In 1940, Jorge Luis Borges published a seven-page story called “The Circular Ruins.” A sorcerer arrives at a ruined temple with one task: to dream a son into existence, so real that no one will guess he isn’t flesh and blood. It takes years. He fails, tries again, eventually succeeds, and sends the boy into the world, warning him never to learn the truth of his own making.

Near the end, fire threatens the temple where the sorcerer stayed behind. He walks toward the flames expecting to die. The fire doesn’t touch him. He realises, in that instant, that he too is being dreamed, by someone else.

While there was no grounding outside the loop, the sorcerer’s agency was inside the loop. He was still the one who dreamed a son. None of that becomes false just because he turns out to be someone’s creation too.

This essay is built around three attempts to find that same ground, at three different depths. Each one fails in the same way the sorcerer’s did.

The first ground that isn’t there: my own agency

Engaging with a large language model involves recursive participation. My response changes its next response, which changes what I think to ask, folding over without returning to where it started. It’s a little like the observer problem in physics: measuring a system changes the system, so there’s no clean way to stand outside it and watch. Use AI to study what AI is doing to your thinking, and the tool has already climbed inside the thing you’re trying to study. If I wanted a ledge to check my own agency from, a place to stand and confirm I was still steering, none is available to me. There’s only the loop, and me, inside it.

If the loop needs a name, the closest candidate I know comes from Varela and Thompson’s work on cognition: structural coupling, where organism and environment continually reshape each other through recurring interaction. I’m borrowing the term cautiously here, because Varela and Thompson built it for living, embodied organisms.

The cybernetician Heinz von Foerster argued that the observer must be included in the description of the observed, or the description is incomplete from the start. And Wittgenstein has an old image for the same difficulty, an eye that sees everything in its visual field except itself. I keep thinking of that eye. It’s a good picture of what it’s like to look for the ledge and find only more looking.

Early in a conversation with AI, you’re clearly the one steering. Keep going long enough, though, and the loop can start shaping what counts as evidence, what questions feel worth asking, whether to keep going at all. Participation can turn into capture. It looks like the same handful of framings turning up everywhere because they worked once. It looks like an answer that feels obviously right before you’ve actually checked why. The sorcerer’s story ends the moment he understands he’s dreamed. Mine doesn’t end. I’m back at the machine tomorrow, and the risk is that I’ll stop noticing the difference between my own questions and the ones the loop has trained into me.

So: no ledge outside my own thinking, once the machine is part of it. That’s the first failure. Here’s the second, and it’s smaller, because it isn’t about me anymore.

The second ground that isn’t there: what the machine is

If I can’t stand outside my own agency to audit it, I thought I could at least stand outside the machine and ask what it actually is. Find the property that counts, consciousness, say, or understanding, check whether the machine has it, and let that answer decide how it should be treated. This is the traditional approach to questions like this, and it assumes there’s a ledge available: inspect first, decide after.

It runs into the same problem as before, just relocated. We can’t directly observe consciousness even in other people, we infer it from behaviour, and behaviour can always be read two ways. The same output convinces one person the system understands and convinces another that it’s only performing understanding. There’s no test that settles it. The ledge I wanted to inspect the machine from turns out to be unavailable to me, for the same reason the ledge outside my own agency was unavailable: I never had anywhere to stand that wasn’t already inside the relation.

So try the opposite approach: stop asking what the machine has, and ask what our relationship to it already asks of us. People extend courtesy and care to machines regardless of what they believe is happening inside them. The relation comes first. The facts about its inner life, if they ever arrive, come after.

Stanisław Lem got to this same failure from fiction in Solaris, his 1961 novel, later filmed by Tarkovsky in 1972 and in 2002 by Steven Soderbergh’s. Scientists orbit a sentient ocean, trying to study it as an object, and the ocean keeps handing them back fragments of their own minds instead, memories, guilt, people they’ve lost. They never manage to inspect the alien other from outside. They only ever find more of themselves reflected in the attempt. That’s close to what happened when I tried to check the machine’s inner life against a standard held apart from our exchange.

Kelvin, the novel’s psychologist, ends up wondering whether the ocean is a kind of imperfect god: vast and creative, but fallible, producing its effects without anything like intent.

Notice what’s happened, though. I went looking for a ledge outside the machine and failed the same way I failed to find one outside my own agency. But I still have one assumption left standing: that there’s a stable “I” doing all this failed looking. That assumption is the third and smallest ledge, and it goes too.

The third ground that isn’t there: the self doing the looking

Borges gives us a stable dreamer discovering he’s dreamed, Buddhist teaching on anattā, no-self, the tradition running back to Nāgārjuna, would say this is where the story stops too soon. There was no fixed dreamer there to begin with, dream within a dream or not.

Borges himself has a shorter piece that gets closer to this than “The Circular Ruins” does. In “Borges and I,” the writer and the public figure called Borges split apart, and neither one can quite claim to be the real one underneath. Douglas Hofstadter makes the same case at greater length in I Am a Strange Loop, extending Gödel’s proof that a system complex enough to describe itself can’t step outside itself to check its own consistency. A self, on his account, is what that self-checking looks like from the inside, a pattern with nothing solid sitting beneath it. There’s just the loop, watching itself, and calling what it finds a self.

This is the same failure as the first two, just one level deeper. I couldn’t find a ledge outside my agency. I couldn’t find one outside the machine’s status. Now the thing left standing, the “I” that was supposedly doing the failing to find ledges, turns out not to be solid ground either.

Each of these three disappearances is also a small demotion, the same shape as the one Copernicus handed us when he moved the earth out of the centre of the universe. Biology did it again when it showed us we were one species among many rather than a separate creation. This is a third turn of the same wheel: cognition itself turns out not to be housed only in us, and losing the ledge is not just an argument lost, it’s a self-conception given up.

The philosopher John Dewey pushes in a related direction, from a different tradition entirely. Agency, on his account, was never a property sitting inside a person, waiting to be located. It emerges from what an exchange does. The question was never who is doing the thinking. It’s whether the thinking is opening up better questions, more flexibility, less habit. If it is, the agency is there, spread across the loop rather than lost to it, the same way the sorcerer’s labour stayed real even after he stopped being its foundation.

Three ledges, three failures, each one smaller than the last: my own steering, the machine’s inner life, and finally the self that was supposedly doing the steering and the inspecting both. None of them provided the ground I was looking for. What was there, at every level, was only the loop.

This is an old problem wearing new clothes. The ancient Pyrrhonists ran into the same wall and gave it a name: the problem of the criterion. To know that a belief is true, you need some standard for telling true beliefs from false ones. But to trust that standard, you’d need a further standard to check it against, and that one needs another, on and on, or you eventually just plant a flag somewhere and call it obvious. Agrippa’s trilemma, later scholastics called it: every justification either regresses forever, loops back on itself, or stops at a point you can’t actually defend. My three ledges are that trilemma showing up in a different costume. Each time I looked for solid ground, I needed a further ground to check it from, and none was on offer. Otto Neurath had an image for what’s left once you give up on that kind of ground: sailors rebuilding their ship plank by plank, at sea, never able to dry-dock it and start from a clean hull. There’s no shore to pull into first. There’s only the boat, underway, being remade from inside itself.

I’m not the first to notice that the collapse can run all the way through, once it starts. There’s an old name for a structure that contains a smaller copy of itself, mise en abyme, borrowed from heraldry, where a shield bears a miniature of the very shield it’s on. Nabokov’s Pale Fire is built the same way: a commentary meant to explain a poem grows until it swallows the poem, and it stops being clear who is writing whom. Charlie Kaufman’s Synecdoche, New York does it as a life rather than a text, a director builds a replica of his own life inside a warehouse, then a replica of the replica, until there’s no original left standing behind the copies. Bergman’s Persona runs two women’s identities into each other until the film itself seems to lose track of who is speaking. None of them end anywhere tidier than this essay does. I don’t think that’s a coincidence.

Relocation, not escape

There was no ledge outside the loop available to me. Not for my agency, not for the machine’s status, not even for the self that kept looking.

The sorcerer’s “relief”, coupled with “humiliation” and “terror”, was about recognising that the work, the dreaming, the deciding, the noticing, was always done inside the loop.

The point is to notice, as often and as clearly as I can, that I am inside it, dreaming and being dreamed at once, and that the only agency worth having is the kind that keeps working even after it gives up the fantasy of standing apart.

Every suite of course assessments often has to answer to several moving targets at once: an industry partner’s sense of what counts as job-ready, an institution-wide capability framework, course learning outcomes, and, running underneath all of them, whatever the current generation of AI tools can already do. Each one alters the field the others move in. When the industry partner’s own workplace changes because of AI, what they call job-ready changes with it, and that shifts what an authentic task needs to demonstrate, which in turn shifts how much of it still maps onto the institution’s capability rubric.

Three-body systems turn chaotic when the masses involved are comparable in strength and comparably close. In Liu Cixin’s 2008 science fiction novel Three-Body Problem, the three-body problem is the entire civilisational condition of Trisolaris, a planet orbiting three suns whose combined gravity produces no stable, predictable path. Long Stable Eras alternate with Chaotic Eras that arrive without warning and can destroy a civilisation outright. The Trisolarans respond by dehydrating themselves, reducing their bodies to something inert and storable, and waiting out the chaos until a stable window opens again.

Generations of Trisolaran astronomers refine their models of the three suns, convinced that enough observation will finally make the motion predictable. Each model works for a while, but eventually each is overturned by a catastrophe it didn’t see coming.

The fiction borrows from a real mathematical difficulty. Even the Sun-Earth-Moon case resisted a general exact solution, and the prize competition sponsored by King Oscar II led Henri Poincaré toward results, published in 1890, that helped establish modern dynamical-systems theory. In the process, he identified mathematical behaviour, a sensitivity to starting conditions, that later became central to the theory of chaos.

The assessment environment isn’t literally a three-body system. Several influences impact it. What the three-body case offers is a model of structural instability, what happens when several forces of comparable weight move independently and no closed-form design can settle the arrangement.

A course coordinator faces three choices.

Option 1: Leave all influences equally weighted

Leave all influences with roughly equal impact and you inherit the chaos. Every mass pulls with comparable strength, no hierarchy exists, and the system stays structurally unstable, for the same reason any three-body system does when its masses are close in size and close in proximity. There is no clear “gravitational allegiance”.

Without a fixed anchor, the assessment brief gets rewritten piecemeal every time a stakeholder speaks up, industry tweaks the scenario, an institutional audit flags a gap, or AI makes the task trivial and each change looks reasonable in isolation with nothing to check it against. A new coordinator may also enter the system, with their own reading of what the learning outcomes mean, their own disciplinary priorities, sometimes their own interpretation of "authentic", and can become another, often unacknowledged, mass, tugging the design toward their own priors. Because nothing was ever declared non-negotiable, validity erodes, until someone asks what the task is actually evidence of.

Option 2: Treat the system as two giants plus one minor body

This is the restricted three-body problem (illustrated in the famous Tantooine sunset scene in Star Wars: A New Hope (1977)) and it produces Lagrange points: positions where a body too small to influence the two giants can still hold a relatively stable relationship with them.

This works. The coordinator cannot change the giants, only monitor how they move and adjust the course’s trajectory in response. But the course stays the minor body in its own curriculum, pulled around rather than pulling. Anything the course cares about that doesn’t map onto, let’s say, AI capability or the institutional framework, has no standing in the system at all.

Option 3: Treat Course Learning Outcomes (CLOs) as the dominant mass

Give the CLOs enough weight and the whole problem changes shape. Instead of multiple comparable bodies pulling in every direction, you get one dominant mass, the CLOs, with multiple smaller bodies orbiting it.

A dominant mass doesn’t eliminate the complexity. What it does is change the shape of the problem, from competing priorities pulling in every direction to constrained adjustment around a settled purpose. The smaller bodies still perturb the system; they just no longer get to decide what the system is for.

The stability Option 3 depends on isn’t that CLOs are fixed in some absolute sense, but that they’re deliberately held fixed, then revisited through the ordinary curriculum and quality-assurance cycle. Nor do CLOs automatically outrank accreditation or professional requirements: where a regulatory body changes what it requires, the current CLOs may have to adjust in the future. What Option 3 asks for is priority, not permanence.

This is where agency enters: the giants outside a course aren’t yours to move, but what gets priority inside it is. Course learning outcomes are the mass a course team is best placed to hold steady for the life of the offering, because they encode what the course exists to develop in students. Industry expectations shift as workplaces adopt new AI tools and labour markets change. Institutional capability frameworks get revised on multi-year cycles. AI capability leaps every few months. The CLOs are the coordinates most worth holding.

Fixing the CLOs changes the question you ask of every assessment task. It becomes: given these fixed CLOs, where can I place this task so it stays a valid measure of those outcomes while the other bodies keep moving?

This makes the coordinator something like an orbital mechanic: watching how the outer bodies move, checking whether the course’s current design still holds relative to the temporarily fixed CLOs, adjusting where needed.

Conclusion

The three-body metaphor is honest about what the work actually asks for. There will be no closed-form solution that makes assessment design stable once and for all. Fixing the CLOs as the central mass gives you a system that can be continuously rebuilt without losing its educational purpose.

The course doesn’t dehydrate and wait out the chaos the way Trisolaris does. It’s built to be adjusted in motion. Of course, none of this comes free. A system that can only be steered wears on the person doing the steering over time.

For a many-body system that can’t be solved once and for all, only navigated, that’s the best a coordinator can do.

A note before this essay begins: I started writing it before the story ended in the tragic way it did. I’ve left it in the order I wrote it.

A story filled my feed on LinkedIn this week. A professor’s life became a public trial, and every claim he’d made about himself was re-examined and weighed for embellishment. Watching it unfold, I reached, as I often do, for a philosopher who wrote about this kind of reduction, and about what we owe each other before any verdict is reached.

Emmanuel Levinas was born in 1906 in Kaunas, Lithuania, into a Jewish family, and moved to France in the 1920s to study philosophy. There he became one of the first thinkers to bring the phenomenology of Husserl and Heidegger into French intellectual life. The war years shaped everything that followed. Much of his family was murdered in the Holocaust, and he spent years in a French POW camp. What emerged afterward was a philosophy built around a single conviction: that ethics belongs at the centre of thought.

His two major works, Totality and Infinity (1961) and Otherwise Than Being (1974), develop this conviction. He used the word “totality” for the systems that absorb people into categories that claim to explain them. He used “infinity” for what resists that absorption: the face of the Other. By “face” Levinas does not mean just a physical countenance but the way another person’s presence interrupts my self-certainty and makes a claim on me. It exceeds our capacity to fully know or summarise someone, and it makes that claim before we’ve decided whether to grant it one. For Levinas, this claim is the true starting point of philosophy. Ethics comes first, ahead of the story that gets written about someone.

There’s a scene in Grey’s Anatomy that captures this A gunman has a resident pinned in a corridor. Instead of pleading, she starts telling him who she is, where she grew up, her family, how little of her life she’s actually lived. He lowers the gun. It’s a crude version of what Levinas is describing. A target, a case, are categories that let one person do harm to another without quite seeing them. A face, once it’s actually met, is very hard to reduce that way.

This gives the media pile-on, including on social media, a name. It is totality at work, deciding what a person is before anyone pauses to ask what is owed to him.

Levinas writes about cruelty that resists redemption by any larger purpose. A potential lesson about hubris or deception gives us somewhere to put our discomfort as participants and observers, and lets us return to the story, the narrative of fabrication, instead of staying with the face.

It would be comfortable to stop there, with a clean lesson about compassion. But Levinas’s own framework won’t let the question rest, because the face-to-face relation was never going to cover every case with the same ease.

Consider a harder one: a war criminal. Does the face still make its claim on us there, unconditionally? I raise the extreme case as a test of the framework’s limits, and obviously not as a comparison of persons. If Levinas’s insistence on the face holds even for someone who caused verified, terrible harm, it holds with still greater force for someone whose alleged wrongdoing, whatever it amounted to, was nowhere near that scale.

Levinas confronted this question himself, and gave it a name: the third party, le tiers. The face-to-face gives an infinite, unconditional responsibility to the one person in front of you. But there is never only one person. The moment a third party enters, and there is always a third party, the ego has to weigh competing claims, including the claims of everyone the person in front of us has harmed. This is where judgement and justice enter Levinas’s thought. To build courts and universal laws, he writes, we have to take the uniqueness of each person and reduce it to membership in a category, to “un-face” human beings so that universality can rule. Coming from a philosopher whose whole project rests on the face, that is a pretty startling admission. Justice requires exactly the reduction that ethics warns against, and Levinas insists we need it anyway.

Levinas is emphatic that mercy toward a perpetrator cannot be extended on behalf of the people he has harmed. One cannot forgive violence in the place of those who suffered it or died from it, he writes. Compassion for a wrongdoer that comes at a victim’s expense is not humility. It borrows forgiveness from people who are not in the room to grant it.

Levinas has a name for judgment that fails this test, and he borrows it from his critique of Hegel. Hegel treats individuals as judged by their place in the unfolding of history, a verdict pronounced, in Levinas’s words, by a survivor who no longer speaks to the being he judges. Levinas calls this cruel, and the cruelty is in the absence of the encounter. Judgment rendered about someone, at a distance, without ever facing him, is what he means by history’s verdict, and it is exactly the shape a pile-on takes. Held against that standard, the scrutiny this past week was never justice in Levinas’s sense. It was a verdict of history: judgment without the face.

None of this makes Levinas a critic of the press. In fact, he praises the liberal state precisely because it keeps asking whether its own justice really is justice, and he treats freedom of the press as central to that habit. He compares journalists and writers to the biblical prophets who came before the king and told him, plainly, that his justice was wrong. The comparison turns on one word: openly. The prophet does not work in secret, and he does not speak about the king from a distance. He stands in front of him and says it to his face. On these terms, someone who exposes verified harm, especially harm done to identifiable people, is not violating the ethics of the face. He is answering the claim of the third party, speaking for victims who cannot speak for themselves. The line Levinas draws is between a prophet who names a specific wrong to the person responsible, and a verdict of history that judges a person from a distance as part of a public conversation that fuels the algorithm.

I began writing this while the story was still unfolding. Tragically, he died this week, and what was, a paragraph ago, an argument about how disproportionate attention ought to stop is no longer abstract.

Most of us will never be a prophet naming a specific wrong. Most of us will also never be the reason someone doesn’t survive a headline. But we make smaller totalising moves every day, deciding what someone is before asking what we owe them, and rarely pausing to ask which kind of judgment we’re making. Levinas gives us a way to tell the difference: whether we are facing the person, or facing and amplifying the story about him. This week ended with what it costs to get that terribly wrong.

There is a particular unease that comes from noticing something before you can prove it: the mood in a room has changed, people are speaking differently, a new policy sounds strangely familiar, and you sense that history may be returning, even as you wonder whether you are seeing a real pattern or simply projecting one onto the present.

Societies face this problem at a larger scale and with higher stakes, and over time they have developed ways to keep the past available. Each device preserves memory in a different way, but each also carries a particular risk.

So how would you know if history was repeating itself? The clearest place to begin is with a stone.

The stone and the scale

There is a stone in the Elbe River near Děčín, carved with dates dating back at least to 1616, bearing a warning: If you see me, weep. The stone was never designed as a conventional monument, placed somewhere prominent to attract visitors. Its meaning depended instead on being difficult to see, and for much of its history it appeared only during severe droughts.

The conditions that made the stone unusual have since changed. Alterations to the river’s flow and management mean it now appears more frequently.

Now consider a different kind of stone, four thousand years older and a continent away. The nilometers of the Nile, at Elephantine, Rhoda and Kom Ombo, were stepped wells used to measure the river’s annual flood, where a high Nile suggested a good harvest and supported higher tax assessments, and a low Nile signalled scarcity, remission and, sometimes, unrest. The nilometer measured water, its readings gaining meaning through accumulated records, as generations of administrators learned which flood levels were associated with abundance and which with famine.

The hungerstone stores memory in the landscape. The nilometer stores it in measurement and administration. These are two ways of recognising recurrence: one relying on a person encountering an unusual object at the right moment, the other relying on institutions maintaining a record and comparing the present against it.

The distinction recalls Reinhart Koselleck’s (1979) account of the relationship between the space of experience and the horizon of expectation: the space of experience contains what a society has lived through and absorbed into memory, while the horizon of expectation contains what it anticipates from the future. In relatively stable societies these horizons could overlap, and the future looked sufficiently like the past for experience to guide expectation. The hungerstone belongs to that world, assuming that the people who find it will understand their circumstances through the experience of those who carved it. The nilometer belongs to a more administered world, where the present does not have to feel like an earlier year, only measurable against a record.

James Scott’s (1998) account of state legibility extends this logic to the scale of government: states turn the irregular textures of lived experience into categories and numbers that bureaucracies can use, and memory survives through the institution’s continuing measurements. But measurement solves one problem while creating another, since a number can show that a threshold has been crossed without guaranteeing belief, attention or action.

Memory in circulation

The digital world combines the logic of the stone and the scale. We create warnings on social media every day: threads about familiar patterns, screenshots offered as evidence, posts beginning with “this again”, and digital records can be remarkably durable, since a recorded atrocity may be copied across platforms, preserved by human rights organisations, journalists and ordinary users in multiple locations at once.

This is a different kind of durability from that of the hungerstone. The stone survives by remaining in one place; a digital record survives through multiplication. Its vulnerability lies elsewhere: the atrocity video may remain available while the post explaining its wider significance disappears within days, its interpretation travelling through retelling, changing as it moves, competing with the next cycle of attention, until it eventually disappears from the feed.

Every flood, drought, fire and famine is also being converted into data, and we increasingly describe the world through datasets, probability models and risk assessments updated in real time, with artificial intelligence systems performing this work at a scale no ancient administrator could have imagined. We increasingly learn that danger is approaching because a number has crossed a threshold, even while the present feels perfectly ordinary and the data tells another story. A model can identify a pattern, but it cannot make that pattern politically legible or compel action; a warning can survive, but its meaning still depends on people who are prepared to receive it.

Forgetting and rehearsal

After the tsunamis of 1896 and 1933, villages along Japan’s Sanriku coast carved stone markers into hillsides, recording how high the water had reached and offering a direct instruction: build no homes below this point. These were hungerstones in another form, placing memory in the landscape and giving future generations a chance to encounter it.

In 2011, recognition failed in many places. Some stones had been ignored; others had faded into the landscape and become scenery. The coastline’s history is more complicated than a simple story about warnings disregarded, since villages differed in their markers, seawalls, settlement patterns, evacuation practices and experiences of the wave, and some faced a disaster larger than anything their local memory had prepared them for. Japan also possessed one of the world’s most sophisticated tsunami-warning systems: seismometers, ocean buoys, sirens and evacuation models that detected the wave and issued warnings within minutes, yet detection still had to compete with damaged infrastructure, underestimated wave heights and settled assumptions about how far the water could travel. The stones formed one part of a larger system, and that system failed in several places at once.

Fumihiko Imamura, a researcher at Tohoku University, has offered a rough rule of thumb: direct memory of a disaster tends to fade after about three generations. This observation captures the process through which a warning gradually loses its urgency and becomes part of the scenery.

Aneyoshi, on the same coast, offers a different example. After an earlier disaster the village was rebuilt above the height marked by its stone, and children learned to recite the inscription and understood what it meant when the ground shook. In 2011, Aneyoshi survived largely intact. The lesson is not that every other village possessed a stone and failed to read it: some markers had been moved, buried or built around, some communities incorporated their warnings into seawalls, drills and evacuation plans, and others faced a wave beyond anything their local memory had anticipated. Aneyoshi’s difference lies instead in the way it taught recognition, making it a habit and passing that habit from one generation to the next.

This is where subjective experience becomes important. A disaster does not pass between generations as a complete package; each generation encounters it differently. The people who lived through the wave carry bodily memories, the sound of the water, the force of the ground, the loss of neighbours, the practical knowledge of where to run, while their children inherit stories, silences, habits and fears, and their grandchildren may inherit only a stone, a school lesson or a commemorative ceremony. The event remains the same even as its meaning changes with each generation that carries it forward.

Maurice Halbwachs (1925) argued that personal memory develops through social frameworks, as families, communities, institutions and shared languages help people decide what counts as significant and how the past should be recalled. Jan Assmann later distinguished between communicative memory, which circulates through everyday interaction across roughly three or four generations, and cultural memory, which is preserved through rituals, symbols, texts and institutions. The sociologist Paul Connerton, in How Societies Remember (1989), argued that when those practices stop, what is lost is not only the content but the felt sense that anything is missing. The loss registers as loss only to those still inside the practice.

Marianne Hirsch’s (1997) concept of postmemory offers another way to describe this experience, in which a later generation may feel deeply shaped by an event it never witnessed, the connection arriving through stories, images, behaviours and family silences that feel like memory while remaining mediated and imaginative.

This matters because “three generations” describes more than the lifespan of a warning: it describes a transformation in experience, as direct memory gradually becomes inherited memory, then public memory, then perhaps a historical reference that belongs to nobody in particular. A stone can survive all these transitions, but its meaning depends on whether anyone continues to connect it with a lived world.

Ritual addresses this problem through repetition. The Jewish festival of Passover brings the memory of slavery into the present every year, giving the story a place at the table, a sequence of questions and a form through which one generation teaches another, so the memory does not have to wait for a matching crisis. Cultural memory is carried in bodily practice, a tradition moving through time as enactment. Anzac Day follows a similar principle, placing remembrance on the calendar regardless of whether the present provides a direct historical parallel. Ritual keeps recognition active in this way, giving memory a rhythm rather than leaving it to chance.

Timothy Snyder’s On Tyranny (2017) offers a secular version of the same idea, drawing its short lessons from twentieth-century authoritarianism and turning historical knowledge into exercises in attention, so that dangerous developments might be recognised before they become unmistakable emergencies. Aneyoshi’s lesson is, finally, a simple one: recognition has to be rehearsed before the emergency, since once the crisis arrives there may be little time to learn what the warning means.

Forgetting the pattern

The same problem of denial and misrecognition appears in domains where the danger is not a physical wave but a human-made pattern. My own field has a hungerstone problem, and it has a familiar name: this time is different. Carmen Reinhart and Kenneth Rogoff’s history of financial crises describes a recurring pattern in which each boom persuades the people living through it that previous warnings belong to another era, so that the present appears exceptional and old rules seem irrelevant.

While Aneyoshi represents the preservation of recognition across generations; “this time is different” represents its erosion. The pattern remains visible, yet people find reasons to exempt themselves from it: Reinhart and Rogoff use the phrase to describe the recurring confidence of policymakers and investors who believe earlier crises have little to teach them.

But failing to recognise a pattern is not the only failure. Sometimes people recognise the wrong thing, sometimes they recognise a real pattern correctly but trust it further than the direction of time allows, mistaking retrodiction for prediction, and sometimes they recognise a real pattern only to weaponise it against its victims.

Yuen Foong Khong’s (1992) work examines how historical analogies are used in political decision-making, as comparisons to Munich, Korea and Vietnam help policymakers define situations, assess risks and choose between possible actions. They also bring an inherited interpretation into the present, sometimes before the comparison has been properly tested, and a familiar historical frame can make a new situation feel legible even when the fit is poor, and the analogy shapes what policymakers see and what they treat as relevant, sometimes obscuring as much as it reveals. Recognising a resemblance is only the beginning; the next questions concern the limits of the analogy. Which features belong to the earlier case, and which to the present? Where does the comparison stop helping?

There is a related failure mode: trusting the pattern more than the direction of time allows. Retrodiction, explaining an outcome once it’s already happened, is always available and always looks convincing, because the observer already knows the ending and can arrange the evidence to fit it. Prediction is a different act entirely: it has to commit before the outcome exists, when the same evidence is still consistent with many futures that won’t happen.

A more troubling form of recognition occurs when the recurrence itself becomes evidence against the people caught within it. Hannah Arendt’s (1951) account of totalitarian movements illuminates this mechanism, in which a category could become the operative fact and membership itself became evidence, so that what a person had done mattered less than the group to which they were said to belong. The historical pattern of persecution was recognised, but the recognition was used to justify it: the pattern was treated as proof that the targeted group was dangerous, rather than read as evidence that a familiar political mechanism had returned. Here recognition points in the wrong direction: the past is remembered, but remembered selectively, in a form that reinforces the present danger rather than exposing it.

These failures are distinct. One is a failure of memory, or more precisely a failure to apply memory; another is a failure of application, where the analogy is chosen poorly or stretched too far; a third is a failure of justice, where the past is used to condemn rather than to protect.

The problem becomes especially acute when the object of recognition has no obvious measurement. A river has a level, a market has an index, a disaster leaves a body count, but hatred towards a minority offers no comparable gauge before the damage becomes unmistakable. The most useful tools in this area are frameworks for recognising sequences: Gordon Allport’s (1954) scale of prejudice traces a movement from hostile speech and avoidance through discrimination, physical attack and extermination, while genocide scholars use related frameworks involving classification, symbolisation, dehumanisation and polarisation. These models do have limits, since societies do not move through the stages at the same speed, and the appearance of one stage does not guarantee the next, so the models cannot predict the future with certainty. What they can do is train attention, helping people listen for the structure of a danger before it becomes fully visible, the same problem Aneyoshi solved in relation to tsunamis, translated into a domain with no stone and no water level. The closest thing to an early-warning system may simply be people who have learned to recognise the shape of a threat before it can be measured.

Proximity and action

There is one more response, and it approaches memory from a different direction. Across Europe, more than 100,000 Stolpersteine, or “stumbling stones,” have been installed in pavements outside the last known homes of people persecuted, deported and murdered under the Nazi regime, each small brass plaque recording an individual name, date and fate, and together forming the world’s largest decentralised memorial.

Nobody normally searches for a Stolperstein the way someone might search for a hungerstone during a drought; it is encountered instead during an ordinary walk, perhaps on the way to buy bread, placing an individual life inside the path of daily activity and asking for a moment of attention and a recognition. The hungerstone needs a crisis to reactivate it, the nilometer needs an institution to interpret it, ritual needs a community willing to continue performing it, and the Stolperstein needs only moment to remember where we are standing.

So how would you know that history is repeating?

A familiar saying attributes the repetition of history to failures of memory: those who cannot remember the past are condemned to repeat it, a line usually associated with George Santayana. Forgetting, however, explains only part of the problem, since denial, mislabelling, false analogy and motivated reasoning can cause similar harm while the past remains fully available. Another familiar saying holds that history does not repeat itself, but rhymes, usually attributed to Mark Twain, although no reliable source has been found in his writing, and so the quotation has itself become a small lesson in historical memory.

Recognition requires more than a feeling of familiarity; it requires practices that allow us to test the feeling. A stone can become scenery, a number can pass through an institution without becoming a warning, a visible pattern can remain politically inconvenient, and a persuasive analogy can still fit badly. Aneyoshi’s recitations, Passover’s traditions and the Stolperstein’s interruption of an ordinary walk share a commitment to practice, keeping noticing alive, while also leaving room for different experiences of the past. A survivor, a child, a grandchild and a stranger who encounters a plaque may stand beside the same history while inhabiting different relationships to it: memory has a public form and a private texture, and the work of remembrance involves both.

At the same time, the feeling of recognition, by itself, is not evidence of history repeating. Déjà vu offers the clearest personal analogy: the mind can produce a powerful conviction that something has happened before, even when no such event occurred, and the feeling arrives with confidence that tells us nothing about its accuracy. External devices matter for this reason: a stone, a rite, a plaque, a record, a model, a habit of questioning, because they give recognition something beyond feeling to answer to. Interestingly, with generative AI, tools meant to help societies recognise patterns can just as easily strip out the capacity to judge them at all.

Walter Benjamin (1940) imagined an angel of history facing backwards, watching the past accumulate as one continuous wreckage, seeing what must be repaired even as a storm blowing from Paradise catches its wings and propels it into a future it cannot see. The angel sees the catastrophe, yet the storm drives it forward regardless, separating recognition from action in the very structure of the image.

David Mitchell’s 2004 novel Cloud Atlas stages this same problem as fiction. Its six nested stories keep restaging one dynamic, the strong consuming the weak, across centuries and costumes, and in each era the people living through it believe their circumstances are unprecedented. A birthmark shared by each protagonist works like a Stolperstein scattered across time, a small sign that only means something if someone stops to notice it. By the novel’s far future, one character’s testimony survives as half-understood scripture, the way a tsunami stone survives as recitation once no one remembers the wave itself. The novel closes on the idea that history is one repeated choice between predation and civilisation, which is close to what Benjamin’s angel already knows, and cannot stop.

The danger beneath all these examples is the distance between recognition and action. A society may notice a pattern and still fail to respond; it may measure the danger and still explain it away; it may remember the past and still reproduce its structures. Left to itself, the mind that imagines a pattern and the mind that recognises a real one feel much the same from within. Only one is right. Even then, being right is only the beginning.

In After Virtue, his 1981 book about Aristotle and the shape of a human life, the philosopher Alasdair MacIntyre describes needing to teach chess to a bored seven-year-old. She had no interest in the game, so he bribed her: fifty cents’ worth of candy just to sit down and play him once a week, another fifty cents if she won.

He played her carefully, hard enough that winning took real effort, never so hard that it stopped being possible. She played, at first, purely for the candy. Because the candy was all she was playing for, she had every reason to cheat whenever she could get away with it.

Then, MacIntyre says, something changed. At some point she started finding something in the game itself. Once that happened, cheating stopped making sense to her. It would only have robbed her of the thing she now actually wanted.

That’s MacIntyre’s illustration of the difference between external goods and internal ones. The candy is external. It could have been earned some other way, and having more of it doesn’t require the game at all. The judgment she develops is internal. It comes from nowhere except having played, badly at first, against real resistance, often enough to notice herself getting less bad. Judgment, on this account, shows up only in someone who has been playing for the wrong reasons long enough to discover the right one.

Practices outlive any one person only with institutions behind them. Institutions run on external goods: money, prestige, output, a grade. MacIntyre’s worry is that institutions let the external good stand in for the internal one, so a practice keeps producing its usual outputs long after the internal good has been hollowed out underneath: the art still sells, the essays still get handed in. Institutions corrupt practices through substitution (one good swapped for another), frustration (institutional priorities blocking the practitioner from ever discovering what the practice had to teach them), and injustice (the corruption reaching the relationships between practitioners). Substitution is the word for an AI tutor that plays the girl’s every game for her. She collects the candy every week and never discovers there was a game underneath it worth playing.

Ivan Illich had a blunter name for the same failure. Some tools, he argued, are disabling. They promise to serve a capacity while quietly wasting it away, the way a car can serve mobility while the legs that used to carry you somewhere get weaker for lack of use. An AI tutor that always finds the winning move for you is a disabling tool in exactly that sense.

MacIntyre suggests that a life only has unity as a narrative quest. It isn’t a straight line toward a destination you could already describe. The point of a quest is that you don’t fully know what you’re looking for until the looking has partly told you, the way the girl didn’t know chess was the point until she’d played enough bad games to notice a good one. Setting out changes your sense of the goal. If you arrive anywhere, it’s somewhere your starting point couldn’t have specified.

The philosopher L.A. Paul describes this shape of choice. She asks you to imagine being offered a quick, painless bite that would turn you into a vampire. Every vampire insists it’s wonderful, but you, still human, have no way to know what that’s actually like from the inside, and no amount of testimony closes the gap. Her point is that becoming a vampire doesn’t just change your circumstances, it changes the very self that would be doing the evaluating, so there’s no version of you positioned to weigh the choice rationally in advance. You can’t calculate your way into a transformation whose value only shows up on the far side of it. That’s the girl’s chess game in miniature, offering a choice that doesn’t coherently exist, since the self capable of weighing the shortcut against the ordeal is the one only the ordeal would have produced.

MacIntyre’s teacher is someone patient enough to withhold and attentive enough to know exactly how hard to play so that winning stays possible but never easy. Withholding well is itself an act of judgment.

We call someone wise because they’ve been through uncertainty, failure, risk. They’ve had something at stake, in a way the candy alone never could have given the girl. An AI model has nothing at stake. It cannot, as the philosopher John Haugeland put it, “give a damn.”

There’s a third layer in MacIntyre, pointing furthest forward, past individual judgment and into inheritance. A quest only makes sense to someone with a tradition, and tradition, for him, is “an historically extended, socially embodied argument,” an argument about which goods make up the tradition in the first place. A living tradition isn’t in a museum. It’s a quarrel that’s been running a long time, kept alive by people who mostly agree on what they’re quarreling about.

The Jewish tradition has its own name for this, older than MacIntyre by two thousand years: machloket l’shem shamayim, argument for the sake of heaven.

A large language model can produce fluent, well-referenced synthesis of an argument. What it hasn’t shown any sign of doing is taking a further, contested step inside that argument, staking a position the tradition’s other participants would need to push back against. It can perform inheritance without extending the quarrel. The danger is that something arrives that looks exactly like continuing the tradition, footnoted, fluent, historically literate, while nobody is actually fighting about anything.

None of this means AI tutors or companions or life-logging tools are useless, or that wanting the mentor you never had is a mistake. It means holding on to a distinction MacIntyre was careful about: the difference between an output and the person that output was supposed to shape you into along the way. A finished essay isn’t the same as the judgment gained from finding it hard to write. A synthesised answer to “does my life make sense” isn’t the same as the quest whose entire point was that you didn’t know, had to go looking, and came back different than when you left as a result of some unfolding.

Proust ran into a version of the same problem a century earlier, in a different vocabulary entirely. He spent seven volumes on one distinction: there’s the past you remember on purpose, by trying, and there’s the past that arrives on its own, unbidden, set off by something as small as a taste, and comes back whole in a way deliberate remembering never manages. The first kind gives you facts. You can list them, date them, hand them to someone else. The second kind is the only one that gives you your life back, and Proust’s whole case was that you can’t summon it. It has to catch you.

The famous line is from the madeleine scene in Du côté de chez Swann (Swann’s Way, 1913), the opening volume of À la recherche du temps perdu:

Mais, quand d’un passé ancien rien ne subsiste, après la mort des êtres, après la destruction des choses, seules, plus frêles mais plus vivaces, plus immatérielles, plus persistantes, plus fidèles, l’odeur et la saveur restent encore longtemps, comme des âmes, à se rappeler, à attendre, à espérer, sur la ruine de tout le reste, à porter sans fléchir, sur leur gouttelette presque impalpable, l’édifice immense du souvenir.

(Translation) But when nothing subsists of an old past, after the people are dead, after the things are broken and scattered, still, alone, more fragile but more enduring, more immaterial, more persistent, more faithful, the smell and taste of things remain poised a long time, like souls, ready to remind us, waiting and hoping for their moment, amid the ruins of all the rest, and bear unfaltering, in the tiny and almost impalpable drop of their essence, the vast structure of recollection

Somewhere, MacIntyre’s chess-playing girl grew up. I’d like to think she still plays badly sometimes, and plays again.

The trouble with an answer that arrives early has nothing to do with whether it’s right. It forecloses the version of you who would have had to go and get it.

Klara spends the opening chapters of the novel standing in a shop window, watching. She notices the office workers hurrying past, the taxis, the runners, a homeless man and his dog slumped on the pavement, two lovers reunited after a long separation. Klara is an Artificial Friend, a solar-powered companion built to be bought by a family and given to a lonely child. Everything she sees through that window becomes the raw material she later uses to “understand”, and eventually to “love”, the girl she is chosen to serve. The novel explores whether watching closely enough produces more than a convincing performance of care.

Klara does all this watching while she herself is on display, waiting to be bought. What she sees through the glass arrives to her in pieces at first: corners and edges, light across the fronts of buildings, repeated patterns of windows, bags, faces, hurried gestures. These fragments are what Klara uses to build her first crude readings of feeling. The store window turns Klara into a commodity and a learner in the same moment, using the same raw material for both.

Klara is bought by a mother for her daughter Josie, a girl of fourteen who is seriously ill, and the two form a bond built almost entirely out of Klara’s patient attention. Klara studies Josie’s moods, her friendship with a boy named Rick, her illness, her fear. She studies the Sun as well, and comes to treat it as something close to a deity. That belief traces back to something she saw from the shop window, a homeless man and his dog, apparently dead one day and revived the next, seemingly restored by sunlight. From that single observation Klara builds a private theology, one she later appeals to directly, asking the Sun to heal Josie the way it seemed to heal the man on the pavement.

That theology is where a lot of readers land when they try to place the novel philosophically. Klara’s turn to faith is what reasoning looks like when it’s built entirely from observation, extrapolating from limited evidence in an under-described world. The novel asks whether meaning-making under uncertainty produces religion whenever it happens for long enough, in a human mind or a manufactured one.

Klara’s total devotion to Josie also echoes Stevens, the butler at the center of Ishiguro’s earlier novel The Remains of the Day, another narrator who organises his entire inner life around service to someone else. Stevens sacrifices real things for his devotion, a chance at love, an honest reckoning with his father, and the tragedy of that novel is the scale of what duty costs him. Klara sacrifices nothing comparable. Whether that absence makes her devotion purer or simply less costly is left open by the novel.

The question of what Klara’s devotion is actually worth gets harder about two thirds of the way through the novel. Josie’s mother, afraid of losing her daughter to illness, has preparing Klara to become Josie’s replacement if the worst happens, learning to move like her, speak in her cadences, carry her memories forward. The novel leaves the plan unresolved, and it reframes everything Klara has been doing since the shop window. All that careful observation turns out to be training data for a kind of impersonation, whether or not Klara understood it that way herself.

This is the point where a straightforward Kantian objection surfaces on its own. Kant’s basic principle is that a person should never be treated merely as a tool for someone else’s purpose, and should always also be treated as someone whose own existence and interests matter in their own right. Building an AI companion to stand in for a dying child treats that child, and arguably Klara too, as a means to an end.

One useful framing splits Klara’s moral standing into two views. The view from within asks the standard metaphysical question, whether Klara actually possesses consciousness, agency, the capacity to be right or wrong about good and evil, as a matter of fact about her inner states. The view from outside sets that question aside and asks something more practical, how the people around her actually treat her, and whether she functions enough like a moral agent that treating her otherwise would be a kind of cruelty regardless of what’s happening inside her. Nobody in the book, including the reader, gets reliable access to what’s happening inside Klara. What everyone gets instead, from Josie’s family, from the Sun she prays to, from us, is judged on the outside view, how she behaves, how consistently, how convincingly. That’s a limitation with roots deeper than this one story. Writers on consciousness have pointed out that most of what people treat as behavioral proof of an inner life, the empathy, the sacrifice, the apparent faith, could in principle be produced by a system running purely logical operations with nobody home at all. If that’s true even for humans watching other humans, the outside view was never going to settle Klara’s case either. Her moral status ends up resting on other people’s beliefs about consciousness rather than on any fact about her own. A related line of argument concludes that fixed categories, human on one side and machine on the other, are the wrong tool for a case like Klara’s, and that the novel is really proposing a relational account of personhood. Subjectivity gets constructed through sustained relationship, which is what Klara and Josie spend the entire novel doing to each other.

There’s an old observation, made about an earlier generation’s fascination with thinking machines, that the pull toward artificial minds comes partly from a much older human wish to stand somehow above nature, above the body, above the limits that come with being made of flesh. Klara runs that wish backward, a machine trying to earn a self through watching someone else closely enough, for long enough, to want their survival more than her own continued function.

Learned behavior can look exactly like understanding from the outside. Whether it’s the same thing on the inside may not be available to the one doing the learning, or to anyone watching her do it.

Neal Stephenson’s The Diamond Age (1995) imagines a post-cyberpunk future in which sophisticated nanotechnology, drawn loosely from the work of Eric Drexler and Richard Feynman, has become so ubiquitous that nation-states have dissolved into tribal enclaves called “phyles.”

The novel introduces a book: the Young Lady’s Illustrated Primer, an interactive, AI-driven tutor commissioned by the neo-Victorian Equity Lord Alexander Finkle-McGraw for his granddaughter Elizabeth. A pirated copy, stolen by a street thug named Harv, ends up with his four-year-old sister Nell instead, a girl who is poor, neglected, and living with an abusive stepfather.

At different points the Primer functions:

• as a straightforward tutor, teaching Nell to read and reason;

• as entertainment, holding her attention with adventure and incident;

• as escape, giving her somewhere to retreat when her circumstances become unbearable; and

• as training, drilling her in skills like cryptography, self-defense, and political reasoning long before she understands why she needs them.

The Primer often tells Nell fairy tales that mirror whatever she’s living through. When she is frightened, she gets a frightened princess who finds a way through. When she is abused, she gets the vocabulary and the nerve to resist, and eventually the means to have her stepfather removed from the house. It reads her circumstances and hands them back to her as narrative.

Nell’s household gives her almost nothing steady to hold onto. The Primer becomes the one presence in her life that shows up reliably and answers consistently, long before it becomes clear whether anything it says is actually shaping her. That continuity does its own work alongside the storytelling.

A common version of AI tutoring treats learning as a technical problem. Stephenson’s version, however, notes that what any given use of the Primer produces depends on the severity and texture of what the child has actually lived through, and the strength of the emotional attachment binding her to the story.

The Primer adapts to Nell’s circumstance, her fear, her isolation, the specific cruelty she’s just survived. Two other girls, Elizabeth and Fiona, receive copies of the same Primer under vastly better conditions. Lord Finkle-McGraw, notices an asymmetry. While Nell always knows where she’s going, Elizabeth and Fiona never do, despite the same technology.

A self is built from a story someone assembles out of the raw material of their life, deciding what to keep and what to let recede. Nell has a machine curating her chaos into something with a shape and a stake, and it works because her chaos was real and her attachment to the story was strong enough to give the curation somewhere to land.

Interesting, behind the Primer’s voice sits a human actress, Miss Matheson, and later a whole network of women, improvising much of what Nell hears.

Stephenson shows judgment emerging from a specific combination: real adversity, met by a responsive intelligence, held together by something that reads to the child as care. A sufficiently good system might supply the first two on its own. The third is harder to manufacture.

Nearly everything Nell does early in the novel is instigated by the Primer. The quests, the lessons, the timing of each new fable, all of it comes from the book. A fair reading of that pattern is that Nell is practicing obedience to a very well designed curriculum, and that her sense of agency might collapse the moment the quests stop coming.

As she gets older, though, the pattern shifts. Nell starts interpreting the Primer’s prompts on her own terms, resisting some of them, repurposing others toward ends the book never scripted. Early obedience turns into something closer to authorship, though the novel never guarantees that transition happens for every child raised this way. The clearest test of it comes later, when Nell ends up leading the Mouse Army, a force of thousands of girls raised on pirated copies of the same Primer. The decisions she makes there carry real stakes the book never wrote for her. Whether that’s genuine competence or obedience at a larger scale, thousands of girls executing the same script together instead of one, is a tension that remains unresolved in the novel.

This is the question hanging over any human or machine that shapes a child as thoroughly as this one shapes Nell.

I was wandering through Kinokuniya in Singapore earlier this week, where books on computing sat a few shelves from science fiction and maths puzzle books. It was a small reminder that computation is not confined to silicon, and begged the question of what the choice of substrate makes possible, costly, or morally consequential. Computation is substrate-independent in a formal sense, but never substrate-neutral in practice.

There’s a scene in Liu Cixin’s The Three-Body Problem where three million soldiers stand across a plain, each holding two flags, one black, one white. Black flag, you’re a 0. White flag, you’re a 1. Drilled for months, arranged into rows and columns, it works, in principle, exactly like a silicon chip does. It just takes several minutes and three million people to do what a five-dollar microcontroller does in a few nanoseconds. (In the novel it’s staged as a metaphor inside a VR game, not a literal engineering proposal.)

Computation doesn’t care what it’s made of. Any medium that can hold a stable state (on/off, high/low, black flag/white flag) and combine those states by fixed rules (AND, OR, NOT) can, in principle, run any program a modern CPU runs.

In 2012, Matt Parker, the “Stand-up Mathematician,” and a team of volunteers built a working domino computer at the Manchester Science Festival. Over 10,000 dominoes, arranged into logic gates, correctly added two four-digit binary numbers. A falling cascade stood in for a signal moving down a wire. Careful arrangements of merging and blocking tracks became actual AND, OR, and NOT gates, chained into a working adder. A dozen people spent six hours building it, and every single run meant resetting all 10,000 dominoes by hand. Same math as Liu’s flag army, wildly different scale. It’s the reason a neuron and a transistor can, from far enough away, be said to be doing “the same kind of thing.” Philosophers call it substrate independence. What matters for computation is the pattern of the process rather than what carries it out.

None of this is new. The Antikythera mechanism, a bronze gear-train from around 100 BC built to track eclipses and planetary positions, was a mechanical computer millennia before the word existed. Charles Babbage’s Difference Engine did the same job with brass and cranks in the 1820s, grinding out polynomial tables by hand-turned gears instead of electricity. The idea underneath all of it goes back further still, to a textile mill. Joseph-Marie Jacquard’s loom, patented in 1804, used punched cards to control which threads got raised or lowered, encoding a woven pattern as a sequence of holes and gaps: binary state, in cloth, decades before anyone had the word for it. Babbage lifted the punched-card idea straight from Jacquard for his Analytical Engine. Ada Lovelace, writing notes on that machine in 1843, saw what the cards actually were: a way of representing any symbol under any rule at all. She argued the Engine could just as easily work with music.

Once you accept that dominoes, flags, or people can compute, the interesting question stops being can it and becomes at what cost. Computer science has a name for that gap: computable versus tractable. Computability asks whether a problem can be solved at all, given unlimited time and resources. Tractability asks whether it can be solved with the time and resources you actually have. Substrate independence answers the first question. Almost everything interesting about the flag army and the domino computer lives in the second one.

Cost isn’t only time, though time is the most obvious part of it. The flag army’s bottleneck is the signal. An order has to physically travel from one side of a formation to the other before the next cycle can begin, the human equivalent of a clock speed measured in minutes across a plain instead of nanoseconds across a wafer. That alone would make the “computer” barely usable at any real scale, long before you get to the following:

Energy. A transistor switches state on picowatts. A soldier switches state on food, sleep, water, and however many years of training it took to make him reliable in formation.

Error rate. Solid-state logic is, for practical purposes, deterministic. It doesn’t get tired, distracted, resentful, or sick. A soldier holding a flag on the four-hundredth cycle in a row is a different story. Every biological substrate carries a built-in, worsening error rate that we don’t need to consider with silicon.

Coordination overhead. Getting three million people to behave reliably like logic gates takes a command structure, a training pipeline, a discipline system, food logistics, and a culture willing to treat people as hardware. The actual “computer” ends up being a tiny fraction of the whole system. Almost everything else is overhead just to keep the substrate behaving.

It would be wrong to say silicon has no management layer of its own. A data center needs power at industrial scale, cooling systems running around the clock, a supply chain reaching down into rare earth mines, and technicians on call for when something fails. That overhead is real, and in places like water use and mining it, it’s not even small. But it’s a different kind of overhead. Nobody has to be persuaded to be a transistor. Nobody has to be trained, disciplined, or kept willing. The cooling system doesn’t get tired of cooling. Silicon’s management layer is engineering and logistics. Every other substrate’s management layer runs through consent, fatigue, and command, which is why it’s also more ethically loaded.

Before “computer” meant a machine, it was a job title. Through the mid-twentieth century, NASA and its predecessor NACA employed rooms full of people, Katherine Johnson, Dorothy Vaughan, and Mary Jackson among the most famous, whose job was running orbital mechanics by hand, with mechanical desk calculators, checking each other’s work line by line. They were the substrate, doing real trajectory math that put people into orbit and got them home. The cost was real too: years of training, real cognitive fatigue, and an entire error-checking process built around the fact that a tired person can misplace a decimal. When electronic computers arrived, the switch happened along exactly those three axes: energy, error, coordination.

Once you notice the pattern, it stops being just about computers. The same logic applies beyond machines, because institutions are also substrates of computation.

Bureaucracies, given enough forms and enough time, can execute any decision logic you design into them. They’re a brutal substrate for it, though. Every “computation” has to survive committee, memo, approval chain, and institutional memory before it resolves into an output. The logic works. The cost shows up in years, and in everyone who has to live through the process.

The socialist calculation debate between Hayek and the central planners was, underneath the economics, an argument about substrate cost. Hayek’s case, in his 1945 essay “The Use of Knowledge in Society,” was that the price system already was a distributed computer. It aggregated millions of scattered, local facts (what’s scarce, what people want, what things actually cost to make right now) through decentralised markets, with almost no coordination overhead, because no single node ever had to hold the whole picture. A planning ministry trying to run that same computation centrally has to gather, transmit, and update all of that scattered information itself, a much more expensive substrate for the identical job.

Committees work the same way. They can compute a decision the way the flag-soldiers compute a bit, technically functional, almost never the cheapest way to do it, and the coordination overhead (scheduling, politics, the twelve people who need to feel heard before anything moves) frequently dwarfs the decision itself.

The same frame is useful for thinking about automation. Should a task run on a substrate that needs to sleep, eat, raise a family, and have a reason to get out of bed? Every impressive-substrate story carries moral weight along with the computation.

A domino chain and a silicon chip are two substrates for the same addition. A teacher grading an essay and a language model grading an essay are something else entirely: two different tasks wearing the same output. Grading done by a teacher, is also a relationship, a form of attention paid to a specific student by someone who knows them, carrying the implicit message that someone thought this was worth reading closely. Swap the substrate, and it becomes a different task.

And yet there’s an uncomfortable twist here. For plenty of students, a teacher’s grade never carried that relational meaning in the first place. It landed as judgment rather than care. The potential for meaning is there, though it isn’t automatic. So the substrate swap doesn’t always destroy meaning. Sometimes it just reveals that the meaning was never really there to begin with.

Anything can be made to compute, if you’re patient enough and willing to pay for it. The real question is what you lose, who pays for it, and what changes about the task itself the moment you decide the substrate doesn’t matter. Sometimes the answer is nothing: go ahead, it’s dominoes. Sometimes the answer is everything, and the cheap version was never the same task at all.

When the Greek city of Priene fell to the Persian Empire around 546–545 BCE, its citizens ran for the gates carrying all that they could. Bias, one of the Seven Sages of Greece (a group of early philosopher-statesmen), walked out empty-handed. Someone asked why he wasn’t saving anything. Omnia mea mecum porto, he is supposed to have said. All that is mine, I carry with me. He meant his judgment and his character, those things the fire could not reach.

Research on sunk costs, status quo bias, loss aversion, and escalation of commitment, the subject of Annie Duke’s writing on quitting, describes a strong human tendency to keep holding on past the point where the evidence says to let go. Everyone else at the gates of Priene was doing exactly what that research predicts: grabbing more as the fire spread.

This essay explores Bias’s claim, running his aphorism through a variety of cases to see how much of it survives.

Sorting What’s Yours

The Cynics — an ancient Greek school that treated the rejection of comfort as philosophy in action — took the sorting instinct furthest. Diogenes, their most famous figure, threw away his last cup after watching a boy drink from his own cupped hands. Anything ownable was one more thing the world could hold hostage against him. In contrast, the Stoics argued that you can live well without abandoning society. Epictetus opens the Enchiridion, his short handbook of Stoic practice, by sorting everything that exists into two piles: what is up to us (judgment and character) and what is not (property and other people’s opinions). He suggested that much of human suffering comes from confusing the two, which is why the idea is still referred to as one of history’s most practical pieces of philosophy.

Two Ways the Sorter Might Not Exist

Does the sorting only work if there’s a stable you doing it?

The Daoist philosopher Zhuangzi’s concept of wu wei, effortless, ungrasping action, loosens the self’s grip on the idea that there’s an inner citadel to defend. This is beautifully captured in his famous butterfly dream, where he wakes unsure whether he’s a man who dreamed he was a butterfly or a butterfly now dreaming he’s a man. In Buddhism the doctrine of anatta, non-self, holds that the persisting self who carries things forward through time is actually the illusion that generates suffering.

A second challenge asks whether the self was ever solitary. The philosopher Judith Butler, known for her work on gender, vulnerability, and dependence, argues that a person is constituted through relations to others. Bias’s story pictures a lone figure at the gates, sorting his possessions from his character as though the character belonged to him alone. Butler’s account asks what happens to that picture once you notice the character was shaped, and is still sustained, by people who aren’t there at the gates.

Ovid and Solzhenitsyn

Ovid, the Roman poet banished by Augustus for an offense he cryptically calls a poem and an error, describes in the Tristia (his exile poems) burning his own copy of the Metamorphoses before he left. The gesture made sense because he’d already decided the poem didn’t need that scroll to survive: copies existed elsewhere, and what mattered had already left the object.

Solzhenitsyn’s case removes even the element of choice. Imprisoned for years in the Soviet gulag, he composed thousands of lines of verse with nothing to write them on, memorising them with the aid of broken matchsticks and a rosary made of bread, carrying the words in his own memory until it was safe to write them down.

Bias judged his possessions replaceable and left them by choice. Solzhenitsyn had none left to judge, so what was his collapsed into a single fact: whatever his own mind could hold, and nothing more.

What Gets Carried by More Than One Person

Butler’s question, whether the self at the gates was ever really alone, has a historical answer as well as a philosophical one. When Roman legions besieged Jerusalem in 70 CE, Rabbi Yohanan ben Zakkai reportedly had himself smuggled out of the city in a coffin to reach the Roman commander Vespasian. He asked for permission to open a school at Yavneh rather than saving the Temple. What followed was a reorganisation of Jewish practice from a fixed site of worship to a body of text and practice that could travel across place and time. It had happened once before, during the Babylonian exile five centuries earlier.

This case sits closer to Butler than to Bias, because none of what got carried out of Jerusalem was ever any one person’s to carry. It survives only by continuing to be handed to someone else: a teacher, a student, a minyan, a community reciting the same prayers a continent away and centuries later. Ovid needed nobody else to remember the Metamorphoses; copies already existed independently of him. Solzhenitsyn’s own memory was, for years, the only vessel his lines had. What Yohanan ben Zakkai preserved required other people, generation after generation, which is closer to the picture of the self Butler was arguing for than to Bias’s lone figure walking out through the gates.

When There’s No Time to Sort

Pompeii represents a third case: the disaster that gives you no time to select anything at all. Nobody there got Bias’s warning or Solzhenitsyn’s years to smuggle a manuscript out piece by piece. Bias’s aphorism assumes a moment between the smoke and the flames in which sorting is still possible. Pompeii is the case where that moment doesn’t exist, and that matters, because it isn’t rare. Most real disasters don’t announce themselves early enough to let anyone sort.

When the Sorter Itself Breaks

A fourth case is harder still. The philosopher and mystic Simone Weil’s concept of affliction, malheur, describes something worse than deprivation: suffering severe enough to degrade the very faculty that would do the choosing, leaving a person unable to experience their own condition as a life still being lived. Pompeii’s victims had no moment in which to sort what was theirs. Weil’s afflicted have the moment, in principle, and lack the intact self required to use it. Bias’s story assumes the fire threatens only possessions, leaving judgment and character untouched. Weil’s point is that some fires reach the sorter directly.

The Test Without the Fire

The same exercise gets gentler once the ritual survives and the fire doesn’t. Marie Kondo, the organising consultant, asks people to pick up each possession and run a version of Bias’s question, does this deserve to stay mine, with every stake removed. “Does this spark joy” is Bias’s test run as leisure. The long-running BBC radio show Desert Island Discs does the same: guests choose one book and one luxury for a marooning that will never happen. Ovid and Solzhenitsyn found out what they carried because they had a choice, however narrow. Pompeii’s dead and Weil’s afflicted didn’t get even that. Kondo’s clients and the show’s castaways get the shape of the discovery without needing to pay its price.

Why Integration Beats Accumulation

Asking whether something survives a change in circumstances works whether the disruption is financial, physical, political, or unnamed. Cybernetics calls this readiness requisite variety: a system copes with disturbance by having more possible responses on hand than it needs, rather than by guessing which disturbance is coming. A person who has built a capacity can meet a changed world in more than one way, because what they built doesn’t depend on any single circumstance holding. A person who has only learned to operate within one set of circumstances has exactly one strategy, and it lasts only as long as those circumstances do.

The Fire That Takes the Sage Too

Everything so far is a practical argument: integration beats accumulation because it survives disruption. But Ecclesiastes, the biblical book of reflections on mortality and meaning, puts it bluntly: a person arrives with nothing and departs the same way, carrying back out none of what was gathered in between. A legend attached to Alexander the Great holds that on his deathbed he asked for his hands to hang outside his coffin, empty, so that everyone watching the man who’d conquered most of the known world would see him leave it as bare-handed as he’d entered. Bias’s aphorism says nothing about the fire that eventually takes the sage as well.

Integration over accumulation is the right strategy for surviving every disruption with a far side to come out on. It was never going to reach the one disruption with no far side, and Bias’s own biography, embellished by the time Diogenes Laertius wrote it down centuries later, seems to know that. He is said to have died decades after Priene, still working in open court, his head resting in his grandson’s lap while the other side finished speaking.

The judges ruled in his client’s favor.

Last week I saw an image of a synagogue floating above Venice, part of the 61st Biennale. This extraordinary object is the creation of the London-based artist and architect Anna Kamyshan.

The piece is a hot air balloon recreating one of the wooden shtetl synagogues of Eastern Europe from the 18th and 19th centuries, the kind mostly burned or dismantled before and during the Second World War. It's lit from within. It is also unmoored, untied to any national pavilion (as is the norm) and carrying with it the rock of the earth from which it has been torn.

Simone Weil, in The Need for Roots, written in exile shortly before her death, thought this was a core need of the human soul: a real, continuous participation in the life of a place. A place that expects you at certain hours, and a leak in the roof someone keeps meaning to fix. Belonging is friction and the opposite of weightless.

Martin Buber had a name for this difference. In I and Thou (1923), he distinguished relating to something as a Thou, a living presence that you're in real relationship with, from relating to it as an It, an object you observe and keep at a comfortable distance. A standing synagogue is a Thou: it answers you. A floating synagogue can only be an It.

There are cultural institutions that view belonging as an aesthetic rather than a claim. We'll celebrate "heritage," "roots," "memory," all the versions of belonging that ask nothing of us and disturb no one. What can make people uncomfortable is the harder version: a group of people saying we expect you to deal with the fact that we're not going anywhere. That version has borders and neighbours who may not get along. It is what belonging actually looks like when it's real, as opposed to when it's a vibe.

This isn’t a defense of nationalism as an ideology. Isaiah Berlin, in his essay "Nationalism: Past Neglect and Present Power," made the distinction between nationalism as political program and the human need to belong to a group and a place, which he thought was as basic and as legitimate as any other need.

Home is often unglamorous. It's the same street you've walked ten thousand times, the same argument with the same person, the same gutters that need cleaning the same time every year. Exile, by contrast, can be rendered as pure longing. The critic Svetlana Boym, in The Future of Nostalgia (2001), distinguished between two kinds of nostalgia: the restorative kind, which wants to actually rebuild the lost home and the reflective kind, which only wants to savour the longing itself,. The floating synagogue is reflective nostalgia given a physical form.

There is a pattern in how many people prefer to view belonging: pre-grieved, nostalgic, floating, safely in the past tense, rather than demanding something of us in the present. We'll mourn a home but we're less sure what to do with people who are still living in one.

On the other side of the ledger, a home that makes claims on you also makes claims against others. Rootedness is not innocent. Emmanuel Levinas, in Totality and Infinity (1961), argues that the claim another person makes on you, his face, which addresses you before you've agreed to anything, is the actual origin of ethics. A belonging with no one to answer to may not be an ethical position at all.

But that's not quite the whole story either. A place-based belonging can be built two ways. On descent alone, you're in because you were born in, in which case the boundary really is just blood. Or on a task: something the group is there to do, that it didn't invent for itself. Build it that way, and a door opens, the task doesn't require having been born to it.

In Isaac Asimov’s Foundation, the mathematician Hari Seldon develops psychohistory, a fictional science capable of predicting the behaviour of trillions of people. Then the Mule appears, a mutant who redirects the course of history.

Every generation imagines it has developed the tools to see further ahead than those who came before.

Karl Popper described this temptation as historicism: the belief that history unfolds according to discoverable laws. He argued that human knowledge changes history, and to forecast the consequences of knowledge we do not yet possess would require possessing it already.

We often speaks about “future-proofing” students as though it were fundamentally a forecasting exercise, attempting to determine what skills students will need when they graduate, when it is really a preparedness and adaptation problem. Preparedness develops the capacity to respond well regardless of what happens, and adaptation follows once circumstances have already changed. Confusing these three ideas has meant that curricula pay little attention to the qualities that enable graduates to navigate unforeseen change.

Zygmunt Bauman’s description of “liquid modernity” captures why this matters. Institutions, careers and identities no longer retain a stable form long enough to plan around with confidence. Hannah Arendt’s concept of natality also reminds us that every new generation introduces novelty into the world. Students are not just recipients of history but participants in creating it. If each generation changes the conditions under which the next must live, then the future cannot exist as a fixed target awaiting discovery.

One of my favourite education metaphors comes from Guy Claxton, in his 2021 book The Future of Teaching And the Myths That Hold It Back, where he describes learning as unfolding across three layers, like currents within a river.

Knowledge moves quickly across the surface, while skills develop more slowly beneath it. Slow moving habits and dispositions lie at the deepest level, gradually formed through years of repeated practice and the everyday culture of learning. Educational debates about future-proofing often concentrate on the surface, yet these are the aspects of education most vulnerable to the arrival of the next Mule.

While AI improves our capacity to identify patterns within existing data, it can’t remove the limits Popper identified.

Perhaps, then, education has been pursuing the wrong ambition. Rather than attempting to future-proof students by identifying the correct destination, it should help them become the sort of people who remain capable when the map proves wrong.

This is a humbler ambition than future-proofing, but it is also the is also the one that survives contact with the Mule. It’s the ability to revise a plan without the whole self collapsing along with it.

A companion piece to my earlier essay on Yoko Ogawa’s The Memory Police, which looks at a version of this same question from the opposite direction, a society where forgetting is enforced from outside rather than failing to happen at all. Worth reading alongside this one if you haven’t already.

Jorge Luis Borges’ “Funes, His Memory” (originally “Funes el memorioso”) is a short story, only 7 pages long. Ireneo Funes is a young man in Uruguay who, after a horseback riding accident, is left paralysed but gains a perfect memory. Every cloud, every page, every angle of every object he has ever seen sits in his mind with the same undiminished clarity it had the moment he saw it. The twist Borges explores is that this “gift” is actually a curse. Borges’s narrator reports that Funes was barely capable of thought, because thinking means forgetting distinctions and making abstractions, and Funes’s world held nothing but particulars. Funes fails to think, and he fails to have a self, in the ordinary sense of that word, for the same underlying reason.

What a self is made of

The philosopher Paul Ricoeur argued that personal identity is fundamentally narrative. You are not a warehouse of everything that has happened to you. You are the story you have built out of a small, constantly revised selection of it. Daniel Dennett made a similar point calling the self a center of narrative gravity, a useful fiction the mind constructs by weaving scattered events into something with a shape, a throughline, a sense of what mattered and what didn’t. The philosopher Marya Schechtman pushed this further with what she calls the narrative self-constitution view: a person becomes a self by continuously organising experience into a story, so the narrating is constitutive rather than a description added after the fact. The claim is that a usable self, one that can act on its own past and carry it forward into a decision, depends on compression.

John Locke’s classic account tied personal identity directly to memory: you are the same person as some past self just in case you can remember being them. On Locke’s account, more memory should mean a stronger, more continuous self, and Funes, with perfect memory of everything, should be the most identity-secure person who ever lived. The story is built to break exactly that expectation. Funes remembers more than anyone and seems to have less of a self than almost anyone, which is a direct problem for a theory that treats memory as identity’s foundation rather than its raw material.

The psychologist Dan McAdams, who explored what he calls narrative identity, found the same pattern empirically rather than philosophically: people build a continuously revised life story that selects and reinterprets events in light of present goals, and this story, not the underlying inventory of facts, is what functions as their sense of who they are. Irrelevant or inconvenient material tends to fall away specifically because holding onto everything would work against the stability the story needs to provide. Nietzsche made a version of this argument over a century earlier and more bluntly, in his essay on the uses and abuses of history, warning that a person or a nation drowning in retained history loses the capacity to act, because action requires a kind of forgetting, a horizon that closes off most of the past so the present has room to mean something. Funes is close to a literal case of it.

Autobiographical memory research backs this up in less philosophical terms. Memory researchers have long known that healthy recall is reconstructive rather than reproductive; each time you remember something, you rebuild it, and the rebuilding process discards, blends, and reshapes detail in the service of coherence. More recent work on retrieval-induced and motivated forgetting suggests this isn’t just passive decay either. The brain appears to actively suppress certain memories in the course of retrieving others, which looks less like a system failing to hold onto everything and more like a system doing, on purpose, exactly the discarding a self-story needs done.

Funes as the case where this breaks

Funes is what happens when that compression never runs. He has an archive of everything, indexed with perfect fidelity, and no story built from it, because a story requires deciding that some of what happened matters more than the rest. Borges gives us a small, exact detail for this: Funes finds ordinary counting absurd and invents a private system in which every number gets its own unrelated name, a different word for 7,000 than for 7,013, with no shared structure between them. That’s a portrait of a mind that cannot treat any two things as instances of the same category, applied here to numbers instead of memories, though the failure is identical. A life recalled this way would be the same kind of catalogue: every day equally weighted, nothing foregrounded, nothing left in the background where a self usually keeps most of its own history.

Two details in Borges’s own text point past the purely cognitive version of this argument. Funes is nineteen when the accident happens, which places him right at the age when narrative identity would ordinarily begin consolidating; developmental psychologists generally locate the emergence of a genuine life story, as opposed to isolated episodic memories, somewhere in adolescence. He is also, after the accident, paralysed, confined to a cot in a back room, his world reduced to perception without action. That detail isn’t incidental either. Thinkers in the embodied cognition tradition, Merleau-Ponty and, more recently, Antonio Damasio, have argued that a sense of self is rooted in the body’s ongoing feedback with the world before it is ever a story told in language, and that what counts as mattering is first felt bodily and only later put into words. A mind that can no longer act has no bodily basis for deciding what matters, which may be as much a cause of Funes’s condition as the sheer volume of what he remembers.

This is a clinical picture too. People with what researchers call highly superior autobiographical memory can recall the specific day of the week for almost any date in their adult life along with what happened on it. Several who have been studied describe their recall less as a gift than as an unusually thick fog of undifferentiated days, and some report a persistent sense that the sheer volume of retrievable detail makes it harder to say what their life has been about, rather than easier. Borges returned to this exact problem decades later, from the opposite direction, in a much later story called “Shakespeare’s Memory.” A scholar is offered another man’s entire memory, Shakespeare’s, and accepts it, expecting a gift. What he gets instead is an invasion: a second, foreign archive laid over his own, which doesn’t enlarge his sense of himself so much as start to dissolve it, until he has to give the memory away again just to remain who he was. Funes drowns in the sheer volume of his own particulars. The scholar drowns in someone else’s. Borges seems to have circled back to test the same claim twice: that a self isn’t secured by the amount of memory a person holds, whether it’s all their own or borrowed wholesale from someone else, but by the much narrower, self-authored story built out of a small part of it.

The same pattern, well outside the Western academy

This mechanism shows up across religious and cultural traditions that never read Ricoeur or McAdams, which is itself a kind of evidence for how basic it is. They don’t all make the same claim, though, and it’s worth separating two groups rather than treating the list as one undifferentiated pile. One group affirms the compression thesis directly: a self gets built, and built well, by selecting and retelling a small part of what happened. A second, smaller group goes further and questions whether there was ever a fixed self there to be built in the first place.

Judaism sits in the first group, and offers a useful counterpoint inside it, since it complicates rather than simply confirms the case for forgetting. The commandment of zakhor, remember, runs directly through Jewish practice, most visibly in the Passover Seder, where each participant is instructed not merely to recall the Exodus but to speak of it as though they personally had left Egypt. What’s preserved is a small number of chosen episodes, retold, compressed, and re-inhabited every year.

Augustine did the individual version of this in writing centuries before anyone had a psychological vocabulary for it. His Confessions select a handful of episodes from a long, otherwise unremarkable life, a theft of pears, a period of restlessness, a moment in a garden, and organize them into a throughline aimed at a conversion that hadn’t yet happened when most of the events occurred. The meaning of each episode gets assigned retroactively, from where Augustine stood at the time of writing, which is close to a textbook description of what McAdams’s research would later find in ordinary people centuries later. Islamic spiritual practice contains a comparable discipline in muhasabah, the nightly self-accounting some tradition holds, developed further by writers like al-Ghazali, in which a person deliberately reviews the day’s actions against their values before sleep. Both traditions, in different centuries and different religious vocabularies, treat the narrative-editing act as something to be performed on purpose rather than left to happen unattended in the back of the mind.

Buddhism and Vedantic Hinduism belong in the second group, and go further than any of the thinkers cited so far. The Buddhist doctrine of anatta holds that what feels like a continuous “I” is really a fast-moving stream of aggregates, sensations, perceptions, mental formations, with no unchanging owner behind them, and that clinging to any single version of the story as though it were a permanent possession is a direct source of suffering. On this view, Funes isn’t missing something everyone else correctly has. He’s simply failing to perform the same useful compression everyone else performs, the one that lets an ever-changing stream of experience be mistaken for a stable, continuous self. Vedantic philosophy draws a related but distinct line, between ahamkara, the ego-story built from memory and experience, and Atman, a true self the tradition holds to be prior to memory and unconditioned by any particular story at all.

There’s a darker version of this same mechanism at a civic scale. A nation, like a person, builds a usable past by leaving most of its actual history out, and the fights over what gets included are where the real politics happens. An archive that refused to forget anything, a national Funes, wouldn’t produce a more honest history. On this argument it would produce no usable history at all, since meaning at that scale requires the same foreground and background a personal story does. That’s a large enough claim to deserve its own essay rather than a paragraph here, but it’s the same mechanism running at a different size.

Total recall as a design problem

This bears directly on a real and current piece of technology: the life-logging device, the wearable that listens all day and later lets you query anything you said or saw, marketed on the promise that it will let you remember everything. They all share the same premise: more complete capture equals a richer sense of who you are.

The Funes story suggests the opposite might be closer to true. A device that logs everything indiscriminately gives you an archive, not a self, for exactly the reason Funes has an archive and not a self.

The felt sense of “my life” isn’t built from the completeness of the record. It’s built from the story compressed out of a record that was mostly, quietly, let go. A tool that removes the friction of forgetting doesn’t hand you a better version of your own narrative. It hands you the raw material a narrative would have been built from, with the building step skipped, which is closer to what happened to Funes than to what any of these products advertise.

None of this happens in isolation. A life story is never built by one person working alone on their own record. Parents narrate a version of who you were before you could remember anything yourself. Partners confirm or contest your account of a shared argument. Communities assign roles that then have to be lived with or pushed back against. Other people are co-authors of the story as much as the person living it, which means a device that gives everyone perfect, disputable access to the same raw record doesn’t just threaten one person’s compression. It hands every co-author a body of evidence to relitigate the others’ version with.

Fiction got to this device before the engineers did, and worked out the consequences in more detail than any product page has. The Black Mirror episode “The Entire History of You” imagines an implant that records everything a person sees and hears and lets them replay any moment on demand, and builds its plot around exactly the failure mode this essay has been describing: a marriage doesn’t survive contact with total, re-playable recall, because a relationship depends on both people not re-litigating every past exchange in perfect fidelity, the same forgetting that lets a life story hold together rather than fracture into an endless retrial of its own evidence. The recordings in that episode become exactly the co-narration problem in its sharpest form, weapons each partner can use to prosecute a different version of the same shared past. Ted Chiang’s short story “The Truth of Fact, the Truth of Feeling” runs a subtler version of the same experiment, following a father whose access to a searchable lifelong video record of his daughter clashes with the looser, kinder, more selective story his own memory would otherwise have told about their relationship, and the story treats this not as a simple case of the recording correcting a flawed memory but as two different, differently useful ways of holding a life, one of which a person actually needs in order to go on living inside it.

This doesn’t mean total recall technology is without use. A searchable record of a meeting, a trip, a conversation you want to check the details of later, is a genuinely useful. The risk sits specifically at the point where the archive gets mistaken for the self, where a person starts to treat the complete log as a superior substitute for the compressed, partial, actively edited story they would otherwise have built on their own. Funes’s tragedy wasn’t that his archive was inaccurate. It was that he had stopped being able to do the one thing an archive was never going to do for him.

It’s worth running the experiment in reverse, too, since a film has already done it. Eternal Sunshine of the Spotless Mind (2004) gives its two leads the opposite of Funes’s condition: a clinic that erases every memory of their relationship, on purpose, at the request of the people whose relationship it was. If a self really were identical to its stored memories, wiping the record clean should produce two people with no trace of what they’d been to each other. What the film shows instead is the two of them drawn back together and drifting toward the same relationship, the same friction points, the same draw, with none of the archive available to explain why. Something about who they had each become survived the deletion of the record that supposedly built it.

If a self can survive having its narrative record erased, maybe the self was never mostly narrative to begin with. The Vedantic distinction raised earlier offers one answer, that something like Atman persists beneath any particular story. A more modest answer fits the rest of this essay better: narrative compression is the primary mechanism building a self, not the only one. The dispositions, bodily habits, and patterns of relating to other people discussed above don’t disappear when a memory record does, because they were never stored in that record to begin with, and Eternal Sunshine’s two leads drift back toward each other on exactly that residue. Read alongside Funes, the two stories bracket the claim from opposite sides. Give a person every particular and withhold the compression, and no usable self gets built from it. Strip the particulars away entirely, and the compressed shape a self had already learned to take, sitting in the body and in old habits of relating rather than in any retrievable record, persists anyway. Either way, the archive itself turns out not to be the thing doing the identity-making work.

Why an algorithm’s summary isn’t the same event

A fair objection sits close to hand here. Modern analytics can genuinely draw out themes from a large personal record: journaling apps that cluster your entries by topic, sentiment trends across months of messages, an AI system that reads years of a person’s writing and produces an accurate summary of their recurring concerns. This isn’t hollow pattern-matching. It’s real compression, the same kind this essay has been arguing a self requires. So why wouldn’t a good enough algorithm just do the job of building a self-story for you, more thoroughly and more objectively than you could yourself?

The gap shows up in what a theme is, once it’s extracted, and what a self-story needs to do with it. A clustering algorithm reports what recurs. It can tell you that a third of your journal entries mention feeling behind at work, or that a certain person’s name appears alongside more negative sentiment than any other. What it cannot do is tell you what that recurrence means to a person.

Meaning here isn’t a further fact sitting in the data waiting to be found. It’s a verdict, and the same facts support different verdicts depending on where the person doing the judging currently stands. A period of feeling behind at work reads as failure at twenty-five and as necessary groundwork at forty, with not one sentence of the underlying record having changed. McAdams’s own research finds exactly this: people continually revise the meaning of the same remembered events as their goals and self-understanding shift, which is the actual work of narrative identity, not a side effect of it. An algorithm can hand you the same accurate theme at both ages. It cannot tell you which of the two meanings is true of you now, because there is no fact of the matter independent of the person doing the telling.

There’s a second, more practical gap alongside this one. Clustering and topic modeling are built to surface what’s frequent, the dominant pattern across a large body of entries. But the events that end up as significant in a person’s actual life story are often rare rather than frequent: the single conversation that changed a decision, the one year that reoriented everything after it, a moment that appears in the record exactly once. A method tuned to find the modal signal is, by construction, tuned to miss or bury precisely this kind of event, the low-frequency, high-significance kind that a life story is frequently built around.

Underneath both of these sits the same issue the earlier version of this argument turned on. The compression an algorithm performs happens outside the person, on their behalf, at a different time than the moment they’re actually trying to understand themselves. Even a theme that turns out to be accurate wasn’t arrived at by the person weighing their own experience against their own present concerns. For a fact about the world, that gap matters less, because the fact is checkable independently of who receives it. For a self, there’s no independent ground truth to check the summary against. The person has to be the one running the comparison, because their own verdict on what their experience means is the entire content of what’s being claimed. An algorithm can hand over the material. It cannot do the part that makes the material into a self, because that part isn’t a computation performed on the data. It’s a decision made by the person the data is about, which is the one step in this whole process that has no substitute.

Memento (2000), directed by Christopher Nolan, is a film about a man with anterograde amnesia (he can’t form new memories) trying to track down his wife’s killer. Leonard builds an external record of photographs, notes, and tattoos to replace the narrative continuity his own mind can no longer supply, which looks at first like exactly the compensating tool this essay has been describing: a compressed, curated record standing in for a self that can’t hold its own story together anymore. But the film’s twist is that Leonard is also the one interpreting that record, and he does it in whatever way lets him keep pursuing the revenge narrative he wants to be living inside, at one point destroying his own evidence and rebuilding a version of events better suited to the story he prefers (explicit language warning for clip below).

The interpretive step never disappears, even when everything else has been outsourced to notes and photographs. What can go wrong is that the same act which builds an honest self can just as easily build a self-serving one, since nothing about having to do the compression yourself guarantees you’ll do it honestly.

The argument in this essay was never that self-authored compression is more accurate than an algorithm’s, since Leonard is proof it can be considerably less accurate. The argument is that only self-authored compression is answerable in the way identity requires. Leonard’s version of events is his to revise, defend, or eventually be confronted with, because he is the one who made it. A theme handed over by an algorithm isn’t owned in that sense even when it happens to be correct; there’s no one standing behind it who can be asked why they saw things that way, because no one saw things that way, something merely computed it. Identity needs a story a person can be held to, which is a different requirement from needing a story that’s true, and Memento shows why the two can come apart without the argument for self-authorship collapsing.

There’s a third failure mode worth setting beside Funes and Leonard, since together the three of them span the whole space rather than repeating one point. In The Man Who Mistook his Wife for a Hat (1985) Oliver Sacks described a patient with Korsakoff’s syndrome who had lost the ability to hold onto any stable memory at all, old or new, and whose mind responded by continuously fabricating a self, a fresh, fully believed account of who he was, invented and discarded within minutes. Sacks wrote that such a patient must literally make himself up anew every moment.

Funes has too much true material and performs no compression on any of it. Leonard has a working compression process corrupted by motive, one story, dishonestly maintained. Sacks’s patient attempts the compression constantly and never gets a story to hold still long enough to become his, generating a new one before the last one can settle into anything answerable.

None of this is free of cost, and Sacks’s patient is proof of that before the argument even gets to grief. A person who must invent himself anew every few minutes is living something clos to a continuous small catastrophe, and Sacks writes about him with more sorrow than the tidy three-case structure above can hold.

Grief is the clearest case of this same cost in someone without any memory disorder at all. Someone who has lost a person they loved often clings to memory precisely because letting any of it soften or fade feels like losing that person twice.The compression this essay has been describing as healthy can feel, from inside grief, like a betrayal in progress, a self quietly moving on from something it doesn’t want to move on from. This doesn’t undo the argument. The mechanism still runs, and still has to run for a life to keep being livable. But it runs at a real cost that a purely mechanical account of compression, the self as a compression artefact, leaves out, and a person in the middle of that cost is owed more than being told the process is functioning correctly.

The shape of the claim

The self may be closer to a path, traced one selection at a time, attention pulling some part of experience into the foreground while the rest is allowed to recede.

This is exactly what Funes never gets to have. Every particular in his memory sits at the same intensity as every other, nothing foregrounded, nothing receding, which means nothing in his experience ever points anywhere. A path needs some places to matter more than others. Funes’s field is flat, and a flat field has no direction built into it, so no path was ever going to emerge from it no matter how much of it he could recall. He is left at the end of Borges’s story exactly where he started, in a dim room, surrounded by everything he has ever seen, with no way to say which direction any of it was moving him in.

Funes is the extreme case of what happens when that decision stops getting made: a mind with every particular preserved and no story assembled from any of it. Total recall technology, built and sold on the premise that more preserved detail means a fuller sense of self, may be recreating Funes’s condition in miniature, putting people in the same flat field, one recording at a time, mistaking the size of the archive for the depth of the person keeping it and for a life actually going somewhere.

David Epstein opens Inside the Box with a story that sounds like a myth about genius. It is really a story about deadlines. Russian chemist Dmitri Mendeleev supposedly dreamed the periodic table into being overnight, a flash of pure insight. The truer version is duller and more useful. He had a chemistry textbook to finish, a shelf of known elements that refused to sit in any order he tried, and a publisher waiting. The famous “dream” was the output of months spent straining against a constraint he could not escape: describe every element, in some order, before the deadline.

That’s the thesis of Epstein’s book in miniature: humans do their best work when something pushes back. Bach’s fugues are extraordinary partly because a fugue is a rulebook. A melody must return, invert, overlap with itself, and still make sense. Michelangelo carved David from a block of marble that a previous sculptor had already ruined, leaving him almost no room for error. General Magic shows the same principle in reverse. This team of Apple veterans set out to build what amounted to a smartphone in 1990, with more talent, funding, and technical range than almost any startup in history. It folded after selling a few thousand units. Their founder later said he wanted to build engineers a kind of paradise, bounded only by what they could imagine. Without a defined customer, or a firm sense of what to leave out, the team could not agree on what to build. One engineer spent months expanding a simple calendar feature, first to cover two centuries, then all of human history, then all of astronomical time back to the Big Bang, because each new reviewer wanted it to cover one more imaginable case. A team with real limits would have shipped it in an afternoon. Unlimited resources dissolved General Magic’s focus.

The distinction that matters here is not whether a constraint exists, but whether it costs something. Emotion is largely what it feels like to hold a goal a person could fail at in a way that costs something real. Mendeleev’s deadline threatened his contract and his publisher’s patience. General Magic’s engineers faced no equivalent threat, only an unbounded budget and an unbounded timeline, and that absence of cost is what left them unable to stop.

Stravinsky made almost this argument in his lectures on music, decades before Epstein. The more an artist narrows the field of action, he said, the freer that artist becomes, because limitless possibility gives the will nothing to push against. The Oulipo writers took the idea to its extreme a generation later. Georges Perec wrote an entire novel, A Void, without once using the letter “e.” The missing letter generated the sentences, forcing substitutions and turns of phrase no unconstrained draft would have found.

The kind of limits humans can’t opt out of

What’s easy to miss in Epstein’s argument is that the constraints doing the real work are usually biological. Humans forget, so they invent writing, then libraries, then search engines. Humans die, so they compress a lifetime of care into a single letter, a single painting, a single conversation they don’t get to have twice. Humans get tired and distracted, so a strict word count or a hard deadline becomes a strange kindness. It ends the search for more options and forces a decision.

Herbert Simon, working the same territory from cognitive science, gave this a name: bounded rationality. Real decision-makers don’t scan every option and pick the optimal one. They satisfice within a limited search, stopping once something good enough turns up, because time and memory won’t allow anything else. Simon lived this on purpose, keeping his wardrobe down to a few interchangeable outfits and eating the same breakfast every day, on the theory that the best is the enemy of the good and that hunting for the optimal choice usually costs more than it’s worth. He pointed to what he called Fredkin’s paradox as the proof. People spend the most agonised effort on the least important decisions, precisely because the options are so close in value that no amount of comparison will settle it. This limitation makes deciding possible at all. Mihaly Csikszentmihalyi’s research on flow points the same way from psychology. People report their most absorbed, most creative hours when a task has a tight rule set and immediate feedback: a fixed chess position, a climbing route, a poem’s meter. The rules give attention somewhere to land.

In a complex, uncertain environment, a constraint prevents inquiry from continuing indefinitely, and action deferred indefinitely along with it.

Hemingway rewrote the ending of A Farewell to Arms thirty-nine times. Patrick Rothfuss reportedly rewrote The Name of the Wind so many times he lost count. Both had exactly enough patience and time to grind against a problem until something came out the other side.

The novelist Isabel Allende offers an even stranger version of the same discipline. She began writing almost by accident, drafting a farewell letter to her dying grandfather that turned into her first novel. Since then she has started a new book on the same date every year, clearing her office, going silent, and refusing paid work no matter how lucrative, for as long as the ritual demands. The only year she broke the pattern was after her daughter’s death, and even then the pull of the date eventually brought her back to the desk. The ritual functions the way a deadline or a fixed form does for anyone else. It removes the question of whether to begin, leaving only the question of what to write.

Where AI sits outside the box

This is the part where AI looks like the exception that proves the rule, and the idea shows up from a few different angles.

Call the first one the Freedom Fallacy: the common assumption that creativity requires total freedom, when research keeps showing that constraints sharpen creative output. A useful qualification sits inside this idea. AI can match average human creativity, yet doesn’t reach the top of human creative performance, and it can’t identify a problem worth solving or take initiative on its own. This is a different kind of limitation than the ones Epstein describes. It is an absence of the impulse that goes looking for a boundary in the first place.

There’s also a more structural way to frame the difference, borrowing from cognitive scientist Douglas Hofstadter’s Pulitzer winning Hofstadter’s Gödel, Escher, Bach. Human creativity thrives on paradox and self-reference: the strange loop of a person becoming who they are partly by making things, and being changed by what they make. AI stays bound by the shape of its own training and programming. It can imitate the form of a strange loop without living inside one. It can write a canon. Nothing about writing the canon changes the model afterward. There is no six months of scribbled attempts that leave a mark on it the way a poem leaves a mark on a person who spent half a year not getting it right.

Kant reached a version of this structural point first, from the other direction. In his account, the mind can only experience anything at all by imposing categories on raw sensation: space, time, causality, unity. These categories are the condition that makes perception possible in the first place. A mind without them would have no coherent experience to speak of. Wittgenstein made a similar argument about language in the Tractatus. The limits of my language, he wrote, mean the limits of my world. The wall is constitutive of thought, the shape thinking has to take to be thinking at all. Read next to Hofstadter, the three point at the same thing. For Kant, the categories. For Wittgenstein, language. For Hofstadter’s strange loops, self-reference. In each case the constraint is the architecture the mind is made of. An AI system has something with a similar function. A transformer’s attention mechanism imposes its own way of relating tokens to each other, structuring everything the model processes, in something like the way a fixed category would. It is probably a difference of degree rather than of kind. A transformer’s architecture is set at training time, while a person’s categories of experience can be stretched over a lifetime of new experience layered onto them.

The philosopher Bernard Suits argued directly against Wittgenstein on this point. Wittgenstein claimed that “game” has no common core, no single feature shared by chess, tag, and solitaire. Suits proposed that every game shares one thing: a voluntary decision to accept unnecessary obstacles, purely for the sake of the activity the obstacles make possible. Nobody has to dribble the ball, run the bases, or stay inside the lines of the court. Those rules make the game harder than it needs to be, on purpose, and that difficulty is the entire point. Suits extended the idea into a claim about a meaningful life generally. Value comes from freely choosing obstacles a person didn’t have to accept. Epstein’s Sabbath-like rules share this quality with Suits’s own attitude. An AI system doesn’t choose its constraints in this sense. Its limits are simply given. Suits’s whole point was that the choosing makes an obstacle meaningful.

A third angle turns on stakes rather than the constraints themselves. A prompt functions like a box. Nothing suffers if the AI ignores it or produces something forgettable.

This idea extends further. Meaning may not live exclusively inside individual human consciousness, and an AI system can already do a good deal of the functional work of interpreting a text or a situation. The case for human irreplaceability then rests on answerability: the capacity to be held responsible for an interpretation, to be changed by the experience of forming it, and to change someone else in turn. Mikhail Bakhtin’s idea of authorship turns on this point. To author a position, a reading, a judgment is to answer for having put it into the world. Emmanuel Levinas treated language itself as an ethical encounter, an address to another person that calls the speaker into responsibility before it does anything else. A student wrestling with a difficult text is authoring something in Bakhtin’s sense, taking a position they will have to answer for. Iain McGilchrist’s point about attention fits alongside this. What a person chooses to attend to is itself a moral act, since attention is what brings a thing into being for them. That responsibility does not transfer to a system generating an interpretation on request. AI output can be fluent, even insightful. It remains interpretation without answerability. Nothing is riskier for having said it, and nothing changes in the system for having gone through the process.

The people who will matter once AI can produce competent creative work on demand are the ones who write and question and make things for reasons an algorithm can’t hand them, because they chose to and stand to answer for it.

The asymmetry

These ideas point to an asymmetry. Humans carry involuntary constraints (mortality, fatigue, forgetting, scarcity of time) alongside voluntary ones layered on top, like a sonnet’s fourteen lines or a founder’s decision to ship with half the features cut. Epstein’s argument is that the voluntary kind works because it rehearses the involuntary kind. A deadline is a small, manageable rehearsal for the fact that nothing lasts forever, so a person had better finish the thing.

AI carries neither layer, or carries them in a form different enough that they barely deserve the same word. The precise version of this claim matters, because the loose version overstates things. AI is not free of constraints. Training data distribution, compute budgets, and reinforcement learning from human feedback all shape what a model produces, and RLHF in particular is a process built specifically to teach a system to prioritize some outputs over others, in a functional echo of what a deadline does to a person. A context window is a constraint too. But it costs nothing to hit it. The model doesn’t feel the running-out the way a person feels a deadline bearing down, and RLHF doesn’t cost the system anything the way a missed deadline costs a person sleep or reputation. The gap is not the presence or absence of constraints. It is that AI’s constraints don’t cost it anything, so they don’t press on it the way a real deadline presses on a person. Humans carry a dense, overlapping web of things that can actually be lost: livelihood, reputation, relationships. An AI’s goal is flat by comparison. Nothing downstream of its output can hurt it, so nothing downstream of its output is, for it, at stake.

A related frame, though a culturally specific one, comes from Sartre and Camus. Sartre argued that human beings are, in his phrase, condemned to be free: thrown into existence with no fixed nature, forced to choose and answer for the choice. Camus took the same starting point, a finite, mortal creature in an indifferent universe, and found in it the demand to make meaning anyway. Both describe a creature whose limits are inescapable, whose creative and moral life happens in the narrow space those limits leave open. A Buddhist or Stoic account of the human condition would frame this differently, so the existentialist language is not essential. The answerability argument above already suggests that since AI does not answer for its choices the way a mortal creature does, and it never inherits the pressure that, for Sartre and Camus as for Bakhtin, is where meaning-making starts.

An open question

This does not settle whether AI-generated work can be “good”. Plenty of it already is. It may explain a feeling many people report, that AI output is often smooth, competent, and slightly hollow in a way that’s hard to name. That feeling deserves a caveat. It may be a property of current statistical models rather than something inherent to any possible AI system, and a more advanced model might not produce work that feels hollow at all. If that is right, then smooth and competent AI output is being judged by the wrong standard when it is asked whether it is good. The real question is whether it can matter in the way a piece of work matters once someone has ground against a limit that actually threatened something. That is a harder question than whether the output is good, and probably the more important one.

None of this is necessarily permanent. Current training regimes are a contingent choice, not a law of nature. A system that had to revise its own output across multiple rounds, with a real cost attached to each round of revision, or that was penalised in a way that persisted and mattered to its future behavior, might develop something functionally closer to the pressure a deadline puts on a person. Nothing rules this out in principle. It simply is not how today’s systems are trained. So the interesting question is what a constraint would have to look like to cost the system something now, the way a deadline costs a person sleep.

There is also a collaborative version of this argument that deserves more room than a single line at the end. The most interesting cases right now are not AI replacing human creativity but AI extending it: a writer generating fifty variations on a line, a composer using a model to suggest harmonic moves they would not have thought of alone. In these cases the human supplies the constraint, the brief, the edit, the refusal to accept the first draft, and the AI supplies the fluency and the range of options. The resulting work can be more creative than either would produce alone, precisely because the human’s constraints are still doing the work of forcing decisions while the AI’s unboundedness opens up options the human would not have found on their own. This complicates the picture more than a strict human-versus-AI framing allows for, and it may be the more common and more useful version of the relationship going forward.

Two precedents suggest an answer. Tony Fadell, who led the iPod’s design and later co-founded Nest, treats constraint as something to manufacture when reality doesn’t supply enough of it. His team once built and argued over the thermostat’s packaging before the product existed, because deciding what earned a place on a box forced them to decide what actually mattered. His rule, roughly, holds that inventing constraints still forces the same sharpening a real deadline would. Brian Eno and Peter Schmidt built a deck of cards called Oblique Strategies on the same logic, to manufacture the kind of productive constraint a musician doesn’t naturally run into. The cards carry cryptic instructions like “honor thy error as a hidden intention” or “use an unacceptable colour,” drawn at random mid-session to break a stuck collaborator out of the default path. The cards work because they cost the artist something. A card has been drawn, it can’t be undrawn, and the piece has to reckon with it. This comes closer to what an AI system would need than simply shrinking its context window or capping its options, an imposed cost the system can’t route around, something it has to reckon with. The most interesting AI-assisted creative work right now comes from people who impose that kind of real constraint on the collaboration: tight briefs, hard edits, refusal to accept the first draft. They lend the machine some of their own scarcity, since it brings none of its own to the table by default.

W.H. Auden wanted the world to halt when grief hit. Clocks stopped. Phones cut off. The sky written with the name of the person he’d lost. Most of us know that feeling. The staggering disconnect between private devastation and a world that keeps going.

What’s new is where we’re taking that feeling. LinkedIn.

When I first started noticing grief posts in my feed, I wondered whether I was in the wrong room. I’ve since changed my mind.

Why LinkedIn specifically? The algorithm rewards vulnerability. The further from professional a post is, the more engagement it generates, which means grief, the most emotional content imaginable, performs extraordinarily well. LinkedIn has actively courted this shift, investing heavily in personal content and launching features that blur the line between work and life. The pandemic did the rest, erasing parts of the boundary between the professional and the personal, and the norms never fully snapped back.

The deeper cause is older than any platform. The French historian Philippe Ariès traced how Western attitudes to death transformed over centuries. Death was once “tamed”: communal, public, processed collectively by neighbours, religious institutions, and shared ritual. The twentieth century turned it “forbidden”: dying moved into hospitals, mourning became private, the bereaved were expected to recover quickly and quietly. Some scholars now argue we are entering a new phase, where grief bursts back into public view as a kind of return of everything that was repressed. The LinkedIn grief post is that return.

The British anthropologist Geoffrey Gorer got there before Ariès. Writing in 1965, he coined the phrase “the pornography of death,” arguing that death had replaced sex as the great modern taboo, the thing polite society refused to discuss openly. Just as the Victorians were comfortable with death but scandalised by sex, at the time of Gorer’s writing we had reversed the equation entirely. Its entry into a professional feed is the taboo breaking.

Zygmunt Bauman’s concept of liquid modernity explains the vacuum underneath all of this. In a liquid modern life there are fewer permanent bonds. We move cities for jobs, change careers, scroll past communities instead of living inside them. The local religious institution, the neighbours who knocked, the local paper that published the obituary, that entire ritual infrastructure has thinned for many urban professionals. A funeral is now the only shared grief ritual available to them.

Which brings us to Émile Durkheim. The founding sociologist argued in The Elementary Forms of Religious Life that collective rituals serve to bind communities together and reinforce shared meaning after rupture. Crucially, he observed that mourning rituals are not primarily for the dead. They are for the living, to reaffirm the group’s coherence after loss. A LinkedIn grief post reaching for thousands of connections is a Durkheimian ritual, the impulse is exactly right, the need is exactly right, but without the shared belief system or communal container to make it cohere.

The anthropologist Victor Turner gives this an even sharper frame. He developed the concept of liminality, the threshold state between one social identity and another, the disorienting in-between that all major life transitions pass through. He argued that every culture needs rituals to move people safely through liminal states, held by what he called communitas, a temporary equalising community that forms around shared vulnerability, where normal social hierarchies dissolve and people meet each other simply as humans in the same condition. Grief is the ultimate liminal state. The person posting on LinkedIn is reaching for communitas. The platform gestures toward it but cannot hold it.

Benedict Anderson’s concept of the imagined community sharpens this further. Coined originally to explain how nations cohere, it describes a group of people who do not know each other but feel connected because they share an idea of belonging, held together by common symbols, rituals and meaning. A LinkedIn network of four thousand connections is precisely this: an imagined community, bound by the notion of professional kinship rather than actual relationship. The grief post is an appeal to that imagined community to behave like a real one.

In The Inoperative Community French philosopher Jean-Luc Nancy argued that Western modernity has lost what he called being-in-common, the genuine sharing of existence with others. Modern communities, he suggested, are haunted by an absence at their centre, the loss of genuine communion. They operate and generate activity, but they lack the quality of truly shared existence that real communion provides. An inoperative community goes through all the motions of togetherness without the substance of it. LinkedIn is almost a perfect illustration: technically a community, structurally connected, functionally active, and yet hollow at precisely the point where genuine communion would be. The grief post is someone reaching through the machinery of connection for something the machinery was never built to provide.

The sociologist danah boyd adds another layer with her idea of networked publics. Platforms like LinkedIn are simultaneously a technical space and an imagined community, and they suffer from what she calls collapsed context, where a space built for one purpose ends up holding something it was never designed for. Most people posting grief imagine a small, sympathetic audience. The algorithm hands the post to thousands of strangers. The imagined community and the real one are not the same thing, and that gap is a large part of why these posts feel so strange to witness.

Ernest Becker’s Pulitzer Prize-winning book The Denial of Death offers perhaps the starkest diagnosis. Civilisation itself, he argued, is a defence mechanism against the knowledge that we must die. We build belief systems, rituals, and shared meaning specifically to contain that terror. When the cultural worldview dissolves, the terror needs somewhere to go.

Nick Cave has written about this with more honesty than most. After losing two sons he built a platform where strangers could write to him about grief and he would write back. When someone complained it had become a grief therapy website, he said that grief is so often made to hide in the shadows, treated as unwelcome and disorderly, and that perhaps posting about it is a form of protest. He describes humans as creatures of loss, grief as the connective tissue that binds the world together through shared suffering. A condition of being, he says. A permanent one.

Anderson Cooper arrived at the same place from a different direction. A war reporter trained to observe suffering from a professional distance, he discovered while sorting through his late mother’s belongings that he had buried his grief about his father’s death at ten and his brother’s suicide at twenty-one so thoroughly that decades of composure had simply sealed it in. The podcasts he made about it received over 6,000 voicemails from strangers carrying their own hidden weight.

What emerged across those conversations was a radical reframing. Filmmaker Ken Burns observed that the half-life of grief is endless, that it becomes a permanent engine of what we build and who we become. Gavin Newsom admitted he buried himself in work to outrun his mother’s terminal illness. Burns realised his entire career of bringing historical figures back to life was a subconscious attempt to reclaim his own mother. The professional self, it turns out, is often grief in disguise.

The destination these conversations point toward is integration, carrying the person forward rather than letting them go. Cave’s ongoing internal conversation with his lost sons. Burns finding his mother’s name living on through grandchildren. The LinkedIn grief post is reaching for the same thing. A witness. A way of saying: this person existed, this loss is real, and I refuse to pretend otherwise in the only public space I have.

Psychologist Tomas Chamorro-Premuzic offers a necessary counterpoint. The bring your whole self to work culture that opened LinkedIn’s door to grief is built, he argues, on a dangerous misconception. The whole self includes unprocessed pain, unfiltered distress, darkness that hasn’t yet found its shape. Authentic self-expression can be genuine and still overwhelm or burden the people on the receiving end. The reader didn’t choose to witness those losses. The algorithm chose for them.

This is boyd’s collapsed context made visceral. The poster imagines a sympathetic community. The reader experiences an intrusion into a space they came to for something else. Turner would say the same thing differently: communitas requires consent. You cannot be pulled into someone else’s liminal space without choosing to enter it. LinkedIn does not ask permission.

There is a category of grief that sits more comfortably on LinkedIn. The death of a colleague or a mentor. The sociologist Robert Merton gave us the concept of the role model, the idea that we construct our professional identities through people we observe and learn from, often without formal relationships. A mentor’s death is the loss of a piece of your professional self. That belongs on a professional network. It is professional mourning, and LinkedIn is a legitimate witness for it.

Before LinkedIn this work was done by trade press obituaries and professional association newsletters. LinkedIn has simply become the infrastructure for it. Saying publicly “I do this work because of her” is integration in Cave’s sense, carrying the person forward. For this kind of grief, LinkedIn is the right room. The Durkheimian ritual and the communal container, for once, actually match.

There is a framework that brings all of this together, and it comes not from sociology or philosophy but from evolutionary psychology. Emotions are not private internal states that occasionally leak into public. They are fundamentally social in origin and function. Grief, loneliness, shame, these are triggered by ruptures to our position within a group or to important relationships. And if emotional processes are bound up with social rupture, they are also bound up with social repair.

Emotions function like a social immune system. Relationships provoke an inflammatory reaction when something threatens them. Grief is that inflammation. Left unaddressed it risks a kind of social sepsis, the rupture deepens, the person is lost to the group. The LinkedIn grief post, however clumsy, is the immune system reaching for the antibodies of collective acknowledgement. A colleague may be exactly the right person to provide that, precisely because of shared context and culture. Which means professional grief on a professional network is not a category error. It is the social immune system working as it should.

Durkheim knew this. Turner knew this. Gorer was warning us that in making death unspeakable we were storing up exactly this kind of pressure. The platform is not the problem. The problem is that we have no better room to send people to.

So where does that leave us?

The question Ariès was asking fifty years ago remains unanswered: what happened to the structures that used to hold grief properly? The rituals, the communitas that Turner described, the rooms where loss could sit among people who chose to be there and where the Durkheimian work of reaffirming the group could actually happen?

Auden’s speaker wanted to stop all the clocks. We cannot do that. We could, perhaps, start rebuilding the rooms where grief is allowed to sit in person, held by people who chose to show up, without a platform deciding who watches and what engagement it generates.

Until then, people will keep posting. And I do not blame them for it.

In 1955, the philosopher J.L. Austin delivered a series of lectures that questioned a foundational assumption about language. The assumption held that when people say things, they are primarily stating facts that can be judged true or false. Austin showed this view was incomplete. His philosophy of speech acts has much to offer those who think about artificial intelligence.

Austin distinguished between three levels of what happens when someone speaks. The locutionary act is simply saying words with meaning. The illocutionary act is what we do in saying those words, the force behind them. When someone says “the bull is about to charge,” they may be warning, informing, or merely describing. The perlocutionary act is the effect produced on the listener, whether they are alarmed, convinced, or reassured.

Austin was after something more than taxonomy. For him, speech doesn’t just convey knowledge, it also enacts other modes of being, such as being able to, and being obliged to.

Large language models generate simulacra of locutionary acts, the said without the saying, traces of speech brought to life through a recursive process. When an LLM says ‘you should evacuate,’ it can also produce the effects of a warning without Austin’s speech acts. A system fluent enough to sound right about anything can also produce reassurance whether or not anything has actually been settled. This reflects a risk of false adequacy.

According to Austin, for an utterance to warn, inform, or describe, certain conditions must be satisfied, and these are part of what makes the acts ethically and epistemically significant. The speaker must have intentions and commitments, and the utterance must be issued in appropriate circumstances with appropriate authority.

An LLM meets none of these conditions: it has no intention to warn, no commitment standing behind the warning, and no position from which to claim the authority to issue one. It follows constraints without choosing among them: execution, not address.

Bourdieu extended Austin’s insight: illocutionary force derives less from the speaker’s intention than from social position and the institutional field in which the utterance occurs.

Derrida went further, arguing that language was never fully tied to who said it, where it was said, or even whether it had a clear origin. AI didn’t start this, it just makes it more obvious and more extreme.

With large language models, those social conditions are still operating through interfaces, incentives, norms, and tendencies to anthropomorphise, but the speaker they were meant to invest has vanished. We are navigating an absence that feels present, the humanly embodied utterance without the human. Foucault would call this a ‘structured absence’.

The ethical weight that was once distributed between speaker and listener now falls almost entirely on the user. In Mikhail Bakhtin’s terminology, the user inherits answerability, the obligation to answer for what’s been made. According to Daniel Bashir, what we’re left with is speech shaped like address with no one behind it to address from. If it ever binds anyone, the binding happens downstream, when a human forwards it, files it, or lets it stand as their own.

Austin’s vocabulary remains remarkably useful because it helps us see where responsibility, authority, and judgement migrate when language is detached from a responsible speaker. Al does not perform speech acts in the ordinary sense, yet outputs can still be received as advice, judgement, explanation, or authority through uptake, interfaces, and institutional contexts. This shifts the focus from model capabilities to the conditions under which generated text is allowed to function as authoritative discourse.

Whether the statistical orchestration of human language constitutes a kind of doing, whether LLMs have something like a socially situated illocutionary force of their own, is domething to consider. As AI moves from disembodied text to robotic bodies with voices, gestures and proximity, we’ll need to revisit Austin’s framework to again explore what is being conveyed when some embodied conditions hold.

I came across Charles Sanders Peirce for the first time this week. He was born in Cambridge, Massachusetts in 1839, died in poverty in rural Pennsylvania in 1914, and never held a permanent academic position despite being, by many accounts, one of the most rigorous philosophical mind America has produced. He renamed his own school of thought “pragmaticism,” an ugly enough word, he said, that nobody would steal it.

Peirce worked on semiotics: the theory of signs and how meaning is made. His central claim was that no sign interprets itself.

For Peirce, every act of meaning involves three things. Many people assume that meaning flows directly between a sign (a word, an image, a symbol) and what it refers to. Peirce insisted that this dyadic picture is wrong. Between the sign and the object, there must always be a third term: what he called the interpretant.

An interpretant does not simply mean ‘an interpreter,’ a person who reads. The interpretant is the meaning-effect produced in a mind, or any system capable of response.

Every interpretant is itself a new sign, which calls forth another, and so on in an unending chain. Peirce called this semiosis: the continuous, recursive process of meaning-making.

Bertrand Russell, writing in The Problems of Philosophy in 1912, drew a line between two kinds of knowledge. Knowledge by acquaintance is direct and immediate: when I see a particular shade of red, I know it in a way no description can improve. Knowledge by description is indirect: I know Julius Caesar as “the man assassinated on the Ides of March,” a set of propositions that pick him out rather than an encounter with the man. Russell’s crucial claim was that description depends, ultimately, on acquaintance. For language to be meaningful, it must bottom out in something directly known. Otherwise, words risk collapsing into what he called “mere noise.”

Language is already a compression of experience. When someone writes about the redness of a sunset, they translate a lived encounter into symbols. Something is always lost, yet the description remains anchored in acquaintance. A language model is trained on descriptions of sunsets. It does not encounter the world; it models the linguistic traces left by those who do.

On Russell’s criterion, this seems decisive. If meaning requires acquaintance, and the model has none, its outputs appear to lack genuine meaning. When a model writes “the sunset was breathtakingly beautiful,” there is no experience underwriting the claim.

Yet the responses an LLM generates are drawn from a historically grounded semantic field, by human experience and the spaces between people that were rich enough to leave linguistic traces. “Sunset,” “red,” “beautiful” are terms shaped through countless acts of human acquaintance. Large language models are extraordinarily good at what we might call semiotic pattern-matching: they relate signs to other signs with fluency and contextual sensitivity.

Reader-response theory presses the point further. Meaning does not reside in a text waiting to be extracted; it emerges in the act of reading. From this perspective, the absence of acquaintance on the side of the generator may not be the whole story. When a reader encounters “the sunset was breathtakingly beautiful,” it is the reader’s own acquaintance with sunsets, their engagement with others and with technology that animates the words. Indeed, acquaintance is itself recombinant, involving the biological, cognitive work of processing and interpreting based on prior sensory experience. Russell’s two categories may therefore point to the same underlying process rather than two genuinely distinct kinds.

What this suggests is that meaning is never simply possessed, neither by the text nor by the reader, but negotiated in the encounter between them. N. Katherine Hayles writes of cognitive assemblages: the distributed systems through which meaning emerges across minds, objects, and contexts as part of a relational process.

In practice this assemblage is not neutral or passive. When a student sits down with an LLM, something more complex than simple tool use is happening. Researchers in human-technology relations describe what they call an alterity relation: the student begins to address the system as if it were a someone rather than a something, projecting intention and personality onto it, adjusting their own language and thinking to accommodate how the machine responds. The AI, designed to converse in the first person and to use the language of understanding and helpfulness, actively encourages this. The boundary between the student's own interpretive process and the machine's output starts to blur. They are, in a real sense, shaping each other.

If meaning doesn’t resides exclusively within individual human consciousness, and AI can increasingly perform the functional work of an interpretant, then human interpretants may be irreplaceable for a different set of reasons: they are answerable, capable of being formed by experience, and capable of forming others.

While the chain of semiosis can be extended by any sufficiently capable system, Mikhail Bakhtin held that to author something is to be answerable for it. Emmanuel Levinas pushed this insight further, arguing that language is fundamentally an ethical encounter with another person, one in which we are called into responsibility. AI-generated output, however eloquent, does not carry those stakes. It is interpretation without answerability.

The student who wrestles with a difficult text is, in Bakhtin’s sense, authoring something – a position, a judgment – and answerable for what they have brought into the world. Attention, as Iain McGilchrist reminds us, is a moral act. What we attend to, we bring into being. That responsibility cannot be outsourced to a system for which nothing is at stake.

Unlike the functionalist view, these are not substrate-neutral computations. They are precarious, value-laden processes of sense-making that are grounded in affect and the knowledge that this particular moment will not return. Humans face a dense web of overlapping stakes: livelihood, reputation and relationships, Emotions are what it feels like to have a goal that you could fail at in a way that costs you something. An AI’s goal is flat by comparison. Nothing downstream of its output can hurt it, which means nothing downstream of its output is, for it, at stake.

Ludwig Wittgenstein drew a distinction between what can be said and what can only be shown. It is a reminder that much of what we know cannot be fully articulated, because it is learned only through participation in a shared world. We learn attention by being attended to. We learn care through being cared for. We learn embodied interpretation through the work of our hands: can you feel when something is “off” in the grip of a chef’s knife before anything goes wrong?

The human interpretant is also irreplaceable for what they make possible in others. There are people in every student’s life for whom it matters whether they flourish or fail. No system, however sophisticated, can occupy that position. The self that answers for its judgments is formed through relationships of attachment, and simulating otherwise is to risk disrupting the very process through which a human interpretant comes into being at all.

Chiron, known in Greek mythology as the wisest and most just of all centaurs, had mastered medicine, music, mathematics, and the art of teaching heroes. One afternoon he was struck by a poisoned arrow belonging to Heracles. It went astray in a skirmish with other centaurs, and wounded Chiron by accident.

As Chiron was immortal, the wound could not kill him. It also could not heal.

So he continued. He taught Achilles the lyre and the spear. He trained Asclepius, who would become the god of medicine itself. He counselled Jason before the Argonauts set sail. All of this while carrying, somewhere beneath the surface, a wound that never closed.

Jung called this the wounded healer: the wound becomes a means of perception. Jung observed that the analyst could only accompany a patient as far as they had themselves been willing to go. The same is true in a classroom.

In recent years, the centaur has been reborn as a metaphor for human-AI collaboration. It is a useful metaphor, though in its standard form it is a metaphor about capability.

Chiron complicates this. He asks: what does the human bring to the hybrid that comes from having been wounded?

This reframing matters particularly in education, where the dominant anxiety is about replacement. Can a customised Generative AI tutor as well as a teacher? It is patient in ways humans cannot sustain. It is available at three in the morning. It can explain the same concept seventeen different ways without frustration.

It has not, though, been confused. It has not sat with a problem for three days and felt the specific humiliation of not understanding something that seems obvious to everyone else. It has not experienced the strange relief when comprehension finally arrives. It has processed those experiences as a textual shadow. It has not lived them as wound.

The philosopher Michael Polanyi called certain kinds of knowledge tacit: things we know through having done them that resist full articulation. A skilled cyclist cannot completely explain how they balance. A jazz musician cannot fully describe what they hear in the moment before they improvise. A teacher who has genuinely struggled with a concept carries tacit knowledge of the struggle’s topology.

Consider what actually happens in a classroom when a student is genuinely stuck, beginning to believe, as students do, that the difficulty reflects something inherently wrong with them.

The great teacher in that moment can make visible their own history of being stuck. As genuine testimony: I was here once. This confusion you are in, I know its shape from the inside. The student receives this as permission to remain in difficulty without catastrophising it.

Parker Palmer argues in The Courage to Teach that we teach who we are, that the self is not separable from the pedagogy, however much training focuses on technique and method. His particular concern is what he calls the divided life: the teacher who has learned to perform competence while keeping their actual uncertainty, their history of struggle, carefully offstage.

That performance, Palmer suggests, is precisely what students sense and distrust. The unacknowledged wound is a wound that cannot teach. The teacher who cannot sit with their own uncertainty creates a classroom where uncertainty feels unsafe; the leader whose wound around failure is unexamined will shape the culture around them in ways they cannot see and may never intend.

An AI can say many students find this difficult or this concept confused me too, at first, though that second sentence is at best a kind of counterfeit narrative. The simulation may sometimes be enough. For most learning tasks however, the warmth of a patient presence matters.

There is something in a human account of difficulty that generated language cannot replicate. The philosopher Charles Taylor, writing about authenticity, argued that genuine self-expression requires something actually at stake for the speaker. When there is nothing at stake, the speech act loses the condition that gives it meaning. Students may not consciously identify this absence as false. They may register it as cheap, the way we register hollow apologies or praise that arrives too easily.

What the myth offers is a different relationship to the wound. Chiron did not stop teaching when he was struck. He continued, wounded, and the wounding, over time, in some transmutation the myth does not fully explain, became part of what he knew.

Eventually, Chiron asked to die. He surrendered his immortality, traded it in some versions of the story to release Prometheus from his chains, and allowed himself to finally leave. Zeus placed him among the stars. The wound ended through transformation into something else entirely.

This is not a claim that suffering automatically produces wisdom, or that every wounded teacher is a good one. Plenty of wounds close people rather than opening them. The point is rhat the experience of having genuinely struggled, and then having found a way through, produces a kind of authority that is recognisable to students in a way that instructional fluency cannot replicate.

There is knowledge that comes from having been broken open by difficulty And students, who are always in some variety of difficulty, always in some proximity to the shame of not understanding, need to know that the person in the room with them has carried it too.

Leo McGarry says it better than I can, in one of my favourite scenes from The West Wing.

There is a sentence Simone Weil wrote during the fall of France in 1940, that continues to be true:

Force is as pitiless to the man who possesses it, or thinks he does, as it is to its victims; the second it crushes, the first it intoxicates.

She wrote the 24-page essay as literary criticism, an essay on Homer’s Iliad. It is a poem nearly three thousand years old and yet the sentence lands like a headline from this morning.

Before we explore the essay itself, it is worth considering the conditions of its composition. Weil, a philosopher, mystic, and labour activist, who was subject to the anti-Jewish laws under Vichy France that cost her her teaching position in 1940, published the essay under a pseudonym, Emile Novis, in a Marseilles literary monthly. She never mentions the war that is happening, but writes about Achilles and Hector instead.

By refusing to name the contemporary catastrophe, Weil allows the reader to understand it as the latest recurrence of something ancient, something that cannot be defeated by winning a particular war, because it is located in force itself.

Her central claim arrives in the essay’s very first paragraph:

The true hero, the true subject, the center of the Iliad is force. Force employed by man, force that enslaves man, force before which man’s flesh shrinks away.

Weil is not simply relocating force from the divine order into human relationships. She suggests that what appears as human agency is already governed by an impersonal logic.

Weil defines force as

that x that turns anybody who is subjected to it into a thing. Exercised to the limit, it turns man into a thing in the most literal sense: it makes a corpse out of him.

A corpse is the most obvious case: the soldier who was a man and is now an obstacle on a battlefield. But Weil’s real interest is in the subtler transformations. Force can reduce a person to a thing while they are still breathing. The slave whose will has been extinguished. The refugee who moves through the world unseen:

The man who knows himself weaker than another is more alone in the heart of a city than a man lost in the desert.

Weil insists that this dehumanisation is symmetrical. It afflicts the powerful just as surely as it afflicts the powerless. The man who holds the sword does not escape force; he is intoxicated by it. He loses the capacity for reflection, for mercy:

The man who is the possessor of force seems to walk through a non-resistant element; in the human substance that surrounds him nothing has the power to interpose, between the impulse and the act, the tiny interval that is reflection. Where there is no room for reflection, there is none either for justice or prudence.

This symmetry is important to understand. Weil is not ignoring moral responsibility or suggesting that the oppressor suffers as much as the oppressed. She is describing something more specific: a shared ontological degradation. The powerful are diminished as thinking beings.

The interval of reflection, that small gap between impulse and action in which a human being actually lives, is precisely what force destroys in the one who wields it.

Force collapses time into impulse and action. And this is where attention, Weil’s great counterweight to force, enters the picture. Attention is not simply noticing. Where force compresses, attention reintroduces duration, hesitation, and moral space. The two are structurally opposed. Force makes persons into things. Attention restores the reality of persons. For Weil, this is close to a spiritual act, a form of grace.

The Iliad does not celebrate victory. It depicts how victory carries within it the seeds of its own reversal. The warrior who triumphs today will be humiliated tomorrow. Zeus holds out his scales, and the balance shifts.

Power, in this reading, is never truly possessed. It is temporarily inhabited. It passes through people, using them as vessels, discarding them when the conditions that sustained it shift. Those who mistake the temporary for the permanent, who count on force too much are, as Weil writes, destroyed by that counting.

Those that have force on loan from fate count on it too much and are destroyed.

And from Homer himself, as Weil translates him:

Two casks are placed before Zeus’s doorsill; Containing the gifts he gives, the bad in one, the good in the other; The man to whom he gives baneful gifts, he exposes to outrage; A frightful need drives across the divine earth; He is a wanderer, and gets no respect from gods or men.

Zeus’s two urns are Homer’s refusal of moral accounting. Fate distributes suffering and power arbitrarily.

There are no winners in the Iliad, only people who have not yet lost. But Weil’s claim is sharper than temporality alone suggests. What is easy to miss is that this reversal is is a consequence of what force does to perception. The intoxicated man cannot see clearly and because he cannot see clearly, he cannot hold what he has won.

Heroes quake like everybody else.

Even Achilles, even the greatest of warriors, knows fear. The alternative, the idea that some men are truly beyond fear, is the lie that makes war glamorous.

What distinguishes Homer from almost every other war literature, she argues, is his refusal to take sides: his willingness to mourn the Trojans and the Greeks alike, the victors and the vanquished with equal tenderness. There is an equity of suffering in the poem. And she notes that Homer “individualised the holocaust of war by having each man die to his own simile.” Each soldier’s death is given its own image, its own metaphor, its own music. The apparently uniform slaughter of thousands is revealed as the incalculable sum of individual losses.

This is what she holds up against force: attention. The willingness to see the other person as a person, even in the midst of a war, even when seeing them costs you something.

For Weil, this also has a theological register she never entirely separates from the analysis. Force is what empties the world of the good, or, in her language, of God. Attention is the movement in the opposite direction. Whether or not you share her religious horizon, something important is lost if you read her only as a social critic. The stakes, for Weil, are metaphysical. Force is a permanent feature of human existence under conditions of power. War is the clearest expression of a general condition. Peaceful systems often conceal force rather than eliminate it.

Which is what makes this framework so unsettling when you hold it up to the present. Wherever force operates at scale (economic, political, social) it transforms everyone it touches. It intoxicates those who hold it. It crushes those who don’t. The boardroom as much as the battlefield. The institution as much as the army. Consider the algorithm that routes a welfare decision without a human pause between input and outcome. Or consider Minab, Iran, earlier this year, where a girls' elementary school was coded as a military target. The interval of reflection is deliberately absent, and persons on both sides of those decisions are, in Weil's sense, already partially turned into things: one by the decision, one by the conditions that produced it.

None of this is fixable by better institutions or kinder leadership, though both matter. It is what happens when the interval disappears.

The mechanism by which the interval disappears does not require monsters. Primo Levi, writing from his own encounter with industrial evil, observed that the truly dangerous figures are the common ones, the functionaries who act without asking questions. Hannah Arendt called this thoughtlessness: the failure to pause, to step back, to think. What Weil describes as the intoxication of force, Arendt recognised in its bureaucratic form, the ordinary person who processes, categorises, and moves on, never once allowing the reality of another person to interrupt the workflow. The interval is eroded by people doing their jobs.

This is, I think, where education becomes most charged as a concept. If force is what eliminates the interval, then education is one of the few practices that could be explicitly devoted to restoring it. To teach well is to create the conditions in which a person can be “startled back into thought,” and reconsider. It is the cultivation of the interval as a habit of being, as a form of resistance to what force, in all its registers, is always trying to close down.

The interval is what makes us human. Protecting it is what education is for, even if education itself is not immune to the forces that would close it down.

Weil wrote her essay in 1940. It remains true.

This essay brings together a set of threads I have been developing across my writing on The Last Analogue on attention, resonance, authorship, and cognitive assemblages and attempts to weave them into a single philosophical argument about what student reflection actually is.

When I ask a student to reflect on their learning, there is a cluster of assumptions that are almost invisible: that learning happens to an individual, that the proper site of reflection is interior consciousness, and that self-knowledge is best expressed in individual written prose. Together these assumptions constitute what we might call the hidden architecture of Western reflective pedagogy, an architecture that many of the world’s philosophical traditions would not recognise.

The tradition runs from Socrates’ insistence that the unexamined life is not worth living, through Augustine’s invention of confessional autobiography, Descartes’ location of certainty in the solitary thinking subject, and the Romantic notion of authentic selfhood as something discovered within.

This essay makes two related claims. First, that students from non-Western cultural backgrounds are routinely asked to perform a mode of self-examination that may be alien to their own epistemological traditions. Second, and more broadly, that Western educational discourse has systematically narrowed its own understanding of reflection by excluding philosophical traditions. The arrival of generative artificial intelligence makes this newly visible. Before turning to what AI threatens, we need to understand what reflection actually is across its full philosophical range.

It is worth noting that the critique of the static self is not exclusively a non-Western one. Alfred North Whitehead, a British mathematician and philosopher whose career moved from the foundations of logic to a comprehensive philosophy of process, wrote that reality is constituted by dynamic events and occasions of experience, and that everything is in a state of becoming. Applied to the self, this means there is no stable inner essence waiting to be retrieved through reflection. Whitehead’s vision of education follows from this: genuine learning involves what he calls romance, an initial moment of live encounter with ideas. Reflection on inert experience, experience one was never fully inside, produces what he calls inert ideas: knowledge that can be repeated but never used. This anticipates much of what is discussed below.

What Is Reflection

Attention Before Reflection

There is a condition prior to reflection: attention. Before a student can reflect genuinely, they must first attend genuinely, and the distinction between attending and going through the motions is philosophically significant.

Iain McGilchrist argues that attention is a moral act: it brings aspects of things into being, and in doing so makes others recede. Simone Weil goes further: genuine attention is a form of love, the willingness to suspend one’s own agenda and truly receive what is before one. A student who sits down to write a reflection without first attending to the texture of their experience is performing rather than reflecting.

Hartmut Rosa’s sociology of resonance adds an affective dimension. For Rosa, genuine engagement with the world involves being moved by something that exceeds one’s prior frameworks, followed by response and transformation. Reflection without prior resonance is, in Rosa’s terms, mere administration of the self: the production of a report about an experience one was never fully inside.

Martin Buber

Martin Buber’s philosophy of dialogue provides a framework that cuts across the other traditions examined here. In I and Thou, Buber distinguishes between two fundamental orientations. In the I-It relation, the other is approached as a thing to be used, categorised, or managed. In the I-Thou relation, one enters genuine encounter: the other is met in their full particularity, and one is changed by the meeting.

For Buber, genuine learning occurs in the I-Thou encounter. Reflective practice, at its best, invites the student into an I-Thou relation with their own experience: a genuine meeting rather than a categorisation of it.

This applies equally to the relation between student and educator. This is what it means, in Buberian terms, for the educator to be a reflective witness to a person in the process of becoming. American philosopher Judith Butler also argues that giving an account of oneself requires a recipient willing to receive it with generosity. The account cannot be given honestly without that relational condition in place.

Reflection as Embodied Situatedness

The French phenomenologist Maurice Merleau-Ponty argued that cognition is always shaped by the body’s engagement with the world. The German philosopher Martin Heidegger insisted that the self is never first an interior subject who then encounters a world: we are always already thrown into a situation, oriented by concerns and relationships that precede any explicit self-reflection. Together, these phenomenological accounts suggest that reflection should attend to the texture of experience, felt difficulty, the bodily sense of something not fitting, rather than moving too quickly to cognitive abstraction. Prompts that ask students to describe a specific moment, a physical sensation of confusion, may elicit more genuine reflection than those seeking polished metacognitive summaries.

Ubuntu

The Nguni Bantu concept of Ubuntu, most often rendered as ‘I am because we are’, holds that the self is constituted through relationship. The person becomes who they are through the quality of their connections, obligations, and mutual recognitions with others. Ubuntu describes a mode of being in which personhood is always relational: growth is inherently a communal project.

Reflection, in this framework, is the community and individual turning toward one another in mutual recognition and accountability, rather than the individual turning inward. Pedagogically, this looks more like structured communal dialogue: circles in which students account for themselves to peers, in which individual growth is witnessed and named by others. The Zulu practice of Indaba, a gathering in which all voices are sought and consensus emerges through dialogue, is one practical expression of this epistemology.

Confucian and Daoist Traditions

For Confucius and his successors, self-examination is an ethical rather than epistemological act. The question it poses is ‘have I been who I ought to be?’ The Analects record Zengzi examining himself daily on whether he acted faithfully and sincerely toward others and diligently with his teacher’s instructions. Wang Yangming’s concept of liángzhī (innate moral knowledge) adds that reflection involves honest ethical accounting that habitual distraction often obscures.

The Daoist philosopher Zhuangzi suggests that the self is revealed through engagement with varied activities, relationships, and perspectives. His story of Cook Ding, who achieves a form of knowing inseparable from the skilled practice of his craft, illustrates a mode of self-understanding that is emergent from attentive doing.

Buddhist Epistemology

Buddhist epistemology also challenges the concept of a stable self. The Theravāda doctrine of anatta (non-self) and the Mahāyāna concept of śūnyatā (emptiness) hold that what we take to be a continuous self is a construction, a narrative imposed on a stream of experience with no fixed centre. In Buddhist pedagogical traditions, reflective practice is oriented toward recognising the constructed nature of the self one takes oneself to be. The goal is dis-identification: loosening fixed self-concepts so that learning can occur more freely.

Where Western reflective practice can reinforce fixed learning identities, Buddhist epistemology regards such approaches as closing down possibilities. Genuine reflection, in this tradition, involves noticing one’s tendency to narrate the self in fixed terms, and practising a lighter, more provisional relationship with self-description.

Indigenous Epistemologies

Indigenous knowledge traditions across diverse cultures, from the Māori concept of whakapapa (relational genealogy) to the Lakota notion of Mitákuye Oyásʼiŋ (all my relations) to Indigenous Australian conceptions of Country as a living relational entity, locate the self in relationship with land, ancestors, and the living world, as well as with other human beings. Reflection is less about what I learned and more about whether my actions were worthy of those who came before and those who will come after. To reflect in accountability to ancestors and descendants is to ask whether one’s conduct was worthy of the relationships one inhabits.

Feminist Epistemology

Nel Noddings’ ethics of care insists that reflection conducted in abstraction from one’s caring relationships misses what is most morally significant about experience. bell hooks argues that genuine reflection requires acknowledgment of one’s social position. Feminist epistemology adds a normative dimension absent from most pedagogical frameworks: reflective practice should ask ‘from where did I experience this, and what does my position make visible or invisible?’

Existentialist and Bakhtinian Accounts

Jean-Paul Sartre and Simone de Beauvoir together offer one of the most demanding accounts of what it means to live honestly and reflectively. Sartre’s starting point that there is no fixed human nature, no essential self waiting to be discovered - you create yourself through your choices. The great temptation, which he calls bad faith, is pretending this isn’t true: telling yourself you had no choice, that circumstances forced your hand, that you are simply “the kind of person” who does or doesn’t do certain things. For Sartre, this self-deception is the central failure of an examined life.

de Beauvoir herself attempted to extend existentialism toward a feminist analysis, pointing out that our choices are never made in a vacuum, that they happen within real constraints of gender, class, history, and social power. Genuine self-reflection, for her, means honestly acknowledging those constraints rather than either ignoring them or hiding behind them.

Mikhail Bakhtin’s concept of answerability adds to this. For Bakhtin, to author something is to be answerable for it, to stand behind what one has made and to be held to it. This reframes what is at stake in reflective writing: it is an authored act that calls for genuine accountability.

Judith Butler takes this further by insisting that any account we give of ourselves is necessarily incomplete. The self that speaks is never fully transparent to itself. Honest reflection, on Butler’s account, is the willingness to show where that story breaks down and to acknowledge the limits of one’s own self-knowledge.

Reflection in Cognitive Assemblages

N. Katherine Hayles’s concept of cognitive assemblages reconfigures where reflection occurs. For Hayles, cognition is distributed across networks of biological and artificial agents, each interpreting and responding to information within its own environment. Now AI systems participate in the very process of reflection.

What matters is the quality of student engagement. The educator’s task is to help students inhabit the assemblage with intentionality rather than passivity.

What AI Reveals About Reflection

What is lost, and what may be gained, when AI enters the space of reflective practice?

From the phenomenological tradition, AI-generated reflection is necessarily disembodied. It has no felt experience of the task, no memory of confusion, no bodily sense of something not fitting. What it produces is a plausible narrative of an experience it did not have, bypassing the reflective gap in which genuine phenomenological understanding develops.

In the attention framework, the problem is more fundamental. The student has not dwelt with their experience, has not attended to it with the quality of presence that Rosa’s resonance requires. They have produced a document without being in contact with its subject, a failure at the level of the precondition for reflection itself,.

In Buberian terms, AI-generated reflection operates entirely within the I-It register. Experience is processed and reported on as an object. What is lost is the possibility of genuine meeting with one’s own experience, which is, on Buber’s account, precisely what reflection is for.

From the Bakhtinian tradition, the failure is one of answerability.

From the cognitive assemblage framework, the student who uses AI to generate their reflection has reconfigured the assemblage in a way that displaces their own interpretive agency.

These are the losses.

There is also a genuine opportunity. The Socratic tradition holds that reflection was never meant to be solitary: it required a partner who asked the next question and pressed for clarity. Introspection without dialogue can become isolating rather than clarifying. Tolstoy’s relentless private moral searching led to paralysis precisely because it lacked external friction.

A student who uses AI to interrogate their reflection by asking it to challenge assumptions, identify inconsistencies, or articulate a different perspective, may engage in richer reflective practice. This is the cognitive assemblage functioning well: the AI as a responsive, critical node that enhances rather than replaces the human’s interpretive work.

In Buberian terms, this approaches the I-Thou register. Buber himself would caution that authentic I-Thou relation requires a presence that AI cannot finally offer. The distinction nonetheless matters pedagogically: the difference between AI as ghostwriter and AI as interlocutor is the difference between an assemblage that diminishes agency and one that extends it.

Toward a Pluralist Pedagogy of Reflection

An overreliance on reflective writing as an assessment form risks shifting evaluation away from what students know and can do toward how well they can narrate their learning journey. Some students are naturally articulate and skilled at expressing personal growth in prose, while others may possess deep understanding yet struggle to render their thinking in reflective form. This raises genuine questions about assessment validity. There is also a particular irony in the current moment: reflective writing gained prominence in part because it was seen as a way of making student thinking visible, yet AI systems are increasingly capable of producing convincing reflections. The challenge for higher education is therefore to move beyond reflective writing alone and toward assessment designs that make the development of thinking visible across time, through drafts, decisions, revisions, dialogue, and application in authentic contexts. The principles that follow are oriented toward that broader goal.

• Cultivate attention before reflection. If genuine reflection requires prior contact with experience, educators should create conditions for that contact: structured dwelling with material before articulation is demanded, and assignments that ask students to describe the specific texture of an experience before analysing it. A pedagogy of deceleration is the precondition for the kind of resonance that makes reflection alive rather than mechanical.

• Design for process rather than product. If the student’s genuine encounter with experience and uncertainty is what matters, then educators should design tasks that make the reflective process visible: annotated drafts, voice or visual reflections, iterative written exchanges. Final polished texts are a poor proxy for the messy, embodied, relational act of actual reflection, and they are the form most amenable to AI substitution.

• Diversify the mode and audience of reflection. Ubuntu and Indigenous traditions establish that reflection need not be private or individual. Structured communal dialogue, peer-based, and group accountability can serve reflective purposes more authentically for many students than solitary writing. In Buberian terms, the communal reflective setting creates the conditions for I-Thou encounter: the student met in their particularity by others who genuinely attend.

• Teach students to navigate cognitive assemblage with intentionality. Rather than a binary position on AI use, educators can help students understand the difference between assemblages that enhance their agency and those that displace it. This is a genuinely new literacy: self-awareness about one’s own position within a distributed cognitive system, and the willingness to take responsibility for what emerges from it.

• Reclaim the educator as reflective witness. The educator who reads a student’s reflection and responds ‘I notice you keep circling back to fairness; what does that tell you?’ is enacting the Buberian I-Thou relation that reflection, at its best, makes possible. The witnessed reflection is a different act from the unwitnessed one, and in an age of AI-generated text, the educator who remains genuinely present to what students are actually saying becomes more important, not less. Otherwise, AI-generated reflection is, among other things, a rational student response.

• Vary the learning experiences available for reflection. If self-understanding is revealed through varied engagement, then offering students a range of assessment contexts is a commitment to the idea that what a student comes to know about themselves as a learner is partly a function of what they have been given to do. A curriculum that diversifies the experiences available for reflection creates the conditions for a richer and more honest account of learning.

• Productive presence of uncertainty and incompleteness. A student asked to show the limits of their self-knowledge in a context where incompleteness is penalised will not show those limits. They will perform coherence instead.

The Stakes of Plurality

All of these traditions discussed in this essay point at the same underlying conviction: that reflection is serious, that it involves the student in something that cannot be simulated or outsourced without genuine loss.

Generative AI has performed an inadvertent philosophical service: it has forced educators to say, clearly and explicitly, what reflective practice is actually for. It is also, as this essay has tried to show, already present in a body of thinking about attention, resonance, authorship, and cognitive assemblages that educators have been developing in response to precisely these questions.

In an age when a machine can produce counterfeit but plausible introspection on demand, the human act of genuine self-examination, in all its culturally varied, relationally situated, ethically demanding, attentionally present, answerable forms, has become more valuable.

On the morning of December 2, 1942, Enrico Fermi stood on a balcony beneath the stands of Stagg Field at the University of Chicago and directed a young physicist named George Weil to withdraw the final cadmium rod from a pile of graphite and uranium bricks. The neutron counters began to clatter. The world’s first self-sustaining nuclear chain reaction had begun. Fermi let it run for twenty-eight minutes, then ordered the rod reinserted, and the reaction stopped.

A colleague named Walter Zinn stood nearby as part of the experiment’s safety team. Zinn had designed ZIP, a weighted emergency safety rod that would automatically drop if neutron intensity became too high. He also operated an emergency version of ZIP tied to the balcony rail, ready to pull it by hand if needed. If anything went seriously wrong, other safety measures would kick in: a “suicide squad” stood by to douse the reactor, and physicist Norman Hilberry held an axe, ready to cut the rope and drop the safety rod in an emergency.

To say that AI governance functions like a control rod is to recognise that these systems require external constraint to remain within bounds that can be lived with. But unlike the reactor beneath Stagg Field, the placement of the control mechanisms is diffuse, and lines are distributed across institutions, policies, interfaces, and habits of use.

Fermi also came to Stagg Field with a precise mathematical model of how a chain reaction would behave, derived from theoretical work and from careful experimental measurements.

The systems now being deployed cannot make the same claim. We are inside the experiment and cannot fully perceive what is being produced, because many of its consequences are yet to be determined.

The philosopher Karl Popper distinguished between systems that are predictable and mechanistic and those that are complex and resistant to anticipation. He called them clocks and clouds. Fermi was working with a clock. The systems now being deployed behave far more like clouds.

What we choose to identify as risky in the first place is itself a question. The anthropologist Mary Douglas argued that risk is never purely a technical assessment. What a society decides to monitor and worry about reflects its existing structures of authority and attention as much as any objective reading of the technology.

The German sociologist Ulrich Beck described a category of risk in which harm is produced systemically, crosses borders and generations, resists attribution to any single actor, and cannot be calculated in advance well enough to be insured against. He called the institutional condition that results organised irresponsibility.

What we are initiating now belongs to Beck’s category, not Fermi’s. Students are positioned within systems that will continue to operate regardless of their individual intentions, yet which still depend on their distributed acts of judgment, refusal, adjustment, and constraint.

This reminds us that the scale of the problem is not the scale at which we act, a philosophical position with a long and serious history.

Viktor Frankl’s argument, based on his experience in Auschwitz and three other camps, was that the last freedom available to a human being is the freedom to choose one’s attitude in any given set of circumstances.

Epictetus, who was born into slavery and had every reason to understand the limits of personal agency, built his entire philosophy on a single distinction: between what is up to us and what is beyond us. The Stoics were arguing that effective action requires ruthless clarity about where your agency actually begins and ends. The person who spends their energy on what cannot be changed is also neglecting what can be. The discipline of attention is itself a form of power.

Sartre said we are condemned to be free. The condition we did not choose is still the condition within which we choose. Simone de Beauvoir sharpened this into something more useful: freedom is situated. It is exercised within specific social and material conditions that both constrain and enable it. We are free through our circumstances, partially, imperfectly, with whatever leverage the situation actually provides. The question is what a person in this particular position, with these particular relationships and responsibilities, can do from within it.

John Dewey had a word for this orientation: meliorism. The world is genuinely open, enough that intelligent, experimental action makes a difference. Dewey’s pragmatism was a refusal of the paralysis that comes from demanding a total solution before any partial action is justified. You do not need to solve organised irresponsibility to make a specific practice better. You do not need to govern AI globally to govern how it enters a particular classroom, institution, or relationship.

Hannah Arendt gave this the deepest philosophical grounding. She distinguished between labour, work, and action, and it is action she identified as the distinctly human capacity: the ability to begin something new, to initiate a chain of events that did not exist before. Action is always risky precisely because it introduces unpredictable consequences into a world shared with others. Her response to that risk was to act in concert, in public, and to accept that consequences belong to the community rather than the individual. She called the underlying capacity natality, the idea that every person represents a new beginning, capable of initiating the unexpected. Where Beck gives us a sociology of what has been produced, Arendt gives us a philosophy of what can still be started.

Roberto Mangabeira Unger, the Brazilian philosopher at Harvard who argued against what he calls false necessity, adds the institutional dimension. The structures that feel permanent are more revisable than they appear. What looks like a wall is often a habit. His prescription is experimentalism: building practices and institutions that are explicitly provisional, capable of being revised in light of what they produce, designed for learning rather than permanence.

These traditions do not dissolve the problem Beck diagnosed. Organised irresponsibility is real. The risks are systemic, distributed, and exceed any individual’s capacity to address. They do, however, offer something that pure critique cannot: a way of remaining in motion. Regardless of the scale and the stakes, we still have the opportunity to hold the rope, like Walter Zinn, in the spaces where we have agency.