Deep analysis of [[Robin McKenna]]'s "Sophistry on Steroids" paper via the [[Epistemic Defectiveness Model|EDM]] framework from [[Bonard et al.]]. ## Context Linked to morning review session and worked through a ChatGPT summary of McKenna's paper, developing critique angles. ## Insight McKenna thinks the problem with AI persuasion is **structural** (few actors, scale, marketplace domination) rather than **epistemic** (something wrong with the arguments themselves in the "best case"). But the structural problem *generates* an epistemic problem: scale + thin rationality = systematic epistemic defectiveness. The "best case" isn't stable because: 1. The thin conception permits omission-based misleading 2. Scale amplifies selective reason-giving into systematic distortion 3. The result satisfies EDM criteria for propaganda The political/structural concern and the epistemic concern are the same concern viewed from different angles. McKenna's separation of them is artificial. ## Mitchell-Douglas "Thin" Conception McKenna adopts [[Thomas Mitchell|Mitchell]] and [[Thomas Douglas|Douglas]]'s definition of [[rational persuasion]] that only requires: - Providing reasons (facts counting for/against an attitude) - Target adopts attitude based on recognizing those reasons What it **doesn't** require: - Engaging counter-considerations - Representing the reason-space fairly - Acknowledging competing considerations This permits the **omission route to propaganda**: genuine, true arguments can be epistemically defective through systematic omission. The target is led to believe reasons are decisive when they aren't. ## Tensions in McKenna 1. **Epistemic vigilance vs. flooding** — If humans are vigilant (per [[Hugo Mercier]]), why does quantity matter? If quantity overwhelms vigilance, it's not the reassurance McKenna needs. 2. **Sophistry as political vs. epistemic** — McKenna frames sophistry as a marketplace-domination problem. But sophistry is traditionally an epistemic concern. The EDM captures why. 3. **Intentionality workaround** — McKenna attributes intentions to purveyors, not AI. But purveyors may not intend specific epistemic defects of particular outputs. Responsibility gap. ## Paper Structure Options **Option A: EDM Critique of the "Best Case"** Present McKenna's framing → show thin conception permits omission-based defectiveness → argue EDM criteria satisfied even in "best case" → conclude rational/propaganda distinction is unstable **Option B: Epistemic Vigilance Meets Scale** Grant vigilance point → argue AI-scale bypasses vigilance → connect to EDM: mechanisms that prevent propaganda break down at scale → structural and epistemic concerns are inseparable **Option C: The Responsibility Gap** Focus on intention attribution → show it creates responsibility gap → conclude we can't rely on "good intentions" to ensure best case **Option D: Marketplace is Epistemically Loaded** Challenge marketplace metaphor → contrast thin (compete, circulate) vs. thick (deliberate, engage alternatives) → AI suits thin model, poorly suits deliberation → propaganda thrives in marketplaces ## Other Notes - The [[Hui et al.]] finding that AI content appears "more evidence-based and well-reasoned" is interesting: the appearance of rationality may be part of the problem. - Need to decide whether to defend EDM or just apply it (anticipate pushback that "engaging alternatives" is too demanding). See also: [[McKenna Propaganda Paper Plan]], [[McKenna Propaganda Paper - Planning Scraps]]