Essay — April 2026
Something that keeps surfacing: what Walter Gropius understood about industrial machinery, we might now need to understand about artificial intelligence. Both forces promised to liberate human creativity. Both instead revealed how urgently we need to teach humans to think before they make. At least, that’s the hypothesis.
There’s a photograph taken at the Bauhaus in 1926 of students in the preliminary course — the Vorkurs — working with paper and wire, their hands dirty, their faces concentrated. They’re not drawing what they want to make. They’re discovering what materials want to become. This might be a pedagogical philosophy as much as it is a classroom exercise, and it seems to contain an insight so simple it keeps getting lost: that the capacity to create something meaningful can’t be downloaded or inherited. It has to be built, through encounter with the world, through failure, through the specific friction of material resistance.
We might be at a moment when that insight matters more urgently than at any point since Weimar Germany in 1919. Not because the crisis is the same — it isn’t — but because its structure might be identical. A new technology has arrived that can produce, at speed and scale, outputs indistinguishable from those made by trained human hands. Then, the machine was the loom, the press, the factory line. Now, it’s the large language model. And the question that Gropius asked then might be the question we need to ask now: when a machine can generate the surface, what is the human actually for?
“The art schools must return to the workshop. This world of mere drawing and painting must at long last become a world that builds.” — Walter Gropius, Bauhaus Manifesto, 1919
Gropius’s answer was that the human brings judgment — the capacity to know not just how to make but why something should be made, what it should resist, what it should never become. That judgment can’t be mechanized because it’s not a procedure. It’s a world model: a structured internal understanding of how things relate, what matters, and what doesn’t, built through years of sustained encounter with real problems in the real world.
I. A connection to LeCun worth tracing
This seems to be, precisely, what Yann LeCun means when he argues that the current generation of artificial intelligence is working at the wrong level. Large language models predict what comes next at the surface — the next token, the next pixel, the statistically plausible continuation. They might be, in LeCun’s framing, very expensive machines for predicting what something looks like rather than understanding what it means. The alternative he proposes — Joint Embedding Predictive Architecture, JEPA — works differently: it builds abstract representations of the world in latent space, throws away unpredictable surface detail, and operates at the level of structure. It learns what things are before it generates what they look like.
The Bauhaus might have been building the same thing in humans, a century before anyone had words for it. And the school we need now might be the one that continues that project — not as nostalgia, but as a rigorous response to a specific contemporary crisis.
II. What the Bauhaus and Black Mountain understood
The Bauhaus opened in Weimar in 1919, in the ruins of a war and the birth of a republic. Gropius had a specific diagnosis of what had gone wrong with art education, and it seems structural, not aesthetic. The academies were teaching students to draw from models rather than to understand materials. They were optimizing for output quality rather than for the internal capacity that makes output meaningful.
When the Nazis closed the Bauhaus in 1933, many of its teachers fled to America. Josef Albers went to Black Mountain College in North Carolina, where he continued the experiment under radically different conditions — no institutional backing, no defined curriculum, no separation between the act of learning and the act of living.
What Black Mountain added to the Bauhaus model was something equally important: the insistence that the experiment and the education were the same thing. “Our central and consistent effort is to teach method, not content, to emphasise process, not results.” This seems like a pedagogy of world-model building, not knowledge transfer. It might be, structurally, what LeCun is arguing AI systems need to do.
III. The contemporary crisis
The Bauhaus was responding to industrialization. We might be responding to something more intimate. The machine of 1919 sat in the factory; it replaced the craftsman’s hands but left his mind intact. The machine of 2026 sits between the mind and its expression. It doesn’t just produce the object — it produces the thought about the object, the sentence describing the thought, the image illustrating the sentence. Every layer of the creative process can now be externalized before it has been internalized.
This might be the crisis that current AI discourse almost entirely misses. The conversation is dominated by questions of quality — is AI art good enough? Is AI writing indistinguishable from human writing? These are surface questions. The structural question seems different: what happens to the human capacity to create when the struggle of creation — the specific productive difficulty of not yet knowing how to make the thing you’re trying to make — is systematically removed from the process?
IV. What the school would test
The school this moment might require is not a critique of AI. It might be a response to what AI reveals about the preconditions for creative intelligence. A New Land School — a place to build something that doesn’t yet have a name, in conditions that make comfortable defaults impossible.
The thesis under test: as large language models make generation effortless, they accelerate the atrophy of the representational capacity that makes generation meaningful. The solution isn’t better tools. It might be a new practice for building internal world models before, and independent of, AI engagement. This practice can be taught. People who undergo it produce qualitatively different work. And if that capacity lives inside people rather than platforms, it can’t be bought, buried, or regulated away.
That thesis is falsifiable. The school would be the test. The work it produces would be the argument.
Sources: Gropius (1919), LeCun (2022, 2026), Johnson et al. (2025), White-Hancock et al. (2022), Katz (2002), Erickson (2015), Albers, Marcus (2019), Sennett (2008).
April 15, 2026