Presentation slides for WIP seminar (2026-01-20) on [[Robin McKenna]]'s "Sophistry on Steroids" and the [[Epistemic Defectiveness Model]]. **Related:** [[McKenna Propaganda Paper Plan]], [[McKenna Propaganda Paper - Planning Scraps]] --- ## **AI PERSUASION AND PROPAGANDA: A Critique of McKenna** **Nick Young** WIP Seminar, 2026-01-20 --- ## **McKenna's Central Question** > "Should this concern us? I argue that, while some worries about persuasive AI may be overblown, we should be worried for a mix of ethical, epistemological and political reasons. Most centrally, we should be worried because persuasive AI may lead to a small number of powerful actors dominating what I call the 'marketplace of arguments'—the set of arguments that provide the materials we use to discuss important moral, political and societal issues." McKenna (2025, Abstract) --- ## **Not About Superhuman Persuasion** > "...the fundamental reason why we should be worried about AI persuasion is not that AI has, or soon will have, significantly more persuasive power than humans. The reason we should be worried is that any AI persuader or AI-generated persuasive content is (a) relatively easy to roll out en masse and at scale when (b) you have the resources (financial as well as technological) to create it in the first place." McKenna (2025, p. 2) **The worry: Scale + Concentration, not quality** --- ## **Sophistry on Steroids** > "The purveyors of persuasive AI technologies can generate huge numbers of 'arguers for hire' (AI agents) or huge amounts of 'persuasive content' within minutes, with any dilution in the quality of arguments often more than made up for by the massive increase in quantity." McKenna (2025, p. 9) Ancient sophists: arguers for hire AI sophistry: **armies** of arguers for hire --- ## **McKenna's "Best Case" - Description** What [[Robin McKenna|McKenna]] brackets: - Misinformation and deception - Manipulative tactics - Coercive influence What remains: - Reason-giving arguments - Aimed at [[rational persuasion]] - But deployed **at scale** by concentrated actors --- ## **McKenna's "Best Case" - His Words** > "I am interested instead in what you might call a 'best case'—or at least a 'better case'—scenario, in which the purveyors of persuasive AI technologies develop LLMs and chatbots that produce genuine arguments, evidence, and reasons for conclusions that the purveyors of these technologies themselves endorse, or at least would like their intended targets to endorse. That is, they produce genuine arguments with the attempt to rationally persuade, not simply deceive or mislead." McKenna (2025, p. 7) --- ## **What is Rational Persuasion?** McKenna adopts [[Thomas Mitchell|Mitchell]] & [[Thomas Douglas|Douglas]]'s definition: > "A rationally persuades B to adopt attitude α if: (i) A brings it about that B adopts α, (ii) A does so only by providing B with reasons for adopting α, (iii) B adopts α based on recognizing some of the reasons given by A, (iv) A intends each of (i)-(iii)." Mitchell & Douglas (2024), quoted in McKenna p. 7 **Key point:** Only requires providing reasons—NOT engaging counter-considerations --- ## **McKenna's Assumption** His "best case" is meant to be: ✓ Politically troubling (marketplace domination) ✗ NOT [[propaganda]] He explicitly sets propaganda aside as "overtly coercive or manipulative" influence **The question my paper asks:** Does this distinction hold? --- ## **But What IS Propaganda?** > "There is plenty of evidence for the current predominance of a negative sense of the word ['propaganda'] in English. 'Propaganda' can...refer to 'the systematic dissemination of information, especially in a biased or misleading way, in order to promote a political cause or point of view'. This final sense has a pejorative connotation..." [[Constant Bonard|Bonard]], [[Filippo Contesi|Contesi]], & [[Teresa Marques]] (2024, p. 1082) **Paradigm cases:** Nazi propaganda, Rwanda RTML radio, tobacco industry campaigns --- ## **Epistemic Defectiveness Model - Description** [[Bonard, Contesi and Marques]] (2024) refine [[Sheryl Tuttle Ross]]'s account: A communicative act is [[propaganda]] iff: 1. **[[Epistemic defectiveness]]** — message is defective (broadly) 2. **Political intent** — institutional provenance, persuasion for political ends 3. **Social targeting** — targets socially significant group 4. **Awareness** — agent aware or should be aware of defectiveness --- ## **Epistemic Defectiveness Model - Their Words** > "A communicative act C counts as propaganda if, and only if, > (1) C conveys an epistemically defective message...that is false, inappropriate, or connected to other beliefs in ways that are inapt, misleading, or unwarranted, given C's illocutionary and perlocutionary intents > (2) C is produced by, or on behalf of, a social or political institution, organization, or cause with the intention to persuade, for direct political ends > (3) a socially significant group of people > (4) while the producer/s of C either is/are aware of the epistemic defectiveness of the message or should have been aware of such defectiveness." Bonard, Contesi, & Marques (2024, p. 1092) --- ## **Epistemic Defectiveness ≠ Falsehood** > "[Cherry-picking true evidence about a battle] may not be epistemically defective insofar as, e.g. it reports true information. So, it does not fail Grice's Conversational Maxim of Quality. However, it does fail Grice's Maxim of Quantity by omitting relevant information about the heavy losses suffered in another battle. This is misleading. The audience will pragmatically infer that the news conveys all that is relevant about the war in question, which is false." Bonard, Contesi, & Marques (2024, p. 1088) **Key:** Defectiveness through omission, implicature, connections—not just falsity --- ## **The Dilemma for McKenna** **Horn 1 (Benign):** Fully idealized "best case" - Transparent, symmetric, flags limitations - Robust exposure to counterarguments - → No epistemic defectiveness - → Not propaganda - → **But also not troubling!** **Horn 2 (Troubling):** Marketplace domination - Small actors flood argumentative environment - Arguments favor their interests - → Troubling in McKenna's sense - → **But is it epistemically defective?** --- ## **Campaign-Level Defectiveness** **My key move:** Treat the communicative act at campaign/ecosystem level, not individual utterance Three routes to defectiveness: 1. **Omission-by-crowding-out** — counterarguments made inaccessible 2. **Quantity-style misleadingness** — saturation licenses false inferences 3. **Distorted salience** — prominence signals importance/decisiveness Individual arguments can be sound while **system** is defective --- ## **The Collapse** McKenna's middle category (troubling ≠ propaganda) is **unstable** **Either:** - Idealize to remove defectiveness → benign (not troubling) **Or:** - Preserve troubling features → defective → propaganda **No stable middle ground** The mechanism that makes it troubling IS the mechanism of epistemic defectiveness --- ## **But Wait: Can It Actually Happen?** Skeptical turn: Can marketplace actually be dominated? **Finite argument schemas** ([[Douglas Walton|Walton]], Reed, & Macagno 2008): - Immigration, climate, redistribution → recognizable types - Appeals to rights, consequences, fairness, precedent - Variations abound, but **underlying structures finite** **Implication:** Flooding produces diminishing returns --- ## **The Empirical Question** From my abstract: > "I do not want to overstate this point. The empirical questions are open. Even if argument schemas are finite, rhetorical and emotional framing may not be; flooding might succeed at the level of attention and affect even if it fails at the level of argumentative content." **Open question:** Does AI influence work at level of attention/affect rather than argument? --- ## **Upshot** **Main claim:** McKenna's distinction between troubling AI persuasion and propaganda **collapses** under Epistemic Defectiveness Model **But:** Unclear whether his specific scenario (marketplace domination via volume) is achievable **Task ahead:** If real AI influence risks lie elsewhere, identify where --- ## **Questions for Discussion** 1. Is campaign-level defectiveness the right unit of analysis? 2. How finite are argument schemas really? Does flooding fail argumentatively but succeed affectively? 3. Is the Epistemic Defectiveness Model too broad? (Does omission-based defectiveness demand too much of persuaders?) 4. Where DO the real risks of AI-enabled influence lie? ---