In the Novum Organum (1620), Francis Bacon cataloged the forms of systematic self-deception under the heading of idols: the Idols of the Tribe (errors inherent in human cognition), the Idols of the Cave (errors arising from individual temperament and experience), the Idols of the Marketplace (errors arising from the imprecision of language), and the Idols of the Theatre (errors arising from received philosophical systems). Bacon intended the catalog as a practical instrument, not a curiosity: a mind aware of its tendencies toward distortion was better equipped to resist them than one operating in ignorance of its own biases. Ann Blair's historical methodology treats Bacon's project as an early attempt at what every subsequent information crisis has required — a systematic catalog of the specific ways content produced in a given medium can mislead.
Bacon's framework is an instrument for what Blair calls evaluative literacy: the capacity to recognize how a medium can fail, tailored to the medium's distinctive features. Manuscript culture generated certain failure modes (scribal error, selective transmission, authenticity forgery); print culture generated others (typographic error, commercial distortion, pseudonymous authority); AI generates still others — failure modes that Bacon did not anticipate but that his methodology suggests must be cataloged if the medium is to be used critically.
An Idol of the Machine, extending Bacon's taxonomy to the AI era, would name the conflation of fluency with truth that AI-generated content produces at scale. The idol works through a mechanism as old as rhetoric: the persuasive power of well-expressed ideas. But AI intensifies the susceptibility by producing fluency at industrial scale, without the effortful human processes that have historically correlated fluency with understanding.
Other AI-specific idols would include: the illusion of completeness (the AI's smooth surface conceals omissions as effectively as errors); the illusion of authority (statistical averaging of training data produces confident-sounding claims that have no verified basis); the illusion of novelty (recombinations of training-data patterns can appear original to readers unfamiliar with the source material); and the illusion of understanding (the reader's comprehension of the AI's output is mistaken for the AI's comprehension of the subject matter).
Blair's framework suggests that the AI era needs its own Bacon — not as a philosophical system but as a practical diagnostic instrument. The identification of specific, systematic, medium-specific failure modes is the first step toward the development of the evaluative practices that resist them. Without the catalog, individual readers must rediscover each failure mode through personal experience; with it, evaluative discipline can be taught, institutionalized, and collectively improved.
Bacon published the Novum Organum in 1620 as the second part of his projected Instauratio Magna. The work was a foundational text of the scientific revolution and has continued to serve as a general instrument for thinking about systematic cognitive error. Its application to AI-specific failure modes is an extension of Bacon's methodology rather than of his specific catalog.
Error is systematic, not accidental. Bacon's insight is that certain errors are produced by the structure of cognition and media, not by individual failings; they can be cataloged and collectively resisted.
Media-specific idols. Each new medium generates its own failure modes; the catalog must be updated with each technological transition.
Practical, not philosophical. The purpose of the catalog is operational — to make identifiable failure modes resistible in practice.
The Idol of the Machine. AI's specific contribution to the catalog is the production of fluency without understanding, at industrial scale.
Precondition for pedagogy. Evaluative discipline can be taught only when failure modes are named; the unnamed failure must be individually rediscovered.