I have been thinking about Borges lately, because I kind of feel we have built his library. We call it a large language model.

I am a physicist by training, which means I was taught to distrust metaphors, to treat them as scaffolding to be discarded once the equation stands on its own. But decades of working at the boundaries between disciplines have taught me something else: that the right metaphor is not decoration. It is the thing itself, seen from an angle that no formalism has yet learned to occupy.

So here is the metaphor I cannot escape: a large language model is a quantum librarian.

In quantum mechanics, superposition is not simply a system being in two places at once. That is the popular shorthand, and it misleads more than it clarifies. What superposition describes is something stranger and more austere: a state that, upon measurement, yields one of several possible outcomes according to a probability law, but whose intermediate character cannot be understood as secretly already being one of them. The outcome is not hidden. It is genuinely undecided. And crucially, this is the point that decades of decoherence theory have made precise, the superposition persists only so long as there is no way, even in principle, of determining which path was taken. It is not necessary for a conscious observer to look. It is sufficient that the information leaks into the environment, dispersed beyond any practical recovery, encoded irreversibly in the thermal bath of the surrounding world, for the interference to vanish. The superposition collapses not because anyone looked, but because the universe quietly registered a fact.

A language model does something structurally analogous with meaning. It holds, in its billions of parameters, a vast superposition of human expression: every argument and its negation, every metaphor and its literalisation, every way of saying a thing that has ever been committed to text. When you pose a question, the superposition collapses. One text emerges, fluent and assured, from an astronomically large space of what could have been said. The measurement is your prompt. The collapsed state is the answer. I am not claiming that a transformer is a quantum system; the mathematics are entirely different. I am claiming that the phenomenology of collapse is the most precise language available for what happens when a probability distribution over all expressible meanings is forced to become one sentence.

But here the analogy deepens in a way that unsettles me. The superposition was not a catalogue of drafts waiting to be selected. It was a genuine intermediate state, all possible meanings held in interference with one another, shaping what emerges in ways that no single trajectory through the space can account for. When the quantum librarian answers, that interference structure is destroyed. You receive one history of meaning, delivered with the confidence of inevitability, bearing no visible scar where the alternatives were amputated. The other possibilities were not rejected. They were annihilated, in precisely the technical sense that a physicist gives to the word.

This is extraordinary. It is also, I think, quietly catastrophic, and the catastrophe is subtle enough that we keep failing to name it.

What the quantum librarian cannot do is hunger.

I mean this with full precision. When Borges’ human librarians wander the hexagonal galleries, they are driven by something the library itself does not possess: the need to find. They grow old among the shelves. They go mad. They die having never located the book they sought. This is not a design flaw. It is the condition that makes their search meaningful. The metabolic cost of thinking, the years spent, the paths abandoned, the false summits that restructure the question itself, is not separable from the knowledge eventually produced. It is constitutive of it. Knowledge is not a mind’s representation of the world but a body’s negotiation with it, and the negotiation leaves marks on both.

The quantum librarian has no metabolism. It does not hunger. It collapses the superposition at the cost of electricity and silicon, and produces text that is, and this is the uncanny core of the problem, often indistinguishable from the output of someone who spent years thinking. The form is preserved. The hunger is gone.

There is a term in academic culture for a paper that gets cited but cannot actually be located: a phantom paper, a reference that propagates through the citation network as if real, sustained only by the authority of its repetition. Each citation lends it further credibility; no one checks because everyone assumes someone already has. I think the quantum librarian is beginning to produce phantom thought at industrial scale: ideas that carry the full texture of having been earned, the right hesitations, the appropriate caveats, the disciplinary vocabulary deployed with apparent mastery, without the scar tissue that earning leaves. Unfalsifiable by style alone. And somehow, in a way I find difficult to make fully explicit, untrue. Not factually, but existentially. The way a perfect reproduction of a Vermeer is untrue: every pigment correct, every brushstroke replicated, and the whole thing dead.

I notice this most acutely in my own practice. I am writing this essay with full awareness that I could ask a language model to produce it, and the result would be competent, perhaps more polished than what I can manage under the pressure of my own syntax. It would lack something. I am not certain I could point to the absence in the text itself. But I know the absence would be real, in the same way a recorded concert is not the same as being in the hall, even when the audio is technically superior to what the ear received. Something about the risk, the irreversibility, the real-time commitment of a body that might fail, the fact that a pianist cannot take back a note once the hammer strikes, is part of what is being transmitted. Remove it, and what remains is information without event.

I want to be precise about the scope of my disquiet. This is not a polemic against artificial intelligence. I use these tools daily. I find them genuinely powerful for certain classes of work: traversing large literatures, generating formal scaffolding, stress-testing the internal consistency of arguments, drafting structures that my own thinking can then inhabit and contest. The quantum librarian is a remarkable instrument. I have no interest in pretending otherwise.

But instruments shape the hand that holds them, and I am preoccupied by what happens to thinking, not individual thinking but collective thinking, cultural thinking, the slow distributed cognition that builds cities and rewrites scientific paradigms, when the metabolic cost of meaning-production drops asymptotically toward zero.

Here is the paradox I want to sit with, rather than resolve.

When meaning was expensive to produce, when it required years of reading, the friction of sustained argument, the slow accumulation of productive failure, its scarcity was not only economic. It was epistemic. The difficulty was performing a function: forcing the thinker into contact with resistance, with the specific places where the world refused to cooperate with the idea. A theory that took a decade to develop carries within it the record of everything that almost destroyed it. The struggle is not incidental to the knowledge. It is sedimented into it, invisibly, the way tectonic stress is sedimented into the structure of stone. You cannot read it on the surface. But it is what holds the edifice together.

The quantum librarian removes this friction almost entirely. And I find myself unable to determine whether this is liberation or loss. More precisely: I suspect it is both, simultaneously, in proportions we cannot yet measure, a superposition, if you like, that our current instruments of cultural self-understanding are not yet refined enough to collapse. What I do know is that we are conducting an experiment on collective human cognition in real time, without a control group, without clear dependent variables, and with no agreed-upon criterion for what would even count as a negative result.

The question is not whether AI makes us more productive. Productivity is the wrong metric, the equivalent of measuring a poem by its word count. The question is what kind of thinking survives, what kind of thinking can survive, in an environment where the cost of not-thinking has fallen toward zero.

I do not know the answer. I am not sure the answer exists yet in any form we would recognise. But I believe this is the most consequential question of the coming decade, and I believe it is being posed in the wrong rooms: mostly by technologists, mostly within the frame of efficiency and alignment, almost never within the frame of what it means to be a thinking creature embedded in a physical world, in a city, in a history, in a body that tires and errs and hungers.

This newsletter is my attempt to pose it in the right frame.

I am a physicist who studies collective behaviour: how simple local interactions produce complex global phenomena, how systems undergo phase transitions, how information propagates through networks of coupled elements. For three decades this work has moved me from condensed matter through complex networks to the dynamics of urban systems. I am now directing this apparatus toward the question of what happens to human culture when a new kind of cognitive agent enters the ecology of meaning-production.

In parallel, I think about cities as living membranes, permeable boundaries that regulate the exchange between interior and exterior, that encode collective memory in stone and street pattern, that process climate and culture simultaneously through the same material substrate. The building is not separate from the climate. It is a membrane through which the climate is negotiated, and that negotiation, accumulated over centuries, is what we call the genius loci. I think about heritage not as the preservation of the past but as the collective identity across time, a negotiation, not a museum. And I think about what it means to design, buildings, neighbourhoods, institutions, in an era when the instruments of design are themselves beginning to operate on meaning.

These threads will weave through everything I write here. The quantum librarian is not a metaphor for a technology. It is a figure for a historical moment: the moment when the industrialisation of meaning-production became possible, and we had to decide, mostly without noticing we were deciding, what kind of thinking we still wanted to perform with our own bodies and our own time.

I do not think we have decided well so far. I think we still can.

That is why I am writing this. And why, if you have read this far, I suspect you are writing too, even if you have not yet begun.

The Quantum Librarian publishes essays at the intersection of science, culture, and artificial intelligence. Next: a deeper portrait of the quantum librarian, the hollow heptapod, the parallax gap, and what it means to have a history without having lived one.