Standardization was the property of print that most directly enabled the collaborative enterprise of science. Darwin could not have developed the theory of natural selection without standardized taxonomic illustrations that allowed comparison of specimens across continents. Tycho Brahe's star catalogs, printed in identical editions and distributed across Europe, enabled the systematic comparison that produced Kepler's laws. Vesalius's anatomical atlas transformed medical teaching from an art dependent on local skill into a discipline organized around shared visual standards. In each case, the uniformity of the medium was the precondition for the diversity of the inquiry.
The AI transition produces its own version of the paradox, in a form Eisenstein's framework illuminates with uncomfortable clarity. Large language models converge toward common patterns — idiomatic structures, standard libraries, conventional architectures — because training data reflects the accumulated conventions of millions of developers. Code generated by AI tends to look similar regardless of who requested it, in the same way that printed books looked similar regardless of which printer produced them. But this standardization of the medium accompanies a dramatic diversification of the content, because the lower cost of building enables a wider range of people to produce a wider range of applications.
The tension between homogenized medium and diversified uses is where the most consequential effects of the AI transition will likely be found. The standardization enables comparison, facilitates exchange, and reduces friction between practitioners — all genuine benefits. But it does not produce innovation. Innovation comes from the individual mind confronted with a problem the standard cannot solve, and a generation of builders trained primarily on standardized AI output may find itself with reduced capacity to produce the unconventional implementations that advance the frontier of a discipline.
Segal's concern about the 'aesthetics of the smooth' — the polished, seamless quality of AI-generated output that conceals the absence of the maker's individual judgment — is a concern about this standardizing tendency. The seam where the maker's hand was visible gets smoothed away, and what remains is clean, functional, and anonymous. Eisenstein documented exactly this tension in the print era: printed books were more legible and accessible than manuscripts, but they were also less individual, less expressive of the maker's specific presence in the artifact.
Eisenstein introduced standardization as the third of her three structural mechanisms in The Printing Press as an Agent of Change, treating it as the property that most directly enabled the emergence of modern science. She was particularly interested in the reproduction of maps, diagrams, mathematical tables, and botanical illustrations — visual and quantitative information whose accurate reproduction had been nearly impossible in the manuscript era.
The paradoxical formulation — uniformity of medium producing diversity of content — was not original to Eisenstein but was most rigorously developed in her work. She drew on earlier observations by Marshall McLuhan and Harold Innis but insisted on grounding the claim in specific empirical cases rather than treating it as a general property of communication media.
Uniform medium, diverse content. Print standardized books but enabled the explosion of genres — pamphlets, vernacular works, experimental treatises — that scribal culture had suppressed.
Enabled collaborative science. Shared visual and quantitative references made systematic comparison possible for the first time, creating the conditions for empirical method.
Standardization does not equal innovation. The standard reduces friction and enables comparison but produces the mean of existing practice, not the frontier.
AI convergence as a new standardization. Code generated by AI converges toward idiomatic patterns in the same way printed books converged toward printers' conventions.
The seam disappears. Standardized output loses the marks of the maker's individual judgment — a loss that is simultaneously a gain in legibility and a cost in expressiveness.