# Dialectical saturation thesis Because mainstream LLMs are trained on corpora saturated with philosophical argumentation, they internalise (i) a repertoire of philosophy-specific move types and move sequences, and (ii) a rough model of the field's success conditions. As a result, they can often generate dialectically appropriate continuations—sometimes non-obvious ones—when the conversational context specifies which philosophical "game" is being played. ## The Claim The object learned is not "facts about philosophers" but **how philosophical text works**: - Move types: distinction, counterexample, repair, disambiguation, synthesis - Move sequences: distinction → objection → reply; counterexample → repair; diagnosis → split-thesis - Success conditions: precision, explanatory power, simplicity/unification ## Three Versions (Increasing Ambition) ### Version 1: Script Competence Thesis (most defensible) LLMs have internalised common analytic scripts—recurring move-sequences—and can execute them fluently when cued. *Prediction:* Good at "philosophy-coded" moves (charitable reconstruction, pressure-point identification) even when the topic is unfamiliar. ### Version 2: Latent-Game Inference Thesis The biggest bottleneck is not the model's lack of "philosophical rules" but its uncertainty about which game is being played. Once you disambiguate the latent task, the model reliably lands in the right region of dialectical space. *This explains:* Why [[The obvious move prompting technique]] works—it pins down the game. ### Version 3: Salience-Not-Frequency Thesis (ambitious but disciplined) The model's "obvious move" is not necessarily the most *frequent* move; it can be a rare but structurally apt continuation, provided the local tension is specified sharply. *Key refinement:* "Salient" ≠ "common." See [[Move 37 and philosophical novelty]]. ## What's Doing the Work 1. The relevant competence is **local, dialectical continuation under constraint** 2. The scoring norms are not arbitrary—[[Timothy Williamson|Williamson]] supplies them: abduction rewards precise theories; vagueness is anti-explanatory 3. Philosophy's evaluative norms are largely tacit and learnable as practice-patterns (see [[Philosophy's implicit value signals]]) Source: [[Web Clippings/chat with ChatGPT about Floridi, LLM reasoning, Move 37|ChatGPT conversation on Floridi et al.]]