# Philosophy as internal verification domain The key sidestep for [[Luciano Floridi|Floridi]] et al.'s critique of LLMs doing philosophy: unlike empirical domains, philosophy's verification is largely *internal*—validity, consistency, dialectical robustness—so the missing "world verification loop" matters less. ## The Argument Floridi et al.'s deepest complaint concerns verification: > "In statistical terms, LLMs perform prior predictive sampling but lack an external feedback loop for posterior evaluation." (Floridi et al. 2025, p. 7) And further: > "They can help generate hypotheses and support human reasoning, but their outputs must be critically examined because they cannot discern truth or verify explanations." (Abstract) But this critique has different force depending on the domain. In medicine, posterior evaluation ultimately requires contact with clinical facts. In philosophy, "posterior evaluation" often reduces to: *does the package of commitments hang together under pressure?* From the ChatGPT conversation analysing Floridi: > For many philosophical tasks, you don't need a microscope or a census. You need things like: Validity: do the premises entail the conclusion? Consistency: do you end up committed to both P and not-P? Scope discipline: are you illicitly generalising? Equivocation checks: are you using a term in two senses? Counterexample resilience: does the view survive plausible cases? Dialectical adequacy: does it address the best objection? ## Why This Matters This is the basis for the "philosophy is special" move in the "Generating Philosophy with LLMs" paper project. Even granting Floridi's diagnosis about truth/verification, LLMs can play a substantially more constructive role in philosophical generation than in domains where verification is irreducibly empirical. > The distance between "linguistic competence" and "dialectical competence" is smaller than the distance between "linguistic competence" and "clinical competence." ## Trainable Internal Checks The checks that matter in philosophy can be scored without world access: - "Spot the equivocation" - "Find the hidden assumption" - "Produce the strongest objection" - "Repair the argument while preserving the conclusion" - "Formalise then check validity" This suggests philosophy-specific training could internalise these norms. ## Nuance This is NOT cheap coherentism ("truth = coherence"). The weaker, more defensible claim: > Even if philosophical truth is not identical to coherence, a lot of philosophical error is detectable by internal checks, and the discipline devotes huge energy to those checks. Source: [[Web Clippings/chat with ChatGPT about Floridi, LLM reasoning, Move 37|ChatGPT conversation on Floridi et al.]]