Here is a contradiction worth sitting with. Byung-Chul Han, reading Heidegger, insists that artificial intelligence does not think because it does not have a hand. And yet the same journal issue that hosts this claim is full of architects using generative AI to think—branching, displacing, semantizing, producing meaning at speeds no hand could survive.
Both cannot be simply true. Either the hand is doing something computation cannot, or the philosophers are guarding a shrine that no longer holds anything.
Let me concede the strongest version of the case against the hand first, because it is stronger than most craft defenders admit.
A human hand cannot out-produce a diffusion model. It will never match its variation-per-second. Per the case study in Sanchez Morales and Lopez Cervantes, feed a model a deformed carton and, through a sequential protocol, it can branch into typological, atmospheric, and affective semantization.
If you defend drawing on the grounds of output—more images, better images, faster images—you have already lost. If you defend it on beauty, you have lost, because beauty is a property of finished representations, and finished representations are exactly what the machine now floods.
So drop that fight. It is being scored on the wrong axis. The interesting question is not what the hand produces but what the hand knows in the act of moving.
Giancarlo Carnevale draws the distinction that reframes everything. There is representation—the description of what is already defined. And there is prefiguration—the drawing that accompanies thought in its formative stages.
Representation is what's undergoing radical reconfiguration right now. Cartography, surveying, restitution, rendering: all of it accelerating, all of it in the machine's domain.
Prefiguration is something else. It is imperfect, personal, and resistant to codification. Carnevale notes that Michelangelo's sketches still resonate, unexpectedly, with those of Aalto or Niemeyer; the reading offered here is that this is not because they share a style, but because they share a status: marks made before the object exists, in a zone that centuries have not standardized.
Carnevale calls this intimate drawing, as opposed to exhibited drawing, and the distinction is the whole game.
Exhibited drawing is output. It is meant to be seen, judged, communicated—and it is exactly what AI is transforming, because it lives in the world of already-defined representations.
Intimate drawing is prefiguration in its rawest form: an auroral field of morphemes, figurative obsessions, non-linear logics. Carnevale reaches for Matte Blanco's "symmetric thinking" to name it—a mode where the ordinary asymmetries of logic don't yet hold, where a thing can be its own opposite, where meaning has not yet resolved into proposition.
This is the zone. It is pre-representational. It is thought before it has language.
And here is the structural point that no rendering speed can touch: AI can only operate on representations that are already defined. It transforms, recombines, and proliferates the given. It cannot enter a zone whose defining property is that nothing in it is given yet.
The machine is fast precisely because everything it works on has already been resolved into data. Speed is possible only after definition. The pre-representational zone is slow, or rather it is prior to the axis on which speed and slowness are measured.
Carnevale is careful, and so am I: this is not analog versus digital. It is time versus speed. Slowness, friction, hesitation—the drag of a hand that does not yet know what it is looking for and traces marks anyway.
Now bring in Heidegger, because he supplies the mechanism Carnevale only gestures at.
In Being and Time, the hand is what discloses the world in its original form. Reach for a pen and it does not appear as an object with qualities—it is simply ready-to-hand, absorbed into use. To make it an object, you must withdraw your hand and stare.
The staring is representation. It comes second. "The less we just stare at the hammer-Thing, and the more we seize hold of it and use it," Heidegger writes, "the more unveiledly is it encountered as that which it is."
The grasping hand experiences the thing more primordially than the representing intuition. The hand anticipates every representation—it reaches into being before objectification blocks the way.
This is the whole argument in one line. AI has no hand. It never seizes hold of anything. It only ever stares—it operates exclusively in the register of the finished, objectified, computed representation, which for Heidegger is the derivative mode, the one that comes after the hand has already disclosed the world.
Han, channeling what he says Heidegger would say, draws the blade: "artificial intelligence does not think because it does not have a hand." Not as insult, but as ontology. Per Han's reading of Heidegger, the hand has access to the non-countable, the singular, the sphere that precedes all number—and, on this account, computation is nothing but number.
So the contradiction I opened with resolves. The architects branching through GenAI are doing real work, but it is work in the register of representation—the semantic laboratory, intensified and displaced meaning, the transformation of what is already defined. That is genuine, and it is not thinking in Heidegger's sense, because it never touches the ready-to-hand.
Which sets up the sharpest reversal, and it is not the one craft nostalgists want.
The hand's irreducibility was never really about making marks. In a world where images multiply exponentially, the scarce act is not production. It is the decision about which images deserve to exist.
Carnevale puts it precisely: authorship will reside less in production than in the capacity to select, order, and withhold.
Withhold. That is the word that matters. When a machine can generate a thousand plausible options in a second, the meaningful gesture is not adding the thousand-and-first. It is saying no to nine hundred and ninety-nine.
When a machine can generate a thousand plausible options in a second, the meaningful gesture is not adding the thousand-and-first. It is saying no to nine hundred and ninety-nine.
Selection, synthesis, authorial responsibility—the critical act of judgment that Carnevale says AI expands the field for but does not perform. The Wunderkammer fills itself now; what remains human is the curator's job.
And notice why this is human, on the terms we've built. Judgment about what should exist is not the manipulation of defined representations. It is the pre-representational act par excellence: it happens in the zone of indeterminacy, before the option is fixed, in the hesitation over whether it should be fixed at all.
The hand that withholds is the same hand that traces marks before knowing what it seeks. Both operate in the auroral field. Both are prior to definition. Both are accountable in a way no probabilistic image formation can be, because accountability requires a someone who could have chosen otherwise.
This is where I want to be exact, because the argument is easy to cheapen. Drawing-for-oneself reasserts itself as a critical anchor, in Carnevale's words, "not because it is purer, but because it remains irreducibly human."
Not purer. There is no moral high ground in the pencil. The hand is not virtuous; it is not even necessarily better.
It is irreducible for a structural reason: it has access to a zone that computation is locked out of by definition, not by current limitation. No faster model reaches it, because the barrier is not speed. The barrier is that the machine can only work on what has already been resolved into representation, and the zone is defined by the absence of exactly that.
The contradiction I opened with was not really a contradiction. Both things are true. AI thinks in the only sense it can—on the given, at speed. The hand thinks in the sense that precedes the given, slowly, accountably, deciding what deserves to become given at all.
The image glut is coming regardless. What it makes precious is not the capacity to draw one more thing. It is the older, quieter capacity to stand in the space before the image, and to refuse.