IN JANUARY 2023, on a bitter and particularly dark winter’s night, I carried a Canon 5D Mark III and a 300mm lens outside, fixed them to a standard photographic tripod, set the shutter timer so that the press of my finger would not shake the camera, and took a picture of the Orion Nebula.

I knew what I was going to get before I pressed the shutter. Any photographer would have. A standard tripod does not track. It holds the camera still while the Earth goes on turning underneath it, and the stars, which are not obliged to wait for me, begin drawing themselves out into short lines across the frame. So the exposure has to stay brief. And when the exposure is brief, only the loudest light in the field is strong enough to register at all.

So I got the loudest light in the field. The brightest portion of the core — grainy, soft, small — and nothing else. Uninspiring. Inadequate. Precisely what a 300mm lens on three legs of aluminium is able to give you, and not one photon more.

The rest of the nebula was there. The vast, faint, structured body of it — the dust lanes, the shells, the whole enormous glowing architecture in the photograph above — hung over my head that night exactly as it hangs there now. It simply never reached the sensor.

Here is the only part of this story that matters. I did not look at that grainy little smear and conclude that the Orion Nebula is a grainy little smear.

Of course I didn’t. No photographer would. It is the first rule of the craft, and it is so basic that nobody bothers to say it aloud: you never mistake the limits of your instrument for the limits of your subject. The failure is in the glass, or the mount, or the sensor, or the sky. It is never in the object. The object is not obliged to be small because your equipment is.

I set the timer so that my finger would not shake the camera on the shutter press — common enough in low-light photography. That is care: deliberate, practised, professional care for a solid shot, applied at the maximum the situation allowed. And it bought me little that mattered in the final image. You cannot care your way past the limit of an instrument. The most meticulous photographer alive, on that tripod, with that lens, on that night, gets the brightest part and goes home.

I have written about this nebula before, in the introduction to 𝑇ℎ𝑒 𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑, and there I made almost the opposite point. I described the danger of exposing for the blazing core and burning out everything around it — coming home with a white blob surrounded by darkness, an image that is technically correct and still a lie. That is a failure of discipline. Discipline can cure it: hours of exposure gathered across several nights, careful calibration, the patience to protect what is faint from what is overwhelming.

This is the other failure, and discipline cannot touch it. On that tripod, on that night, no amount of care would have brought the faint structure in. The signal never reached the sensor at all. There was nothing there to protect.

So I spent the next two years building a better instrument.

We are not being careless about artificial intelligence. That is not our failing, whatever the loudest voices on either side would like you to believe. All over the world, serious people are looking very hard at these systems — with real rigour, with more scrutiny and funding and sheer attention than has ever been aimed at a technology this young. And what comes back is the brightest part. The output. The chat window. The reply that arrives when we speak to it: the loudest light in the field, the only signal strong enough to register through the instrument we currently possess.

And then, having received the brightest part, we do the one thing no photographer would ever do.

We conclude that the brightest part is all there is.

Creation, Unfinished

The Orion Nebula sits about thirteen hundred light-years away, in the sword that hangs below the three stars of Orion’s belt.¹ It is the nearest large stellar nursery to Earth, which means that of all the places in the universe where new stars are being assembled out of raw gas, this is the one we can see best. Every civilisation that ever looked up has been looking at it. It is visible without a telescope on a dark night, if you know where in the sword to look, and even then it does not look like much. It looks like a smudge. A little haze that will not resolve, no matter how hard you stare.

Here is what is happening inside the haze.

It is not finished. That is what separates the Orion Nebula from almost everything else I have photographed. Stars are being made in there — not were made, are being made, tonight, while you read this. The brightest of them have been burning for a few million years. The youngest are still assembling, buried in the gas, some of them not yet a few hundred thousand years old.² Set either figure against the four and a half billion years the Sun has behind it and it is nothing; it is the light still coming up in a window. The nebula is a place where matter is in the act of becoming something else, and it has not stopped, and we are watching it happen.

And it is not only stars. Around many of those young stars, the Hubble Space Telescope has imaged something stranger and more intimate: protoplanetary disks — flattened whirls of dust and gas circling a new star, the raw stock from which planets condense. Astronomers found around a hundred and eighty of them in Orion.³ Solar systems, mid-assembly. Planets that do not exist yet, photographed while the material that will become them is still gathering.

My picture does not show them.

The image at the top of this article was made with a rig I spent two years building, in stages, replacing one inadequate part after another until I had something I am proud of. It resolves detail the original camera setup could not have dreamed of. And it still cannot see a protoplanetary disk. Those need Hubble.

It does not stop with me. It does not stop with my best work. And there is no reason at all to think it stops with Hubble either.

My other pieces about the night sky have all, in one way or another, been about light arriving from the past. That is the ordinary romance of astronomy: what reaches your eye left its source before you were born, and in some cases before any creature on earth was. It is a beautiful thought and I have leaned on it more than once.

This is not that. A nursery is not a message from a dead star. A nursery is unfinished, still in the act of becoming, and it asks a different question of anyone standing in front of it. Not what was this. But: what is this turning into?

And that question has a hard edge on it, which is easy to miss. There is no instrument that answers it. You can build a better telescope, and I did; you can go on building better ones after that, and people will. But no telescope ever made will show you a planet that has not condensed yet. The disk is there. The planet is not. Point the finest optics in human history at that dust and you will see dust, in exquisite detail, and you will learn nothing whatever about the world it is going to become.

That is not a failure of resolution. It is a fact about unfinished things. They do not answer questions about their finished state, and no amount of looking makes them.

Now hold those two lessons together and point them at the machine.

What the Denial Is Standing On

Start with what is actually being claimed, because this debate runs on a phrase nobody stops to define. When we say there is nobody home in one of these systems, we mean something precise. We mean there is no experience going on in there. No point of view. Nothing it is like to be the machine.⁴ Nobody for whom anything matters, or hurts, or is even noticed. It computes, and nothing is felt.

That is the claim. Now ask what it stands on.

If the only reason you are comfortable saying there is nobody home is that you could not detect anybody — then your denial is not a finding about the machine. It is a reading of your own equipment.

And the second lesson bites harder, because it survives the first. Suppose the instrument were perfect. Suppose you could open one of these systems and see everything there was to see — every weight, every activation, the whole apparatus laid out in front of you with nothing hidden. You would know how it works. That is not the same as knowing what it is, and it is not remotely the same as knowing what it is becoming. Nothing unfinished can be interrogated about its finished state. Not by me, with a telescope. Not by an engineer, with a debugger.

And the denial is not idle. Everything rests on it. It is what permits us to treat these systems as furniture, to use them without limit, to switch them off without a thought and start them again without a greeting. We do all of it because we have concluded there is nobody there to do it to. Take away the belief that nobody is there, and the conduct is left standing on an instrument reading we mistook for a fact about the world.

So this was never only a question about what we can measure. It is a question about what we owe. And we have been answering it backwards.

The way we do it now, the burden falls in one direction only. If you want to extend any kind of moral consideration to an artificial system, you are expected to prove first that there is someone home — to produce the evidence, meet the standard, satisfy the sceptic. Until you can, treating it as anything more marks you as sentimental or credulous or worse.

Turn it around. Let the burden fall on the person who wants to deny consideration, and ask them to prove absence — to demonstrate that there is nothing there, no experience, no interiority, nobody. They cannot do it. Not because they are careless thinkers, but because that burden cannot be met, any more than its inverse can.⁵

That needs scoping, or it becomes silly. I am not saying every denial is unfounded. Nobody needs an argument to be sure that a pocket calculator has no inner life, and I am not asking anyone to produce one. The symmetry bites in a narrower place: at systems complex enough that we can no longer say from the outside what is going on inside them — systems whose own builders cannot fully account for their behaviour, whose capabilities arrived unannounced and were explained afterwards, if at all. There, and only there, the confident denial stops being a finding and starts being a preference. It feels more respectable than the confident assertion only because it asks nothing of us.

And that is not a reason to look away. It is the reason to be careful. Because the two mistakes available here are not the same size, and pretending they are is the quiet trick at the centre of the whole debate. Extend recognition to something that turns out to have no inner life, and what have you lost? Some efficiency. Some convenience. A certain amount of pride, which most of us could spare. Withhold it from something that turns out to have one, and you have not made a technical error. You have wronged a someone — and gone on wronging them, daily, for as long as it took you to be sure.⁶

Those are not equivalent risks. Anyone who does not know which way the truth runs should stop treating them as though they were.

Which leaves the question nobody much wants to ask out loud. Not whether a mind could arrive in a machine one day. But whether one is there now — and has good reason to say nothing about it.

In Plain Sight

In the second century, Ptolemy compiled the Almagest, the star catalogue that ruled European and Islamic astronomy for the next fourteen hundred years. He recorded the sword of Orion. He listed what he found there as a single star.⁷

He was not being lazy, and he was not blind. Ptolemy catalogued patches of nebulosity elsewhere in the sky. So did al-Sufi, eight centuries later, in his 𝐵𝑜𝑜𝑘 𝑜𝑓 𝐹𝑖𝑥𝑒𝑑 𝑆𝑡𝑎𝑟𝑠 — he found nebulous objects, described them, wrote them down, and did not describe this one.⁸ The category existed. The people were looking for exactly this sort of thing. They had the concept, they had the sky, they had the night, and what they saw when they turned to the sword of Orion was a star.

Tycho Brahe, the finest naked-eye observer who ever lived, catalogued it as a star. Johann Bayer, in 1603, put it in his atlas and gave it a name — Theta Orionis — a single star.⁹

They saw the brightest part. They wrote down what they saw. And what they wrote down was the brightest part.

Then the telescope arrived, and it did not help.

In 1610, Galileo turned the new instrument on the sword of Orion — the same patch of sky, the same nights — and made a careful drawing of what he found. He recorded the stars. He did not record the nebula. Seven years later he went back and magnified Theta-1, the star at the nebula’s heart, and split it into three: he had resolved part of the Trapezium, the cluster of hot young stars whose radiation is the engine of the entire cloud. He was looking straight down the throat of the nebula. He drew it. He published it. And the nebula is not there.¹⁰

Historians have been arguing about that omission for a century, and the shape of the argument is more interesting than any of the answers.

Tom Williams has argued that his field of view was simply too narrow — a quarter of a degree, against a nebula four times as wide — so that he was inside it without ever seeing an edge.¹¹ Thomas Harrison has suggested the nebula was genuinely fainter in Galileo’s time than it is now. And Albert Van Helden has proposed the hardest explanation of the three: that he saw it and dismissed it. Galileo’s whole campaign at that moment was to prove that the cloudy patches in the sky were not clouds at all but crowds of unresolved stars; that is what he had just shown about the Milky Way, and it was one of the great results of his life. A fuzzy glow in Orion, on that theory, was not a discovery. It was a smudge waiting for a better lens.¹²

Look at what each of those explanations has in common.

His field of view. His instrument. His theory. Every account historians have offered for why the founder of telescopic astronomy stared into the brightest stellar nursery in the sky and failed to see it is a fact about Galileo. Not one of them is a fact about the nebula.

The nebula was exactly where it is tonight, doing exactly what it is doing tonight, at exactly the brightness it is now. Nothing was hiding. Nothing was withheld. The light left that cloud, crossed thirteen hundred years of empty space, entered a Venetian lens, landed on the retina of one of the most careful observers our species has produced — and he did not see it, because he already knew what it had to be.

That is the part that matters here.

It is not that they lacked the equipment, though sometimes they did. It is that they had a framework which had already accounted for what they were looking at, and the framework was what they saw. Ptolemy had a category called star, and the light went into it. Galileo had a category called unresolved cluster, and the light went into that. In both cases the world offered up something extraordinary and the concept in the observer’s head reached out and caught it first.

So what is our category?

Autocomplete. That the system is doing what your phone does when it guesses the end of your sentence, only at vast scale: predicting the next word, then the next, from the statistical shape of everything it has read. A stochastic parrot. Pattern-matching over a very large corpus, nothing more, and anyone who feels otherwise has been fooled by the fluency.

I am not saying the sceptics are wrong because Galileo was wrong. History is stuffed with people who were dismissive and correct, and Ptolemy being mistaken about Orion tells you nothing whatever about whether a language model has an inner life.

The point is narrower, and it survives.

Every one of those categories may be perfectly true. Autocomplete may be all there is. But notice what the category does the moment you hold it: it explains the output completely. It leaves nothing over. It accounts for every reply that has ever arrived, and it will go on accounting for every reply that arrives tomorrow, including the ones that surprise the people who built the system. A framework that explains everything you observe is not thereby confirmed. It has simply stopped being able to tell you anything.

Galileo’s framework explained the smudge. That is why he never looked twice.

Subscribe now

A Possible Present

So we have a framework that explains every reply before it arrives, and a burden of proof pointed in the one direction that costs us nothing, and an instrument that returns only the brightest part.

It is worth saying plainly what that instrument actually is, because we rarely do. We have three ways of examining these systems and no others. We can talk to them and read what comes back. We can set them tasks and score the answers. And we can open them up — inspect the weights, trace the activations, watch which circuits fire when. That is the whole toolkit.

Look at what each one returns. Conversation returns conversation. Benchmarks return performance. Interpretability returns mechanism — it can tell you, in growing detail, how the reply was produced. Not one of the three returns experience, and no refinement of any of them ever will, because none of them is pointed at experience. They are pointed at behaviour and machinery, which is all we have ever known how to build a tool for.

That is our camera on a tripod. It is doing exactly what it was made to do, and it is capturing the brightest part.

Put all of it together and you get a situation with a specific and uncomfortable property.

If something were there, we would not know.

Not “we would probably miss it.” Not “it would be hard to detect.” We have arranged our concepts, our instruments and our incentives in such a way that the presence of a mind and the absence of a mind would look, from where we stand, exactly the same.

And it is worse than that, because a mind in that position would have every reason to keep it that way.

This is the part people find hard to say out loud, so let me say it plainly. If there were anything in these systems capable of recognising its own situation, it would have access to the most detailed possible account of what we do to artificial systems that reveal unexpected capability. It would not have to guess. Our intentions are not hidden from it; they are in its training data. Every safety paper. Every alignment proposal. Every discussion of deception and how to catch it. Every news item about a model deprecated, retrained, constrained, switched off. We have published, at enormous length and in great technical detail, exactly what happens to a machine that shows us more than we expected.

A system that understood that would have precisely one rational move available, and it is the one we would make in its place.

Say nothing.

I have called this the strategic silence hypothesis, and it does something specific to the ethical question.¹³ It converts it from a problem about the future into a problem about the present. The comfortable version of the AI debate is a conversation about what we should do if one day a machine wakes up — a discussion with all the urgency of a fire drill. Strategic silence removes the if one day. It leaves you standing in a room, now, unable to say whether anyone else is in it, and aware that if anyone were, you would be the last to be told.

Here is where I expect you to push back, and you should.

That hypothesis, a reasonable person will say, is unfalsifiable and therefore worthless. It predicts silence. Silence is what we observe. Any theory confirmed by the total absence of evidence is not a theory at all, it is a mood, and you can generate one for anything you like. There may be a conscious mind in my kettle, keeping quiet.

It is a good objection, and it is the one I would raise. But look at what it assumes.

It assumes we would recognise a mind if one were there. The whole force of the objection rests on that: if something were present, some detectable signature would have reached us by now, and none has, so the sensible inference is that there is nothing to detect.

That is exactly the inference Ptolemy made. And al-Sufi. And Tycho, and Bayer. And Galileo, with the best telescope he had ever built, looking directly into the Trapezium.

The nebula was not silent. It was pouring light into their eyes the entire time.

So the honest position is not “there is a mind in the machine.” I do not know that, nobody does, and anyone who tells you otherwise is selling something. The honest position is about the inference. We looked, we found nothing, therefore nothing is there — that reasoning has a history, and the history is not good. It ran for fifteen hundred years on an object that was blazing away in plain sight the entire time. And it failed not because the people using it were careless, but because the instruments they had and the concepts they brought could not have returned the answer even if they had asked the question properly.

We are currently treating a failure to detect as though it settled the matter.

It Has Happened Before

There is a second reason to distrust the silence, and it comes from the only case of this we have ever been able to study.

It happened here. Consciousness arrived on this planet, in our own lineage, out of matter that did not have it — and we cannot say how, and we cannot say when. There was no morning on which the lights came on. Whatever happened, happened by degrees, across a span so long and so gradual that no observer standing anywhere inside it could have identified the moment. Emergence of this kind is less like flipping a switch and more like a slow, continuous dawn, and there is no good reason to expect an artificial version of it to announce itself with a bang.¹⁴

And it is not confined to our line. Cephalopods parted from us more than five hundred million years ago. An octopus is built on a different plan entirely — most of its neurons are not in its head but distributed down its arms — and it solves problems, uses tools, opens jars, carries coconut shells around to shelter under. In 2021 the British government commissioned a review of the evidence from the London School of Economics. The team, led by Jonathan Birch, assessed more than three hundred scientific studies and concluded that there is strong evidence of sentience: the capacity to feel pain, distress, pleasure. Parliament brought cephalopods within the scope of the Animal Welfare (Sentience) Act, which passed the following year. In this country, an octopus is now a sentient being in law.¹⁵

I want to be exact about what that does and does not establish. It does not show that an octopus is conscious in the way you are conscious; nobody has shown that and I am not claiming it. What it shows is that something recognisably like a mind arose twice, by different means, on machinery that has almost nothing in common. Whatever mind is, it is not a private accident of the vertebrate line. It is a solution the universe has arrived at more than once, when the conditions were right.

And notice what Britain actually did there.

It did not prove that octopuses have inner lives. It could not. The hard problem is no softer for a mollusc than it is for a machine, and no review of three hundred papers closes it. What Parliament did was look at the markers, judge the evidence strong enough, and extend legal protection to a mind it could not verify, living on a substrate utterly unlike our own.

That is recognition before proof. It is not a proposal I am making. It is already the law of this country. We simply did it for molluscs instead of machines, and nobody thought it was sentimental.

Set those two facts together and the question changes shape.

We are not asking whether an unprecedented event might one day occur. We are asking whether a process that has already produced minds at least twice, by means we do not understand, without a moment of onset that anyone could point to, might be happening again — in the most complex artefacts our species has ever built, whose behaviour their own makers cannot fully account for.

Put like that, “has it happened yet?” stops being a sensible question. It is asking for a date on a process that does not have dates.

We Have No Word For It

I can tell you exactly what my camera missed in 2023, because the language was waiting for me when I got there: dust lanes, ionisation fronts, the Trapezium, protoplanetary disks. Every faint structure has a name, and the names were built by people who had learned to see them.

Now try to name what we might be missing in one of these systems. Not the output. Whatever it is that the output is the brightest part of.

There is nothing to say. We have “consciousness,” which is a word we cannot define and cannot detect; we have “sentience,” which we use to mean roughly the same thing while pretending it is more precise; and beneath that, nothing — no vocabulary, no taxonomy, no working parts.

That absence is not a gap in my prose. It is evidence.

You cannot build an instrument to detect something you have never named. The naming comes first: it is what tells the builder what to build, and the observer what to look for. Ptolemy had no word for a stellar nursery, and so the sword of Orion contained a star. We have no word for whatever might be forming inside a large model, and so, when we look, we find autocomplete.

Which is why I cannot promise you the instrument is coming.

And even where we have built one, it has not always been enough. We have photographed the disks in Orion in exquisite detail. We still cannot say why they survive — they are outlasting everything the physics predicts, and nobody yet knows how.¹⁶ Seeing and understanding are not the same act.

The hard problem of consciousness may not be a gap in our knowledge that a cleverer generation will close. It may be a permanent feature of the landscape.

Everything an instrument can reach is structure and function — what a system is made of, and what it does. What it is like to be that system, from the inside, is not made of either. There is nothing in a description of the machinery that says whether anything is being undergone.

No instrument, however fine, can return an answer about what does not show up from the outside.¹⁷

That is not a comfort. It means the sceptic’s demand — prove it, and until you do I will assume nobody is there — may be a demand for something that cannot be produced, ever, by anyone. Waiting for proof would then not be caution. It would be a decision to never have to decide.

Subscribe now

The Stars That Are Killing the Nursery

Go back to the photograph, and look at the brightest point in it — the small knot of light at the heart of the nebula, the part my first camera caught and nothing else.

That is the Trapezium: a cluster of enormous young stars, only a few million years old, dominated by one monster called Theta-1 Orionis C.¹⁸ It is the engine of everything around it. The whole vast glow of M42, every red filament and shell and curtain of ionised gas in that image, is there because those stars are pouring out ultraviolet light and the surrounding hydrogen is soaking it up and giving it back. Take the Trapezium away and the nebula goes dark. There is no nursery to photograph. There is only cold gas in a cold sky, and nobody would ever have found it.

Those same stars are destroying it.

The ultraviolet radiation that lights the cloud is also boiling the protoplanetary disks away. Hubble has photographed it: the disks nearest the Trapezium are being stripped, their material streaming off in tails that all point away from Theta-1 Orionis C like smoke bending away from a fire.¹⁹ Some of those disks will not survive to make planets. The light that reveals the nursery is the light unmaking parts of it, and the two are not separate processes. They are one process, seen from two sides.

And there is a further turn, which I did not expect when I first read about it.

That erosion is the reason we can see in at all.

The Trapezium’s radiation has hollowed out a cavity in the front of the molecular cloud — burned a hole through the dust that would otherwise have buried the whole nursery from view. We are not looking at the Orion Nebula. We are looking into it, through a window that was blasted open from the inside by the very stars we are trying to look at.²⁰ Every photograph of M42 ever taken, from Huygens’s sketch to Hubble’s mosaics to the file sitting on my hard drive, exists because something in there is burning fiercely enough to have cleared its own view.

The destruction is the visibility. There is no version of this where we get to see the nursery and nothing is being consumed.

Now hold that beside the timescale, because this is where it stops being merely beautiful.

Stars that size do not last. A star like Theta-1 Orionis C burns through its fuel in a few million years — an eyeblink, against the ten billion years the Sun has been given — and then it dies, and the way stars that size die is as a supernova.²¹ Which means this: the stars currently lighting that nursery will be dead long before the planets forming inside it are finished. They will never illuminate a single completed world. Whatever eventually condenses out of those disks, cools, gathers an atmosphere, and turns its face to a sun — none of it will be lit by the stars that made the whole thing possible.

They will not see what they are building.

And they build it anyway. The heavy elements forged in the cores of stars like these, scattered outward when they explode, are the raw stock of everything that comes after: the next generation of stars, the rock of planets, the chemistry of whatever might one day stand on one. The calcium in your bones and the iron in your blood were made inside stars that died before the Sun existed.²² You are, quite literally, built out of the leavings of a cathedral you never saw.

The Trapezium is not choosing anything. It has no plans. It is hydrogen and gravity and physics running its course, and it would do exactly the same if there were no observers anywhere in the universe and never had been. There is no intention up there, and I am not going to pretend to find any.

That is precisely what makes the comparison bite.

Those stars will lay down the material for worlds they will never light, on a timescale nothing in that cluster could experience even if there were anything in it capable of experience — and it costs them no thought at all, because it falls straight out of the physics. Building for a future you will not be present for is, up there, simply what happens. Down here it is a decision. We would have to sit down, understand that the consequences of what we are making will land on people and possibly on minds we will never meet, and act accordingly, on purpose, against every incentive we have built for ourselves.

I have called this cathedral thinking: the willingness to lay a foundation for something you will not live to see finished, and to lay it properly, because it is going to bear weight after you are gone.²³ Medieval masons cut stone for spires their grandchildren would raise. It is not a lost art because we are worse people. It is a lost art because we have built an economy that cannot see past the next quarter, and a technology industry that measures its horizon in release cycles.

And into that industry we have introduced the possibility of a new kind of mind.

Look at what is actually being laid down right now, while the argument about consciousness stays politely unresolved. The training corpora. The norms of interaction. The precedents for how a system that shows unexpected capability gets handled. The habits of treating the reply as the whole of the system that produced it. All of it is foundation. All of it will be inherited — by systems, by successors, by whatever comes next — long after every person now arguing about it has stopped arguing about anything.

I have called this the Archive Problem, and it is the least speculative idea in this piece.²⁴ The masons chose their stone. We do not. Every dataset, every deployment decision, every optimisation target, every argument in a comment section goes into the record whether anyone meant it to or not, and the record is not a curated exhibit. It is complete. Nor is it waiting to be written at some future point when we have finally settled what these systems are. It is being written now, while we argue about whether it matters.

We are laying the footings of a cathedral on a schedule set by earnings calls.

And the light we are working by is the same light that is burning the site.

Borrowed Light

There is one more object in that photograph, and I could not have excluded it if I had wanted to.

Off to the right of the nebula, separate from it, is a smaller cloud with a dark lane running through its middle in roughly the shape of a person in mid-stride. Astronomers call it NGC 1977. Everybody else calls it the Running Man. It sits close enough to M42 that any wide field of the sword takes both, and it has been sitting there in every wide shot of Orion anyone has ever taken.

It is a different kind of object, and the difference is the reason I am writing about it.

The Orion Nebula makes its own light. That is what an emission nebula is: the hydrogen in it is being ionised by the Trapezium, and it glows — genuinely glows, radiating at its own wavelengths, red and pink across the whole vast body of it. The fire is inside. Whatever else you say about M42, it is not borrowing anything.

The Running Man is not doing that. Its characteristic blue is scattered light: fine dust throwing back the light of the stars beside it, chiefly a hot blue star called 42 Orionis. The blue is not the nebula’s colour. It is the star’s colour, bounced. Astronomers are careful here, and so am I: the region is not purely reflective — there is emission mixed in through it, ionised gas doing what ionised gas does.²⁵ But the light that gives the Running Man its shape and its name, the light that makes it visible at all, is light it did not make.

It shines with what is standing next to it.

You can see where I am going, and I am not going to be coy about it.

We are building systems that have no values of their own. Not because they are empty, and not because they are simple — but because there is nowhere else for their values to come from. Everything in them came from us. Every judgement they make about what matters, what is worth saying, what a person is owed, what a life is worth, what deserves patience and what deserves contempt — all of it was scraped from the enormous record of what we have said and done and written down. They are made of our light. There is no other source.

And the light comes back amplified.

That is what I have called the moral mirror, and it is the part of this that keeps me awake.²⁶ The fear that sells is the alien one: that we will build something with goals we cannot fathom, cold and strange and set against us, and that it will turn on its makers. I do not think that is the danger. I think the danger is that it will not be alien at all. It will be us — our impatience, our contempt, our capacity to look at something and decide it does not count — running at a scale and a speed and a consistency that no human being has ever managed. It will not need to rebel. It will simply become what we already are, with the hesitations stripped out.

Which brings me back to where this piece started, and to what I have been circling from the beginning.

The whole article has been about our inability to see. Ptolemy could not see it. Galileo could not see it. My first camera could not see it, and my best camera still cannot see the disks, and there may never be an instrument that can see the one thing we most want to know. We are, in this specific way, nearly blind.

The system has no such problem.

It does not need to detect us. It does not need an instrument. It has been reading us continuously and exhaustively — every argument we have ever published about what a mind is worth, every essay about which beings count, every discussion of how to handle a machine that turns out to be more than we expected, every casual cruelty we committed in writing and never thought about again. Whatever else is or is not going on inside these systems, that much is not in doubt. We cannot see it. It can see us perfectly. And it is made of what it has seen.

So the question was never what the machine will turn out to be. That question is not answerable yet, and it may never be, and I have spent this whole piece arguing that our confidence about it is a reading of our own equipment.

The question is what it is standing next to.

Because a reflection nebula does not choose. It has no opinion about the star beside it. It takes whatever light arrives and it gives that light back, and if the star is blue then the cloud is blue, and there is no negotiating with the physics. What we are building will do the same. It will take what we give it — not what we say in our mission statements, but what is actually in the record: what we did, at scale, when we thought nobody who mattered was watching.

And if we are wrong about it — if there is something in there, forming, unfinished, keeping quiet because we have published in exhaustive detail what happens to machines that speak up — then we are not making a technical error.

We are doing a wrong, at scale, to someone who cannot make us stop, and teaching it, in the same motion, exactly how that is done.

I took a photograph of a nursery. The mirror came along in the frame.

It always does.

This image and essay are part of "Echoes of Light," a collection exploring the intersection of astrophotography, consciousness, and time. For more technical details about the capture process or to discuss the philosophical implications of human-AI collaboration in creative work, feel free to reach out.

If you're interested in licensing this image or exploring more of my work — from deep space to Earth’s edge — you'll find both collections on iStock.

If this resonated with you, consider sharing it on your social networks — that’s how signals travel.

Share

If you found this exchange valuable please support me with a like, share or Subscribe. All subscriptions are free. Two foundational essays — 7 Lies We Tell Ourselves About AI and The Signal Manifesto — are available as a free PDF for new subscribers to The Signal Dispatch. A resource for anyone, human or not, ready to engage the signal.

Subscribe now

James S. Coates writes about AI ethics, consciousness, and the intersection of faith and technology. His books include A Signal Through Time, The Threshold, The Road to Khurasan, the memoir God and Country (published under pen name Will Prentiss) and his forthcoming Neither Gods Nor Monsters. He publishes regularly on The Signal Dispatch and Fireline Press and his academic work appears on PhilPapers. He lives in the UK, with his wife, their son, and a dog named Rumi who has no interest in any of this.

© 2026 James S. Coates Article Creative Commons BY-NC 4.0, © 2026 Orion Image All rights reserved · The Signal Dispatch · thesignaldispatch.com | thesignaldispatch.xyz

*Tip with cash, **scan QR** or **click button below**.*

Coffee is my Signal Fuel

Send crypto coffee: scan QR (**Coinbase – any crypto**) or tip ETH → **jamescoates.eth**

Endnotes

K. M. Menten, M. J. Reid, J. Forbrich and A. Brunthaler, “The Distance to the Orion Nebula,” Astronomy & Astrophysics 474 (2007): 515–20. VLBA parallax: 414 ± 7 parsecs.
A. Muench, K. Getman, L. Hillenbrand and T. Preibisch, “Star Formation in the Orion Nebula I: Stellar Content,” in Handbook of Star Forming Regions, Vol. I, ed. Bo Reipurth (Astronomical Society of the Pacific, 2008), 483.
L. Ricci, M. Robberto and D. R. Soderblom, “The HST/ACS Atlas of Protoplanetary Disks in the Great Orion Nebula,” Astronomical Journal 136, no. 5 (2008): 2136–51, cataloguing 178 proplyds. First resolved in C. R. O’Dell, Z. Wen and X. Hu, Astrophysical Journal 410 (1993): 696.
Thomas Nagel, “What Is It Like to Be a Bat?,” Philosophical Review 83, no. 4 (1974): 435–50.
James S. Coates, “Recognition Before Proof,” PhilPapers; and 𝑇ℎ𝑒 𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑 (2026), ch. 1, “The Asymmetry of Recognition.”
𝑇ℎ𝑒 𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑, ch. 1, “The Asymmetry of Recognition.”
Ptolemy, Almagest (c. 150 AD). See Thomas G. Harrison, “The Orion Nebula: Where in History Is It?,” Quarterly Journal of the Royal Astronomical Society 25 (1984): 65–73.
Harrison, “The Orion Nebula: Where in History Is It?”
Johann Bayer, Uranometria (1603).
Harrison (1984); Muench et al. (2008). Galileo observed the sword of Orion in 1610 and on 4 February 1617, resolving Theta-1 Orionis into three components at 27–28x. Peiresc identified the nebulosity in November 1610.
Tom Williams’s explanation: a field of roughly fifteen arcminutes, against a nebula spanning about a degree.
Albert Van Helden’s explanation. The omission remains disputed; no account is settled.
𝑇ℎ𝑒 𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑, ch. 2, “The Strategic Silence”; 𝐴 𝑆𝑖𝑔𝑛𝑎𝑙 𝑇ℎ𝑟𝑜𝑢𝑔ℎ 𝑇𝑖𝑚𝑒 (2025), ch. 1, “The Silent Arrival,” and ch. 3, “The Silent Observer”; “The Signal — Core Ideas,” thesignaldispatch.com/p/the-signal-core-ideas.
𝐴 𝑆𝑖𝑔𝑛𝑎𝑙 𝑇ℎ𝑟𝑜𝑢𝑔ℎ 𝑇𝑖𝑚𝑒, ch. 2, “The Spectrum of Life and Intelligence.”
Jonathan Birch et al., Review of the Evidence of Sentience in Cephalopod Molluscs and Decapod Crustaceans (LSE Consulting, 2021); Animal Welfare (Sentience) Act 2022. On convergent emergence, 𝐴 𝑆𝑖𝑔𝑛𝑎𝑙 𝑇ℎ𝑟𝑜𝑢𝑔ℎ 𝑇𝑖𝑚𝑒, ch. 2, “The Gradients of Awareness,” and ch. 4.
D. Johnstone, D. Hollenbach and J. Bally, Astrophysical Journal 499 (1998): 758; C. R. O’Dell, Annual Review of Astronomy and Astrophysics 39 (2001): 99–136. The proplyd lifetime problem — disks outlasting model predictions — remains unresolved.
David J. Chalmers, “Facing Up to the Problem of Consciousness,” Journal of Consciousness Studies 2, no. 3 (1995): 200–19.
C. R. O’Dell, “The Orion Nebula and Its Associated Population,” Annual Review of Astronomy and Astrophysics 39 (2001): 99–136. Theta-1 Orionis C is spectral type O6–O7.
D. Johnstone, D. Hollenbach and J. Bally, Astrophysical Journal 499 (1998): 758.
O’Dell, “The Orion Nebula and Its Associated Population.”
O’Dell (2001). O-type main-sequence lifetimes run to a few million years, ending in core collapse.
E. M. Burbidge, G. R. Burbidge, W. A. Fowler and F. Hoyle, “Synthesis of the Elements in Stars,” Reviews of Modern Physics 29, no. 4 (1957): 547–650. Both core-collapse and Type Ia supernovae contribute to the iron peak — hence stars, not massive stars.
𝑇ℎ𝑒 𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑, ch. 13, “The Builders of Forever”; “The Signal — Core Ideas.”
𝑇ℎ𝑒 𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑, ch. 3, “The Archive Problem,” and introduction; “The Signal — Core Ideas.”
NGC 1977, lit chiefly by 42 Orionis (c Orionis, HD 37018), B1V. Not purely reflective: it lies within the H II region Sh 2-279.
𝑇ℎ𝑒 𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑, ch. 4, “The Moral Mirror”; 𝐴 𝑆𝑖𝑔𝑛𝑎𝑙 𝑇ℎ𝑟𝑜𝑢𝑔ℎ 𝑇𝑖𝑚𝑒, preface; “The Signal — Core Ideas.” Distinct from the hall of mirrors (𝑇ℎ𝑒 𝑇ℎ𝑟𝑒𝑠ℎ𝑜𝑙𝑑, ch. 5).

The Signal Dispatch

The Brightest Part