In The Printing Press as an Agent of Change, Elizabeth Eisenstein notes:

“Until a half century after Copernicus’ death, no potentially revolutionary changes occurred in the data available to astronomers.” But Copernicus’ life (1473–1543) spanned the very decades when a great many changes, now barely visible to modern eyes, were transforming “the data available” to all book-readers. A closer study of these changes could help to explain why systems of charting the planets, mapping the earth, synchronizing chronologies, codifying laws and compiling bibliographies were all revolutionized before the end of the sixteenth century. In each instance, one notes, Hellenistic achievements were first reduplicated and then, in a remarkably short time, surpassed. In each instance, the new schemes once published remained available for correction, development, and refinement. Successive generations could build on the work left by sixteenth-century polymaths instead of trying to retrieve scattered fragments of it.…the great tomes, charts, and maps that are now seen as “milestones” might have proved insubstantial had not the preservative powers of print also been called into play. Typographical fixity is a basic prerequisite for the rapid advancement of learning.

Eisenstein, The Printing Press as an Agent of Change, page 113

I’ve been thinking about this with respect to what we’re seeing with so-called AI and large-language models. One of the ways an LLM operates is to ingest gobs of text and then reshuffle and remix that text into plausible-sounding examples of the same. This is the exact opposite of what Eisenstein is talking about: instead of persisting and faithfully reproducing text, LLMs modify and distort it, and then spit it back out at such volume that the original texts may become obscured or lost.

But if, as Eisenstein asserts, typographical fixity is required for the rapid advancement of learning, what happens when that fixity is eroded? What happens when a text can drift into nonsense faster than we can observe or record that drift? Perhaps a possible addendum to Eisenstein’s assertion is that typographical disfixity undermines learning entirely. That the slipperiness and superficiality of an LLM’s output, its ability to cosplay as something written with intention, paints all text with a kind of a priori suspicion. That when text is stripped of its preservative powers, it becomes impossible to engage with, to collectively develop. I hope not; but I’ll admit to being worried.