Typographical Fixity

In the AI Story

Eisenstein argued that fixity operated as a causal mechanism distinct from dissemination and standardization, producing consequences neither could explain alone. In the manuscript era, the fluidity of texts shaped the entire structure of intellectual life: knowledge was local, because versions in Paris might differ from versions in Bologna; knowledge was fragile, because a single fire could destroy the only copy of a work; knowledge was conservative, because only texts of proven value justified the expense of reproduction. The fluidity meant that scholars could not build systematically on previous work, because the previous work was not stable enough to serve as a foundation.

The AI transition presents a paradoxical relationship to fixity. AI-generated code is fixed in the sense that it runs consistently once deployed — every execution produces the same results on every machine. But the generation process itself is stochastic: the same prompt submitted twice may produce different working implementations. This is not analogous to scribal variation, where errors accumulated involuntarily. AI variation is structural, built into the generation process, and often invisible because each variant works correctly.

The deeper consequence concerns the builder's relationship to code. A developer who wrote a function by hand understands its logic in an embodied sense — every variable name reflects a decision, every structural choice a trade-off that was consciously evaluated. A developer who received a function from AI understands what it does but may not understand why, or what alternatives were considered, or what edge cases the particular implementation handles well or badly. Fixity of execution coexists with fluidity of comprehension — a configuration Eisenstein's framework did not anticipate.

The practical stakes are significant. Cumulative knowledge depends on the ability to evaluate, correct, and extend previous work. When the implementation is fixed but the reasoning behind it is opaque, the conditions for cumulative building are compromised in ways that print-era practitioners did not face. Segal's account of the engineer whose architectural confidence eroded after months of AI-assisted development — who was making decisions with less confidence and could not explain why — is a symptom of this new configuration.

Origin

Eisenstein developed the concept of typographical fixity through decades of work in the Newberry Library's rare-book collections and in archives across Europe, tracing specifically what changed when scholars could finally trust that their counterparts in distant cities held identical copies of the texts they discussed. The concept appeared in its mature form in The Printing Press as an Agent of Change (1979), where Eisenstein positioned it as the first and most consequential of the three structural properties of print.

The concept has been refined by subsequent scholars — notably Adrian Johns, whose The Nature of the Book (1998) contested Eisenstein's claim that fixity was automatic and argued instead that it was a social achievement requiring specific institutional practices. Johns's critique sharpens rather than undermines Eisenstein's analytical move: fixity as property versus fixity as achievement reveals that the same underlying technology can produce different outcomes depending on the institutional context of its use.

Key Ideas

Structural, not incremental. Fixity was not a quantitative improvement on scribal accuracy but a qualitative transformation in the relationship between text and reader.

Precondition for cumulative knowledge. Without stable texts, each generation cannot build on the previous one's work with confidence that the foundations will hold.

Enables citation and collatio. The practices of scholarly comparison that define modern science required shared references that only fixity could provide.

Inverted in the AI context. Code execution is fixed while generative variability and opaque provenance produce a new kind of fluidity that print-era frameworks did not anticipate.

A social achievement, not automatic. Post-Eisenstein scholarship has shown that fixity requires institutional practices to realize, suggesting the same will be true in the AI age.

Debates & Critiques

Adrian Johns's 1998 critique argued that Eisenstein overstated the automatic quality of fixity and understated the social and institutional labor required to produce identical copies in practice. The debate has refined rather than displaced Eisenstein's framework, and it has acquired fresh relevance in the AI era, where the gap between potential fixity and realized fixity — between code that could run consistently and code whose behavior is actually trustworthy — depends on institutional practices (review, testing, documentation) that do not automatically accompany the technology.