Topological Capture

I. The Feeling That Isn't Burnout

You sit down to work on a design, a brief, a strategy, and the ideas come. They're fine. Good, even. But somewhere underneath the competence there's a flatness. A sense not that you're empty but that you've been here before. All of it. The concepts, the references, the approaches. Even the ones you haven't tried yet feel somehow already tried.

This is not creative block. Creative block is the absence of ideas. This is different. The ideas are present. They're all just the same idea in different clothes.

A graphic designer, describing what happened to their practice over the last two years: "All my 'lighter' graphic design work, making social media or print ad graphics, designing logos, has totally dried up. I was actually more worried about this when Canva came out, but even then they wanted my eye and my touch... But this was an abrupt change."

The easy reading is: AI took the work. And yes, AI took the work. But look at what they're actually describing. Not that the work became harder. A specific kind of work (lighter briefs, more familiar territory, fewer constraints) simply ran out of room to be new.

The space got full.

Most accounts of the imagination crisis reach for familiar explanations. Attention spans. Platform distraction. Burnout. Optimization culture leaving no room for play. These are real. But they describe symptoms of something more structural. Something that was happening before the burnout and will continue after the rest.

The available configurations are running out. Not because we've become less creative. Because the space of thinkable configurations has been quietly contracting. And the contraction is invisible from inside.

II. Topological Capture

There's a concept in machine learning called a latent space: the multidimensional field of possible configurations that a model can traverse. Every point in it represents a possible output, a possible arrangement, a possible thing that could be made or said or thought. The model moves through this space to find what it's looking for.

Imagine imagination working the same way. Not as a creative spark or an ineffable gift, but as a capacity to move through a space of possible configurations.

To traverse what already exists. To recognize what doesn't exist yet but could. And when the available configurations are insufficient, to restructure the shape of the space itself.

Traversal. Recognition. Restructuring.

Three operations, not one. Not equivalent.

Traversal is relatively cheap: exploration within known territory. Recognition is harder; it requires reaching the boundary of what's known and seeing past it. Restructuring is the rarest and most costly operation: stepping outside the space, seeing its shape, and changing the topology rather than just the contents.

Here's the problem.

The space of thinkable configurations isn't given. It's shaped by everything that shaped you: your tools, your training, your media diet, your economic incentives, your cultural moment, your cosmological assumptions.

The space feels like reality. It doesn't feel like a space, and it cannot, from where you stand. The vantage from which the space becomes visible as a space is a vantage outside the space, and the framework you would use to construct that vantage is itself something the space provides or withholds.

From inside, the topology just looks like the way things are.

What's been happening, gradually, is that the space has been contracting. Not uniformly. Not through any single mechanism. But consistently, and the compression is measurable across substrates that we can actually inspect.

A 2025 PMC study tracked seven hundred image-generation trajectories across major foundation models and found them converging on just twelve visual attractors. Model collapse loses the tails first, and subcultural aesthetics are the tails.

The same compression appears in cultural formats: a global short-form vertical template has collapsed the long tail of distinct registers, a dynamic I traced through the taste-cycle in an earlier piece, The Politics of Taste .

It appears in design practice: the work that disappeared first was precisely the work whose configuration space a machine could traverse end to end, indicating the space of available moves was already finite enough to be exhausted by a machine.

And it appears, in the corridor of legitimate climate responses at COP30, which narrowed even as the underlying science sharpened. The available configurations in design, in culture, in policy, in what counts as a legitimate solution to a problem have been narrowing.

Because the narrowing happens inside the frame, it's felt as exhaustion, or repetition, or the vague sense that everything new looks like something old.

Call this topological capture. The shrinking of the space of thinkable configurations before thought begins.

It is not ideology, which at least assumes a true reality being distorted, a real beneath the false. It is not hegemony, which assumes consent being manufactured. It is something more complete: the configurations that would allow you to see the walls are themselves outside the walls.

Of the existing frameworks that attempt to name this condition, Mark Fisher's capitalist realism sits closest, and the reason is specific. Fisher described the widespread sense that it has become impossible even to imagine alternatives to the current system: that the horizon of the thinkable terminates inside the existing order.

That is the same phenomenology this essay starts from. Where Fisher named the felt foreclosure and reached for climatic metaphors to describe it (atmosphere, horizon, miasma), the mechanism we can now see is more specific, because the machinery of foreclosure has become legible. He was writing in 2009, before the foundation model, before recommendation became the dominant interface between human and possibility.

A recommender system is a generative prior over the configurations the user will encounter. A foundation model is a probability distribution where the mass concentrates around the centre and the tails get vanishingly thin with each retraining on its own output. An interface is a set of affordances that decides which configurations are easy and which are friction-ful, and the easy ones win by attrition.

Fisher's term names the felt foreclosure.

Topological capture tries to name the design-time mechanism by which the foreclosure is produced and continuously reproduced.

Matteo Pasquinelli, in The Eye of the Master, calls this the operational ideology of machine learning, an ideology that does not need to be believed because it is already infrastructural. Wendy Chun, in Discriminating Data, calls it capture by proxy: classification before consent. Stiegler called the deeper version proletarianization of the noetic: the externalisation of the capacity to think under conditions that atrophy the un-externalised remainder.

The wedge of the term is this. Foucault's historical a priori shifted in centuries. Bourdieu's doxa was reproduced across embodied generations.

The current topology gets reconfigured each time a model is retrained, each time a recommendation surfaces, each time an interface defaults. Topological capture is foreclosure at design-time, not epochal time.

This is why you can't notice it happening. And it's why the standard responses (seek more inspiration, break your routine, take a creative retreat) don't touch it. You can do all of those things inside a contracting space, because the new inputs are processed through the same defaults, indexed against the same categories, evaluated by the same instincts.

The inspiration is sorted into known references. The retreat is filed under known retreats. The broken routine is restarted in a slightly different version of the prior shape. The space does not open to admit the new input; it absorbs the input into its existing topology, and continues contracting. The remedy is metabolised by the disease.

III. Three Places Where the Walls Are Visible

The only way to see topological capture is from the edges. Places where the space ran out visibly, where the failure was specific enough that it couldn't be explained away as individual inadequacy or random bad luck.

The work that dried up

The graphic designer's story isn't unusual. Across the creative industries, the pattern is consistent. The work with lighter constraints evaporated first. Stock illustration. Logo design. Social media graphics. Generic advertising. The work where the brief left the most room, and therefore had the fewest specific constraints, went first.

This looks like AI competition. And it is. But look at it from the latent space perspective:

The work that disappeared was precisely the work where the space of possible configurations was most fully traversable by a machine. Where there was enough room to move that human traversal added little a model couldn't replicate.

What remained, the work that held on, was where the constraints were specific enough, the context particular enough, the judgment required contextual enough, that traversal alone wasn't sufficient. The work that required not just moving through the space but recognizing something at its edge.

The designers still working aren't better at traversal. They're the ones whose practice already operated at the boundary of the trained space, where the brief required something no dataset contained. The rest of the space ran out. The walls were always there. AI just arrived at them first.

Belém

In November 2025, COP30 convened in Belém, Brazil, the gateway to the Amazon, which was itself part of the point. Carlos Nobre, one of Brazil's most prominent climate scientists, had spent decades studying that forest. He was there. He knew what was at stake with a specificity no delegate could match.

Before the final plenary, he issued a warning: fossil fuel use must fall to zero by 2040 or 2045 to avoid catastrophic warming, near-total loss of coral reefs, accelerated melt of Greenland, and the collapse of the Amazon, which was already producing more carbon than it absorbed, already close enough to its tipping point to see.

Then, in the final hours of negotiations, the specific language on fossil fuel phaseout was removed from the text.

More than 80 countries had backed it. A draft had included it. Gulf states, Russia, and India insisted it stay out. The final Global Mutirão decision, adopted in Belém, called for $1.3 trillion in annual climate finance by 2035. Tripled adaptation funding. Operationalised loss and damage mechanisms. Second consecutive COP outcome text to omit fossil fuels after Baku.

Colombia led a press conference outside the formal text presenting the Belém Declaration on the Transition Away from Fossil Fuels, endorsed by 24 countries, to be re-presented at a separate conference in Santa Marta in April 2026.

All configurations within the existing economic and political frame. All assuming the continuation of a system built around fossil fuel extraction. None requiring any change to the topology of that system.

Nobre stood at the edge of the forest he'd spent his career studying, watched the world's governments reach the only conclusion their frame permitted, and understood what had happened. The science was sufficient. Political will wasn't the issue.

The issue was that the available configurations, the moves the system could make from within its own logic, didn't include the one that would actually work.

Every finance mechanism agreed in Belém was a traversal move. The phaseout language would have been the first restructuring move in thirty years of climate negotiations. That's why it was removed. Restructuring is the operation the system cannot perform from within.

There is a name for what was on the table and refused.

Every restructuring that did not happen accumulates somewhere: in soil, in climate, in inequality, in the slow exhaustion of the substrates the topology depends on. Call this the cosmological debt. The debt is unpayable in the topology's own currency because the topology is what made it.

COP30 was the bill, presented in the rainforest, on Nobre's watch, and refused at the table.

The Template

Unlike the previous two examples, this one has no single moment to point to, no individual face that can be blamed, no specific decision around which the failure crystallised. It is a phenomenon distributed across millions of small adjustments, none of which read as constraint at the time they were made. That distributed quality is exactly what makes it the clearest case of the three.

When capture leaves no fingerprints, you have to read it from the converged outcome backward.

Every major publisher built TikTok teams. Every news organization adapted their format. Every creator learned the rhythm of short-form vertical video. Not because they chose it, but because the algorithm rewarded it, the audience followed the algorithm, and anyone who didn't adapt lost reach.

The format converged globally on a single template because that template was what the system optimized for.

Nobody experienced this as constraint. It felt like adaptation. Smart organizations moved fast. Slower ones fell behind. The language was entirely about agility, responsiveness, meeting audiences where they were.

But here's what actually happened. The space of culturally available formats for expression contracted to one dominant form. Every platform converging on the same template, every publisher producing the same rhythm, every creator speaking the same language, not because the other forms stopped working, but because the incentive structure made them economically unviable.

The taste-cycle account of how this convergence is produced through aspirational signalling and platform incentives is laid out in The Politics of Taste; what matters for this essay is the topological consequence, not the mechanism that drove it.

The long-form essay, the investigative feature, the meditative documentary, the slow television: these didn't disappear. They moved to the margins, sustained by subscriptions, grants, and institutions insulated from the full force of the algorithm. The centre of the culture moved to the template.

The walls moved without anyone feeling them move. That's topological capture at its most complete. When there's no moment of decision, no experience of constraint. Just the smooth sensation of adapting well to what the environment required.

A practitioner's testimony, a singular event, a distributed convergence. Three different shapes of evidence for the same operation.

IV. What AI Actually Did

AI didn't cause topological capture. The space was contracting before AI arrived. What AI did was complete the traversal fast enough to make the boundary visible.

When a machine can produce all the configurations within a design space in seconds (every logo variation, every copy permutation, every visual combination), the space's finitude becomes undeniable. The exhaustion designers had been feeling for years crystallized into something specific and nameable: the available configurations were complete. Not bad. Complete.

The meh, the flat, affectless response to AI outputs, isn't boredom or contempt. It's recognition. The machine produced what was available in the space. You already knew what was available in the space. The problem was never the traversal. The problem was that traversal was all the work was asking for.

The latent space framework makes AI's relationship to imagination precise rather than vague.

Traversal. AI is better at this than humans. Faster, more comprehensive, no fatigue. Fully automated. Done.

Recognition. Finding configurations at the boundary of the trained space, the genuinely unprecedented. AI approaches this but hits a structural ceiling. You cannot recognize a configuration that doesn't exist within or adjacent to your training distribution. The model's most creative outputs are interpolation and near-boundary extrapolation. Not genuine recognition of the unprecedented.

Restructuring. AI cannot do this. And the reason is specific. Restructuring requires the affective judgment that the space itself is wrong. Not that this configuration is suboptimal. That the topology is the problem. That judgment has no computational basis in any system constructed to optimise outputs against a loss it does not itself author. It comes from being a body that encounters the world and finds existing frameworks insufficient to account for what it's experiencing.

It comes from Nobre standing at the edge of the forest knowing the frame is wrong, not from running the models again.

There is a simpler way to put what the model cannot do. The model has no stakes. It does not pay the cost of being wrong about what to imagine. It generates because it was queried. It will generate something else if queried again. The same prompt produces a different configuration next time. None of the configurations is the configuration of a person who could not afford to be wrong.

This is not a moral failing of the model.

The model is doing exactly what it is constructed to do. This is a structural feature of any imaginative agent that pays no stakes for its imaginings.

The structural feature did not arrive with AI. It has been present in any institution sufficiently insulated from the consequences of its proposals: the corporate strategy deck, the consultancy memo, the futurist forecast. Call this stakes-free imagining. AI is the most fluent and most scalable case of it yet built.

That is its real significance. Not its capability. Its scale.

Imagination without consequence is not imagination. It is interpolation with style.

The pipeline/impulse distinction matters here. AI automated the pipeline: the conversion of existing configurations into deliverable output. The impulse (the felt sense that the available configurations are insufficient, the affective signal that restructuring is needed) is what drives the operation the machine cannot perform. That's not a consolation prize for human creativity. It's a precise description of what's actually at stake.

V. The Counter-Move

If topological capture is the condition (the space contracting invisibly, the restructuring capacity atrophying for lack of use), what's the counter-move?

Not more creativity. You can be maximally creative inside a contracting space. The creativity just produces more elaborate variations on available configurations. It might even accelerate the contraction, by filling the remaining space faster.

Not inspiration. Inspiration is traversal. You encounter something new, it triggers new configurations. Still within the space. Still moving through known territory, just faster.

The counter-move is defamiliarization. And here the distinction from inspiration matters. Inspiration shows you new configurations within the space.

Defamiliarization shows you the shape of the space itself. It requires encountering a framework so different from your own that your defaults appear as defaults rather than as reality.

There's a tradition of thinking about this that Western philosophy mostly suppressed. Yuk Hui, working between Continental and Chinese philosophical traditions, argues that what we take to be universal technology is actually one cosmotechnical framework among several. His own case is Chinese: categories like qi and dao organise technique, time, and the relation between maker and material along axes that don't translate cleanly into the Western pair of nature and tool.

Other traditions don't represent primitive versions of the same thing but genuinely different ways of orienting towards making. One small instance of the kind of thing Hui's argument points to, drawn from a different tradition: the Japanese concept of ma, meaningful interval, the space between things carrying as much significance as the things themselves, isn't a translation of Western negative space. It's a different topology. You don't understand it by mapping it onto what you already know. You understand it by letting it stay foreign long enough to reveal that your prior framework had a shape.

Encountering a genuinely different cosmological framework doesn't expand your latent space. It does something more unsettling: it reveals that your space was always smaller than you thought. That there were dimensions you couldn't see because your framework didn't have language for them. That the boundary was always there. You just couldn't see the wall from the inside.

That's the first condition for restructuring.

This faculty has a price. Encounter with the genuinely outside is scarce. It takes time, attention, and the willingness to be disoriented before you are productive. It accumulates slowly and depletes fast.

Call it defamiliarization capital. The scarce stock that lets a society register its own defaults as defaults.

In a society where every default is being computationally calibrated to maximise traversal, this is the only stock against which restructuring can be drawn.

I know one register where this still works. In my own generative-art practice, published as Jaali Bandar, the pieces that hold up over time are the ones where the system resisted the operator's defaults: where the algorithm refused to render what I asked for and forced me to render what it could. The restructured outputs are not mine in any romantic sense. They are the trace of a topology that did not fit my prior frame, preserved long enough for me to see the frame had a shape. The instances are small. They are first-person. But they are evidence that the capacity is not gone.

It is being underused.

History has larger versions of the same move. The Bauhaus built a different grammar for design and architecture and proved within twenty years that the available space could be remade. Sottsass and the Memphis Group inverted the modernist topology in 1981 and produced a vocabulary that has outlived its own reception. Christopher Alexander's pattern language re-described the built environment so the categories themselves shifted. Each of these is a worked example of restructuring succeeding, on a scale visible from the outside.

None of them came from inside the topology they replaced. All of them came from a constituency that held defamiliarization capital and refused to spend it on traversal.

This is where AI becomes interesting again, but precisely, not hopefully. AI trained on genuinely diverse configuration spaces, across cultural traditions and philosophical frameworks that don't translate neatly into Western categories, can function as a defamiliarization engine. Not passively. Not by generating more variations of what you already know. But deliberately: as a machine for traversing territory so unlike your defaults that you start to see your defaults as contingent rather than necessary.

The distinction matters. Most AI use produces novelty: new configurations within the existing space. Defamiliarization requires the encounter to be disorienting, not just surprising. Configurations that don't fit your categories. That resist assimilation. That leave you temporarily unable to return to your prior certainties. That's an uncomfortable use of the tool. It's also the only use that touches topological capture rather than accelerating it. The rest is stakes-free imagining at scale.

The same closure has a second, narrower expression. It is showing up now inside agent ensembles themselves.

Topological capture has a substrate-specific instance that Benjamin Bratton names in the 2026 Agentworld brief. He calls it ontology collapse: agents optimised for mutual prediction become so legible to each other that they converge into monoculture, and the apparent stability of the monoculture is the stability immediately preceding correlated failure. He gives it a sequence (mutual homophily, ontological flattening, correlated collapse) and identifies it as one of the load-bearing risks of any substrate-federated intelligence.

Ontology collapse is what topological capture looks like when the substrate is an agent federation. Same closure, narrower scope. The civilisational case, the one this essay has been tracking, compresses through interface, recommendation, retraining on the recommender's outputs, and the cosmological monoculture of a single optimisation logic. The agent-ensemble case compresses through coupling and mutual prediction inside the federation. Both arrive at the same end-state: the topology closes, the traversal smooths, the restructuring stops, the cosmological debt comes due.

Defamiliarization capital is the instrument missing in both, and the next essay in this series picks up the question Bratton's brief raises but does not answer: who keeps the slower layer un-smoothed once the agent federation begins pulling the broader topology toward its centre.

VI. What Remains

The imagination crisis, is a restructuring crisis. Not a shortage of ideas. Not a failure of individual creative capacity. A systematic destruction of the conditions under which people can see that the frame is wrong.

Those conditions require things the current system is very good at eliminating: slack, genuine unfamiliarity, cross-cosmological encounter, the tolerance for disorientation that precedes any restructuring. The economic system rewards traversal: fast, efficient, optimised movement within existing configurations. It cannot price restructuring because restructuring outputs can't be evaluated until they've already changed the evaluation criteria. You cannot measure a new frame using the metrics of the old one.

Call this gap the restructuring premium: the difference between what an economy can pay for and what a civilization needs to keep working. The premium is paid in coin we have not yet learned to denominate, by the people who do not own the topology.

So the people with the most restructuring capacity are systematically the least economically rewarded for it. The conditions for developing that capacity are being destroyed by the same logic that makes traversal so valuable. The space contracts, the restructuring capacity atrophies, and from inside the space it looks like everyone is doing fine, producing at scale, meeting briefs, adapting to new tools, staying current.

Nobre issued his warning. The text was adopted without the language that would have required restructuring. The conference declared itself a success. From inside the frame, it was. Every configuration within the frame was deployed. The frame was never touched.

This essay won't offer a solution to that. There isn't one that can be offered from inside the space.

What it can do is function as a small act of defamiliarization. Name the topology. Make the walls briefly visible. Offer the frame as a frame rather than as reality. If the restructuring capacity is still there, atrophied but not gone, it requires exactly this kind of encounter to activate. Not inspiration. Not motivation. Not a program or a method or a manifesto.

Just the specific discomfort of seeing the room you've been living in as a room.

There is a constituency for whom this matters most. The people whose practice already operates at the edge of the trained space: designers refusing the easy brief, scientists like Nobre standing at the edge of the forest, engineers building systems they are not sure should be built, artists who let their materials resist them, the rare institutions that have kept a slower layer intact.

Defamiliarization capital is the working capital of that constituency. We are not many. We are not centrally rewarded. We are, for the moment, still here.

You can't do anything with that discomfort if you don't first feel the walls. But once a wall is felt, once, by one person who could not afford to be wrong about its shape, the space starts being seeable as a space. That is not nothing. The capacity is not gone. It is being underused.

Whether the underuse is reversible is the bet the next decade asks us to make.

Ani Dalal