When the Analogies Mislead: Reading Generative AI Through Its Own Framework
The Comparison That Carries the Wrong Property
Scholars and policymakers reach for historical analogies when they describe generative AI: the printing press, steam power or electricity. Rida Qadri, Vinodkumar Prabhakaran and Remi Denton argue in a new paper that these comparisons are typically imprecise about which property of the technology carries the comparison - and that imprecise analogies produce imprecise governance. A comparison that names the wrong property produces governance aimed at the wrong part of the system. [1]
The paper is not another attempt to settle whether generative AI is good or bad. It is a diagnostic framework for analyzing how generative AI can transform epistemic and cultural practice, and it treats the comparisons we reach for as claims that can be checked rather than images that can be admired. [1]
Three Coordinates in Place of One Metaphor
The framework decomposes each intervention into three coordinates: the epistemic site at which a technology acts, the governing logic by which it organizes its object, and the technical mechanism through which the logic is instantiated. Splitting the three is what turns a comparison into a diagnosis. It separates structural cultural consequences, which follow from the mechanism itself, from contingent ones, which remain open to design and institutional choice. [1]
That distinction is the paper’s sharpest contribution: it draws a line between what a system does because of how it is built and what it does because someone chose it. Only the second category is available to policy. [1]
From Indexing to Inference
Applied to information discovery and knowledge synthesis, the framework traces two shifts: from indexicality to inference, and from editorial authority to statistical consensus. Each produces specific, traceable cultural effects - and, more usefully, reveals governance levers that gestalt analogy obscures.
The second shift is where the familiar worry about surface and substance finds a grounded form. When a statistical consensus takes over from an editorial authority, the question of who vouches for a claim stops having a masthead as its answer. Read through the framework, that is not a vague unease about machines bluffing - it is a named change at a named site, with a logic and a mechanism attached, which is exactly what makes it addressable. [1]
The framework does not argue that any of this is deception. It argues that the shift is traceable, and that tracing it is what makes governance possible where the analogy would have hidden it. [1]
The paper’s own closing move is modest and hard to argue with: applied this way, the cultural impact of generative AI stops being a debate about images and becomes a set of questions a policymaker can answer - which site, which logic, which mechanism, and therefore which consequence is structural and which is still a choice. [1]