# I would like to talk to you about this section of my generating philosophy draft. Okay, before we... ## Retrieval Notes - Session id: `95c42e55-3f26-4ba3-9ec0-3ac7db035c23` - Last activity: `2026-04-21T08:33:13.416Z` ## My Notes <!-- Add your notes here. This section is preserved across syncs. --> ## Conversation ### User I would like to talk to you about this section of my generating philosophy draft. Okay, before we go any further, I want you to familiarize yourself with the project by reading the appropriate session file. Looking at the most recent transcript of the conversation between me and Enrico, it happened in April, so find the transcript from April, okay, not the transcript from March. And yeah, once you've familiarized yourself with the project, bear in mind that many of the things written in the long form project are preliminary. Show me that you understand what the project is and then give me a paragraph by paragraph detailed summary of this section of the draft as it is at the moment. Okay, please just answer in the chat for the time being. ### Assistant The April transcript must be "Transcript - Draft Discussion (Sections 0-3).md" (modified April 7). Let me read it. ### Assistant Before diving in: one honest note. There's no April transcript — only two transcripts in the longform folder, one from 20 March (cleaned 7 April, marked superseded) and one from 31 March. The 31 March discussion is the one that actually introduces the "Challenge from Authorship" framing the current Section 1 uses, so I'm treating it as the relevant recent conversation. If you were thinking of a different conversation (e.g. exported into `Conversations/`), tell me and I'll pull it. ## Project, as I understand it A co-authored paper with Enrico Terrone — "Generating Philosophy with AI" — asking whether LLMs can produce philosophy of sufficient quality to be useful, and on what terms philosophers should adopt them. The "text-internal evaluation" draft reframes the dialectic around three challenges to LLM philosophy, each granting a different set of premises: - Section 1 — **Challenge from Authorship** (constitutive): philosophy is a person-only domain; no philosopher, no work. - Section 2 — **Challenge from Abduction** (causal, Floridi): the structure of good philosophical text can only be caused by a reasoner who actually weighs hypotheses. - Section 3 — **Challenge from Experience/Phenomenology** (causal, Zahavy and the epistemology of intuition): certain text-properties can only be caused by a subject with the relevant phenomenal experience. The 31 March transcript crystallises this three-challenge architecture (with Enrico floating a possible fourth — prompting — as symmetrical to authorship). The paper's positive commitment is that "philosophy is in the text": what is to be evaluated is the paper on the page, not the process that produced it. Section 1's job is to clear the ground for that text-focused reading; Sections 2-3 handle the causal worries that remain once the constitutive worry is disarmed. Much of the draft is provisional, voice issues and LLM contamination are being actively scrubbed, and structural moves (e.g. what to do with Einstein, Machery, Mary, Merleau-Ponty in Section 3) are still live. ## Section 1 — "The Challenge from Authorship", paragraph by paragraph **¶1 (Statement of the challenge, via art analogy).** Introduces the "challenge from authorship" as the idea that philosophy is something only persons (or minds) can do. You note that nobody has quite articulated this in print, but you take it to be a widespread intuition. The worked analogy is art: many would say a purely AI-generated image cannot be an *artwork* because no artist stands behind it; by parity, no LLM text could be philosophy because no philosopher stands behind it. Two footnotes queue up (a) a placeholder gesture toward the AI-image literature and (b) a concessive note that denying artwork-status to AI images does not require denying that LLMs can produce *beautiful* images. **¶2 (Disciplinary evidence that philosophy is individual-focused).** Philosophy, like art, is organised around individuals: undergraduates take courses on Kant's ethics or Lewis's metaphysics, specialists and conferences spotlight particular philosophers. Science is contrasted as idea-focused rather than person-focused — no one specialises in Newton qua Newton, no courses on Crick's rather than Watson's views on DNA. This is offered as circumstantial support for the intuition the challenge tries to express. **¶3 (Transition to precisification — work-in-progress).** A short pivot paragraph asking whether the intuition can be sharpened. It currently contains two `%%comments%%` flagging that the transition is unfinished — one asking for something general about "constitution" or "doing", the other signposting that what follows will adapt Davies' performance theory of art. The paragraph then truncates mid-sentence into the Davies move. **¶4 (Davies block quote).** The canonical statement of Davies' performance theory: the work is the artist's "intentionally guided generative performance", not the product (structure or object) that performance leaves behind. **¶5 (Gloss on Davies).** Unpacks the quotation: the painter's act of producing is the work; the canvas is the product — what the activity leaves behind, and what we attend to, but not itself the work. Appreciation is directed at the product only because the product is, in Davies' phrase, "the focus of our appreciative interest in the work." A `%%comment%%` asks for one more sentence explaining how this works. **¶6 (Transposition to philosophy).** Attempts to carry Davies over: the philosophical work is the philosopher's activity in producing the text — her sustained working-through of a problem, her drafting and revising — rather than the text itself. The text is what the activity leaves behind but is where readerly engagement is directed. A substantial `%%comment%%` here worries that the paragraph is obscure, that "work" is oscillating between noun and verb, and suggests the real problem is that *texts* have not yet been properly introduced as the object of interest — flagging this as a structural rather than local fix (possibly: throat-clearing about texts at the very opening, then run the transposition). There is also a `%%comment%%` calling one of the characterisations "a bit shit" and asking to improve it. **¶7 (Why the transposition fails — art case).** The transposition is only as good as Davies' reasons, and those reasons don't carry across. Davies' argument turns on a feature specific to *art appreciation*: engaging with a painting means engaging with what the artist brought off using the resources of a medium. The Vermeer case is worked in some detail — attending to handling of oil paint, compositional decisions, seventeenth-century Dutch conventions. Davies' indistinguishable-surfaces cases are rehearsed (washing-machine "Rembrandt", Danto-style red squares, molecule-for-molecule forgery) to show that two surfaces brought off under different conditions are different works. The paragraph concludes that this is not a disciplinary stipulation but a fact about how art appreciation operates. **¶8 (Philosophical evaluation is text-directed, not producer-directed).** The contrasting claim. Philosophical evaluation asks whether premises are defensible, whether inferences go through, whether distinctions track real divisions, whether counter-examples hold, whether conclusions survive objections — none of which makes reference to the producer. Validity of modus ponens is checkable without knowing who applied it; counter-examples are testable without knowing how the author arrived at them. The indistinguishable-surfaces case has no parallel in philosophy: two papers that *read* identically make the same arguments, face the same objections, admit the same evaluations. An anticipated objection — that philosophers too speak of what an author has *achieved* — is deflated: achievement-talk in philosophy is parasitic on the text; to say the author achieved something just is to say she produced a text with such-and-such properties. **¶9 (Institutional evidence + Dellsén + Sokal).** Disciplinary practice bears out the text-focused reading. Blind peer review strips author information as a matter of principle, treating information about the author as potentially corrupting of assessment. Dellsén et al. (2024) are enlisted for the normative counterpart: philosophical progress consists in putting people in a position to increase their understanding through publicly available philosophical content — a "for-whom" rather than "by-whom" conception (p. 679). Two reasons are given for why the text matters on this picture: (i) we can only evaluate what we have access to, which is the published paper and not the author's private process; (ii) on Dellsén et al., the publication is what makes a contribution available to the discipline. The Sokal / *Social Text* affair (1996) is deployed as the exemplary breach of the norm: what went wrong was that the journal was assessing Sokal's politics rather than the argument on the page. Closes by declining to follow Davies in locating the philosophical work in the generative performance. **¶10 (Framing for the rest of the paper).** Reclassifies the challenge from authorship as a *constitutive* challenge — one on which the philosopher's activity is not merely causally upstream of the work but part of what the work *is*, so that even a text indiscernible from a philosophical paper would fail to be philosophy absent the producer's activity. That challenge has, on the section's showing, failed. The remaining challenges are *causal* rather than constitutive: they concede that philosophy is in the text but argue that the relevant text-properties can only be *caused* by a human subject. Section 2 is named as taking up Floridi et al. (2024) — the claim that the abductive structure good philosophy exhibits can only be caused by a reasoner who actually performs the comparison. Section 3 is named as taking up a parallel concern from Zahavy and the epistemology of intuition — that certain properties of philosophical texts can only be caused by a human subject who has had the relevant phenomenal experiences. Two footnotes here: one surveying metaphilosophical traditions (Overgaard, Gilbert, Burwood on Wittgenstein, Merleau-Ponty, Dilthey) with a Nick-flagged `%%this is shit%%`, and one expanding on the Sokal affair. A few structural observations worth flagging for the conversation that follows: - The section as it stands is doing two jobs in sequence — (a) arguing against the constitutive reading via the Davies-transposition-and-rejection move, (b) setting up the causal/constitutive distinction as the architecture for Sections 2-3. The Dellsén/Sokal paragraph is where the positive text-focused commitment gets most of its institutional weight. - Your own `%%comments%%` mostly cluster around ¶3 and ¶6 — the Davies machinery. The worry about "work" vs text, and about texts not having been introduced first, is the most structural and probably the highest-priority thing to decide before further local polishing. - The 31 March transcript with Enrico treats this section as close to working — "easy to deal with" via the person-based vs text-based distinction plus the peer-review argument. His outstanding local concern from that conversation (relevant to ¶1-2) was whether the science/philosophy contrast is needed *this early*, given that it reappears around Pigliucci in Section 3 and may sit in tension with Dellsén's continuity-of-progress picture. *Nella filosofia analitica, il foglio stampato non è il residuo del lavoro: è il luogo dove il lavoro prende misura di sé.*