prediction against Actual, on arrivals not yet seen, never inside the Actualizer and never by how calm it feels.
• It comes in two numbers: Reach and Rate.
• No one can raise it by reaching into it. It rises only through the Actuals your actions let in.
• When it works, it vanishes. You will see it only at its edge, or in the gap between a quiet mind and a loud room.

Coda: the third recording

I stopped recording twice on Sunday morning, once at five minutes and once at eighteen. Looking back, both stops happened at exactly the same kind of place. The territory of what I could already predict about intelligence ran out, and I was standing at its edge, surprised by my own argument.

That is where attention goes. It went there on Sunday. It is supposed to.