# Lipton Deployment Decisions — Generating Philosophy > [!info] Legacy > New decisions are created as individual tagged notes in `Notes/`. See [[Recent Captures.base]] or the Project Captures view on [[Sessions/Generating Philosophy]]. This note is preserved as historical archive. Working note capturing refined analysis of how [[Peter Lipton]]'s *Inference to the Best Explanation* (2004) should be deployed in the [[Sessions/Generating Philosophy|generating philosophy paper]]. This note records decisions reached by working through the source chapters on 16 Feb 2026, refining the initial survey of all thirteen chapter notes (made 15 Feb). The chapter-by-chapter analysis notes are at [[Lipton Ch01 Induction - Generating Philosophy Relevance|Ch01]], [[Lipton Ch02 Explanation - Generating Philosophy Relevance|Ch02]], [[Lipton Ch03 The Causal Model - Generating Philosophy Relevance|Ch03]], [[Lipton Ch04 IBE - Generating Philosophy Relevance|Ch04]], [[Lipton Ch05 Contrastive Inference - Generating Philosophy Relevance|Ch05]], [[Lipton Ch06 The Raven Paradox - Generating Philosophy Relevance|Ch06]], [[Lipton Ch07 Bayesian Abduction - Generating Philosophy Relevance|Ch07]], [[Lipton Ch08 Explanation as Guide - Generating Philosophy Relevance|Ch08]], [[Lipton Ch09 Loveliness and Truth - Generating Philosophy Relevance|Ch09]], [[Lipton Ch10 Prediction and Prejudice - Generating Philosophy Relevance|Ch10]], [[Lipton Ch11 Truth and Explanation - Generating Philosophy Relevance|Ch11]], [[Lipton Conclusion - Generating Philosophy Relevance|Conclusion]]. --- ## 1. The Realization Thesis (Ch. 7) — Repositioned **Initial placement:** Section 1, preempting the "just statistics" objection. **Revised placement:** Later in the paper (Section 3 or conclusion), as a *capstone* argument rather than a framing argument. **Reason for revision:** The squash analogy works for Lipton because both technique-thinking and mechanics are real descriptions of the same activity — both IBE and Bayesian reasoning are normatively respectable. But the "just statistics" critic isn't making the analogous move. They're claiming LLMs *aren't playing the game at all* — that statistical pattern matching isn't a normatively respectable epistemic practice. The realization thesis can't counter that claim until the paper has *first* established (via Sections 2 and 3) that LLM outputs satisfy the relevant quality constraints. Only then does the squash analogy land: given that the outputs are in the game, the fact that they were produced statistically is no more problematic than the ball obeying mechanics. **Open question:** Can the realization thesis still do lighter work in Section 1 as a *promissory note* — orienting the reader toward the question "can statistical processing realize evaluative sensitivity?" without answering it? This frames what the paper will argue without requiring the full apparatus. Risk: if Sections 1-2 are slow, the unresolved framing might feel like a dangling thread. Works best if it's brief and explicitly flagged as promissory. **Passage:** > If these suggestions are along the right lines, then arguing that Inference to the Best Explanation is wrong because Bayesianism is right is like arguing that thinking about technique cannot help my squash game because the motion of the ball is governed by the laws of mechanics. Even if Bayesianism gave the mechanics of belief revision, Inference to the Best Explanation might yet illuminate its psychology. (Ch. 7) --- ## 2. "Post Hoc Ergo Ad Hoc" (Ch. 10) — Deployable But Not Standalone **The passage and its structure.** Lipton considers three popular defences of prediction over accommodation and rejects all three before giving his own account. The "ad hoc" defence is the first. Lipton identifies a two-way ambiguity in "ad hoc": > Accommodating theories are obviously ad hoc in one sense, since 'ad hoc' can just mean purpose-built, and that is just what accommodating theories are. To claim, however, that a theory that is ad hoc in this sense is therefore poorly supported begs the question. Alternatively, 'ad hoc' can mean poorly supported, but this is no help, since the question is precisely why we should believe that accommodating theories are in this sense ad hoc. To assume that accommodating theories are ad hoc in the sense of poorly supported is to commit what might be called the '*post hoc ergo ad hoc*' fallacy. The simple appeal to the notion of an ad hoc theory names the problem but does not solve it. (Ch. 10) **The parallel to LLM criticism:** "LLM outputs are derivative" is ambiguous between (a) produced-from-training-data and (b) epistemically worthless. In sense (a), all LLM outputs are derivative — of course they come from training. But produced-from-training doesn't entail epistemically worthless. In sense (b), "derivative" means "worthless," but now you're assuming what you're supposed to prove. **The complication:** Lipton doesn't stop at diagnosing the fallacy — he goes on to offer the *fudging explanation*, which gives a positive reason why accommodation sometimes produces weaker results. His point isn't that the worry about accommodation is baseless; it's that the "ad hoc" label does no explanatory work — you need an actual mechanism (fudging) to explain when and why accommodation goes wrong. He then identifies conditions under which accommodation carries *no* disadvantage. **Consequence for the paper:** Deploying the "post hoc ergo ad hoc" label requires also offering the paper's own account of *when* the LLM's training origin does and doesn't matter. The fudging explanation provides the template: training matters when it produces outputs bearing marks of overfitting (arbitrary combinations, surface-level pattern matching without evaluative structure); training doesn't matter when outputs satisfy the same quality constraints as human-produced philosophy. This is where the paper's artefact-evaluation argument and the [[Notes/Paper-internal constraints as the locus of philosophical evaluation|paper-internal constraints]] thesis do the work. **Deployment:** Use the fallacy label economically in Section 1 (against [[Luciano Floridi|Floridi]]/[[Tom Zahavy|Zahavy]]). Deploy the fuller fudging framework in Section 3 as part of the positive case. --- ## 3. The No-Humean-Gap (Ch. 2) — Constitutive Reading Available But Not Lipton's **The passage.** Lipton observes that we cannot articulate a gap between meeting our standards for explanation and actually explaining: > We do not appear to know how to make the contrast between understanding and merely seeming to understand in a way that would make sense of the possibility that most of the things that meet all our standards for explanation might nonetheless not really explain. To put the matter another way, we do not see a gap between meeting our standards for the explanation and actually understanding in the way we easily see a gap between meeting our inductive standards and making an inference that is actually correct. (Ch. 2) **Lipton's ambivalence.** He reads this as potentially bad news: > We may say that understanding is the goal of explanation, but we do not have a clear conception of understanding apart from whatever it is our explanations provide. If this is right, the absence of powerful skeptical arguments against explanation does not show that we are in better shape here than we are in the case of inference. Perhaps things are even worse for explanation: here we may not even know what we are *trying* to do. (Ch. 2) **Is the constitutive reading available?** Yes, but it's not Lipton's own position. Three features of the text lean against a constitutive reading as *his* view: 1. The chapter's opening question — "What has to be added to knowledge to yield understanding?" — presupposes understanding is a real addition, not constituted by practice. 2. The conditional "If this is right" marks the claim as hypothetical, not asserted. 3. The rhetorical mood is diagnostic: "we may not even know what we are *trying* to do" treats the absence of independent access as a deficit (confusion), not a dissolution (there's nothing more to be confused about). 4. At the next section break, "Whatever the point of explanation or the true nature of understanding" implies there *is* such a nature, even if Lipton sets it aside. But the text doesn't *block* the constitutive reading either. Lipton never says there IS a fact about understanding independent of practice. The "If this is right" conditional leaves the constitutive reading as a live option in an agnostic space. **How to deploy it honestly.** The paper should: - Quote the no-gap observation. - Note that Lipton treats it as ambiguously significant (possibly good, possibly bad news). - Observe that in philosophy (where evaluation is substantially text-internal), the no-gap observation has particular force: the standards for assessing philosophical quality are publicly checkable properties of texts, not proxies for some independent external reality in the way scientific standards proxy for empirical truth. - Frame this as *extending* Lipton's observation into a domain where it gains additional traction, not as attributing a stronger reading to Lipton. **[NOTE — Feb 2026]:** The earlier suggested deployment language ("evaluative constraints are constitutive of quality, not proxies for some independent fact about it") overclaims. The paper does not argue for a constitutive thesis about philosophy's nature. It argues that philosophy's evaluative standards are substantially text-internal — a methodological claim, not a metaphysical one. Lipton's no-gap observation supports the weaker claim without needing the stronger one. --- ## General Assessment Lipton is best deployed as a foil as much as an ally: *here is what the gap looks like in science (Lipton); here is why it doesn't open in philosophy (us)*. The book's limitations for the paper are clear: Lipton writes about science, and his framework presupposes an external world against which theories can be checked. Many of his arguments apply only by analogy or by contrast, not directly. The paper should be explicit about when it is deploying Lipton as structural precedent and when as a contrastive foil. --- ## Passages for Direct Quotation I'm grouping these by paper section where they'd most naturally appear: **Section 1 (Floridi/Zahavy critiques):** - "Post hoc ergo ad hoc" passage (Ch. 10) — against critics who assume training origin vitiates quality - Archer analogy: "this is a commentary on the scientist, not on the theory" (Ch. 10) — provenance-irrelevance in compressed form **Section 2 (self-grounding / appearance-reality):** - No-Humean-gap: "We do not appear to know how to make the contrast between understanding and merely seeming to understand..." (Ch. 2) - Doing/describing gap: "You may know how to do something without knowing how you do it; indeed, this is the usual situation" (Ch. 1) — defuses opacity objection **Section 3 (saturation thesis):** - Background constitutes standards: "the standard itself will be partially determined by the background" (Ch. 8) - "Reliable evaluation entails privilege": "what we cannot have are inductive powers without inductive achievements" (Ch. 9) - Conditions for harmless accommodation: "In a case where we convince ourselves that there is really only one possible explanation... fudging is not an issue and accommodation is no disadvantage" (Ch. 10) - Realization thesis / squash analogy (Ch. 7) — as capstone --- *La differenza tra appropriare un'osservazione e attribuire una tesi è la stessa che passa tra coltivare un seme e rubare un frutto.*