More than half a century later, Tom posted a set of images generated using AI. They showed traditional pottery—plates and vases in a familiar blue and white—decorated not with pastoral scenes or ornamental flourishes, but with collapsing buildings, explosions, and fragments of apocalyptic spectacle. The forms were precise, the decoration convincing, and the contradiction difficult to ignore. The quiet discipline of the surface was intact; what it contained was not. It is not difficult to see why such images capture our attention. The juxtaposition does much of the work: the ordered surface of a long-established craft set against scenes of disorder and collapse. But what they seem to register, perhaps unintentionally, is not simply novelty. They carry with them a more general unease—not only about what is shown, but about how it has been produced.
Conversations about AI increasingly gather this unease into more explicit terms. The concerns vary in emphasis—environmental cost, labour practices, data extraction, concentration of ownership, military alignment—but they are often presented together, as though they converge on a single point. The technology becomes the site at which these different anxieties meet, and where they are most sharply expressed. This has become familiar enough to pass without much notice. A student circulates a letter calling for a ban; elsewhere, in a more developed critique, Tom traces connections between datasets, power, and longer histories of exploitation. The positions differ in tone and scope, but they share a sense that something decisive is at stake, and that the response should match that urgency.
There is, in many of these concerns, a good deal that is difficult to dismiss. Questions of labour, ownership, and environmental cost are not abstract, and they do not originate with this technology. Nor are they confined to it. The conditions under which AI systems are built—outsourced work, uneven access, the aggregation of resources—are continuous with those that support other forms of production. They are more widely distributed and more deeply embedded than any single development might suggest.
What seems particular to AI is not the presence of these conditions, but the way they are encountered. They appear in a more concentrated form. The scale is larger, the speed greater, the scope less easily bounded, and the potential consequences more difficult to anticipate. Processes that elsewhere remain dispersed are drawn together, and in being drawn together, are easier to see. This concentration affects how the technology is understood. It is often treated as a kind of threshold: the point at which existing tendencies become unmistakable and, in that sense, the intensity of the response is not surprising. What might elsewhere be tolerated or overlooked becomes, here, harder to ignore. At the same time, this focus can narrow the field. When multiple concerns are gathered and directed toward a single object, there is a tendency to treat that object as their source. The distinction between the properties of the technology and the conditions within which it operates becomes less clear. What is, in part, inherited may appear as though it has been introduced. This is not to say that the concerns are misplaced. Nor is it to suggest that the actions of those who develop and deploy these systems are irrelevant. It is, rather, to notice that the frame within which the conversation takes place has a particular shape. It brings certain features into sharp relief while others remain on a more distant horizon.
Calls for caution and regulation follow quite naturally from this. If the scale and speed of development appear to outstrip existing forms of oversight, then restraint seems appropriate. Yet those calls sit uneasily alongside the conditions in which such technologies emerge—conditions that favour rapid development, competitive advantage, and the accumulation of capability. The tension is familiar, but here it comes more clearly into view. Against this background, the Delft images return in a somewhat different light. What they show is not only a technical capacity—the ability to combine styles or generate images—but a way in which contrasting elements can be brought into alignment on a single surface. The tradition is intact; the disruption is contained within it. The result is coherent, even as it depicts incoherence.
There seems something similar in the way AI is currently being discussed. A wide range of concerns—some technical, some economic, some cultural—are brought together and treated as though they belong to a single domain. This has the advantage of clarity. It sharpens the sense that something is happening. But it may also risk treating the ground on which these concerns appear as the source of those concerns.
The question, then, is not whether the concerns are justified, but how they are located. To focus on the technology alone is to see one part of the picture with unusual clarity. To widen the frame too quickly is to lose that clarity in generality. The difficulty lies in holding both in view: the specificity of the instance, and the broader conditions it reflects.
None of this resolves the question of how to respond. Positions that advocate refusal and those that advocate engagement are both shaped by the same conditions they address. Each has its rationale, and each its limitations. The more immediate task may be a quieter one: to attend more closely to the way the problem is being formed, and to how that formation directs what can be seen and what cannot.
What remains is a need for artists not only at the barricades or the bargaining table, but in shaping visions that do not shy away from the attractions and apprehensions of an unwritten future—apprehensions that are not always proportionate, but which nonetheless give shape to caution.
