# hello ## Retrieval Notes - Session id: `e9c08243-dab3-4591-a3a6-97ae5c597c2d` - Last activity: `2026-03-20T19:10:12.645Z` ## My Notes <!-- Add your notes here. This section is preserved across syncs. --> ## Conversation ### User hello ### Assistant Hey Nick. Let me grab context from your recent daily notes. ### Assistant All good — just testing the connection? I'm here and caught up. Looks like today's been relatively quiet so far, with earlier sessions around the Generating Philosophy project and a video note on Notebook Navigator. What can I help with? *Tutto funziona — il sistema è sveglio e pronto.* ### User I added a transcript of a conversation with Enrico about the generating philosophy paper today. please summarise the whole thing ### Assistant Let me find that transcript. ### Assistant Hmm, not finding an obvious transcript file. Let me search more broadly — you said you "added" it, so it might be in Notes or somewhere else with Enrico's name or "generating" in the content. ### Assistant Found it — `Notes/Generating Philosophy - Enrico Presentation Transcript (Mar 2026).md`. Let me read it. ### User [Request interrupted by user] ### User if that's not from today, you've got the wrong one ### Assistant Let me look for what was added today. ### Assistant None of those were modified today. Could it be somewhere else — maybe in Attachments, or did you paste it into an existing note? Or perhaps it has a different name than I'm expecting? ### Assistant That's it — `Writing/research/generating-philosophy-text-internal-evaluation/Transcript - Draft Discussion (Sections 0-3).md`, modified today at 18:27. Let me read it. ### Assistant Here's a summary of the whole conversation. ## The generating philosophy paper (sections 0–3) Nick and Enrico go through the current draft section by section, mixing line-level edits with structural rethinking. ### Section 0 (Introduction) - The Douglas Adams quotation feels misleading at the opening — it sets up expectations it doesn't fulfil. Two options: (a) move the quotation to a later section on prompting, or (b) replace the opening with a simpler rhetorical contrast between the science-fiction scenario and the real 2026 situation. They lean toward (b), with the Adams detail reappearing later or in the conclusion. - Missing references and a missing date for Stoljar flagged. ### Section 1 (Philosophy in the Text) - "Crystallographic data" may be the wrong term for the Watson-Crick example — Enrico notes it usually isn't biological. - Several sentences flagged as obscure: the "philosophical corpus shaped by repeated judgments" line needs unpacking; the child-grammar analogy doesn't land; "the properties theorists describe" is unclear. - The transition from section 1 into Floridi is too abrupt — needs a bridge paragraph. - Recaps of Williamson and Lipton later in the paper should be signposted as recaps ("as seen above…"). - There are two senses of "likeliness" in play (statistical probability vs Lipton's truth-tracking sense) and the draft conflates them. - Remove Walton — too many names in one paragraph, and the paragraph is too long. - The paragraph beginning "One may still insist…" feels like a conclusion without enough setup. ### Section 2 (Floridi / abduction) — structural rethinking - The Floridi objection as currently framed is too easy to dismiss: "abduction is psychological → but we care about the text → done." Four pages on that feels excessive. - Enrico proposes a stronger reconstruction: Floridi's real challenge is that you cannot have valuable abduction *in the text* without abduction *in the mind*. Not "abduction lives in the mind" but "without the psychological route, the text can't exhibit genuine abductive virtue." That version connects to the zombie/phenomenology objection rather than being a separate, weaker point. ### Section 3 (Thought experiments / Zahavi) — structural rethinking - The Chinese Room reference anticipates section 3's argument too early and should be cut or moved. - The Austin quotation and the "philosophy begins from what has already been articulated" paragraph are praised. - Enrico raises a philosophical problem: the draft doesn't give a strong reason why physics needs embodied phenomenology while philosophy doesn't. Einstein imagining an elevator, Searle imagining a Chinese Room, Jackson imagining Mary — these all look structurally similar. Either philosophy *also* needs phenomenology (which would weaken the response to Zahavi), or physics could *also* get by on descriptions (which would mean Zahavi is wrong across the board). - The Dummett example doesn't seem like a significant discovery on first reading — needs elaboration or replacement. - The Islamic philosophy example wasn't Nick's — it came from the LLM. Needs fixing. - Too many examples in the section overall. ### The agreed paper structure going forward Enrico sketches a unified architecture: 1. Section 1: philosophy's value is in the text (against the practitioner-focused view). 2. Section 2: can you have good text without abduction in the mind? (Floridi, reconstructed strongly.) 3. Section 3: can you have good text without phenomenology in the mind? (Zahavi.) The reply to both has the same shape but different content: - Against Floridi: abductive *patterns* (comparisons between possibilities, inference structures) are already at work in the corpus. The statistical process over this corpus can yield textually abductive results. - Against Zahavi: phenomenological *descriptions* are sedimented in the corpus. The model doesn't need first-person experience because the descriptions are already there. Enrico calls this "abduction-star" and "phenomenology-star" — functional analogues achieved through corpus-level mechanisms rather than psychological ones. ### Other business - They discuss a second paper: the aesthetics of *functioning* (not just function). Engineering, biology, and nature as three cases where the process, not just the outcome, is aesthetically appreciable. This would expand the existing beauty-in-use paper into a broader framework distinguishing function from functioning, with different roles for agency in each domain. - Practical matters: possibly dropping the Hong Kong conference (funding, Nick's back), presenting online instead, keeping Finland as an option. A brief exchange about contract/recruitment frustrations and weekend plans. *Una conversazione che avanza su due fronti — la struttura del paper si chiarisce mentre un secondo progetto prende forma.* ### User nowhere near enough detail. X 3 ### Assistant Let me reread and give you something properly detailed. ### Assistant Here is the detailed summary. --- ## Part 1: The Douglas Adams quotation (Section 0 / Introduction) Nick opens by saying he realised at the workshop that the Douglas Adams quotation at the start of the paper isn't clear. The problem is its role: a reader encountering it at the opening will expect it to set the direction of the argument, but it doesn't really do that. It's doing something more limited — just introducing the idea that in 2026 you can actually put philosophical questions to machines, and that in the Adams case the question was asked badly. Nick wonders whether to drop it or explain it better. Enrico's suggestion: the Adams quotation makes sense if the paper has a section on prompting, because the quotation is about asking badly. Since the paper currently has no prompting section, the quotation at the opening misleads. Two options surface: 1. Replace the opening quotation with a simpler rhetorical move: a contrast between the science-fiction scenario (asking a machine an absurd question) and the real 2026 situation (where you can genuinely ask machines philosophical questions). Just a sentence or two, no extended Adams material. 2. Move the Adams example with full detail to a later section — either a new prompting section, or the conclusion. They settle on option 1 for the opening. Then the question becomes what to do with the Adams material itself. Nick suggests the Adams quotation could work as a conclusion rather than an introduction — "all we have seen may lead to a situation like that." Enrico agrees: prompting could then be deferred to a separate paper. Nick likes this because it means the paper is nearly finished as-is, though he worries it's too much of a tease ("yes, this is all very important, but we're not going to tell you"). Enrico says they could also go for a longer paper and include prompting, but he's still forming those ideas and they aren't ready today. Nick mentions CEV ("coherent extrapolated volition") as an example of what a prompting section could discuss. Someone on Twitter described CEV as a "hack" — you point the phrase at a paragraph and ask the LLM "give me the thing I want even though I don't know what I want yet." It forces the model to extrapolate toward your ideal version. Nick says you can force it to do philosophy for you this way. Enrico finds this interesting but thinks it could be for the next paper. ## Part 2: Line-level edits on Sections 1–2 Enrico works through the draft page by page. These are the specific items: ### References and terminology - Missing reference in "the other updates." Stoljar citation has no date. - Page 3, "crystallographic data": Enrico flags that "crystallographic" usually refers to crystal structure analysis, not biological work. Nick says it's the Watson-Crick example. Enrico's point is that the term may be technically wrong for what Watson and Crick were doing — crystallographic is "almost the opposite of biological." Nick agrees to check and fix. ### The dependence claim - Where Nick introduces "the dependence claim," Enrico suggests adding something like "stands or not" / "depends or not" in brackets, because the passage also stresses the negative case and it should be explicit. ### Factual error: "two speakers on Twitter" - Enrico says it was one speaker, not two. ### Unclear anaphoric reference - The sentence about "we have some call this overfitting" and the sentence following it: the referent isn't clear. "Otherwise it's not clear" what the pronoun is pointing back at. Needs rewriting for clarity. ### "Philosophical corpus" sentence - The sentence: "a philosophical corpus is not just any body of text; it is a body of text shaped by repeated judgments about whether its arguments…" Enrico finds this enigmatic at the point where it appears. It will be explained later, but at this stage a reader will ask: judgment by whom? What exactly does this mean? "Philosophical corpus" is already a non-obvious notion. Either unpack it here or signal that the explanation is coming. ### Transition from section 1 to section 2 - Too abrupt. Section 1 ends, and then suddenly: "Floridi and others argue that…" There's no bridge. The reader doesn't know why they're now hearing about Floridi. Enrico wants a transitional paragraph at the end of section 1 that motivates the move to the next objection. ### Stray commas in a quotation - LLMs keep inserting commas into the quotation. "LLMs are the sort of thing…" — LLMs will keep doing these things even when told not to. ### Floridi's "external feedback loop" - The sentence: "They produce one plausible continuation without weighing it against alternatives, and they do so without what Floridi calls an external feedback loop for posterior evaluation." Enrico says "external feedback loop for posterior evaluation" is not understandable without explanation. Nick suggests the sentence can do without it. Enrico agrees: either cut the Floridi phrase or make it plainer. ### "Weak work survives" / "strong work may be overlooked" - In the passage about the philosophical corpus, Enrico suggests qualifying the claim about overlooked work: add "sometimes" or "may be" so it isn't too assertive. ### Williamson and Lipton recaps - End of page 7: "Williamson uses to have one way of characterising…" This repeats something already in the paper but presents it as new. Signal it as a recap: "as said above." Same for Lipton: "Lipton distinguishes the likeliest from the loveliest" — we already have a whole section on that. Mark it: "as seen above, Lipton distinguishes…" ### Walton — too many names - Before the Lipton paragraph: "as Walton and the others observe, familiar forms of the feasible… critical questions." Enrico says the sentence isn't clear. Nick says he wants to remove Walton anyway — too many people in this paragraph. Enrico agrees the paragraph is also too long. ### Two senses of "likeliness" - There's a conflation in the draft between statistical likeliness (next-token probability) and Lipton's sense of likeliness (truth-tracking, accuracy). Floridi uses the ordinary statistical sense. Lipton's use is "slightly peculiar" — his example is the opium case ("it makes you sleepy"): you've aimed at truth so single-mindedly that the explanation becomes completely useless (a tautology). That's Lipton's "likely" — not "statistically probable" but "maximally truth-tracking at the cost of explanatory depth." The draft needs to keep these senses separate. ### "The properties theorists describe" - The sentence: "A model trained on philosophical corpus is exposed to a distribution systematically skewed toward the properties theorists describe…" Enrico says he cannot understand this. Not that it's badly expressed — he genuinely does not understand what Nick is going for. "The properties" — which properties? "Theorists" — which theorists? Probably means philosophers, but it isn't clear. The concluding sentence of the pre-Lipton paragraph needs complete rewriting. ### The child-grammar analogy - The comparison: "a child can acquire grammatical competence without being given grammatical rules; exposure to enough well-formed speech is sufficient." That part is clear. But the analogy to the LLM doesn't come through. Enrico says it should say more directly: just as a child is exposed to proper language and learns to speak, the model is exposed to philosophy and learns something analogous. Nick suggests dropping the child analogy entirely and just talking about the LLM itself. Enrico agrees, and adds that "the properties theorists describe" is still bad even with this change. ### "One may still insist" paragraph - The paragraph from "One may still insist that the model has only inherited the results of other people's ability…" through to "…that embodies them." Enrico's problem is not just wording ("the reason a standard…" should be "the reason a theory…") but the whole paragraph. It feels like a conclusion without enough setup. The idea is that in philosophy, unlike in science, the justifications for the evaluative standards are themselves part of the textual corpus — so the model has access to why the standards are what they are, not just the results of applying them. But the draft doesn't make this relation explicit. It just asserts the conclusion. Needs more buildup. ### Blind review — soften slightly - "Blind review would be a defective practice, but it is not…" Enrico suggests "surely it is not" — soften slightly, because some people think blind review genuinely isn't a good practice. Don't sound like you're taking it for granted. ### Lipton quotation — abrupt - The Lipton quotation appears "out of the blue." Some connecting step is missing. Nick agrees they're introducing the quotation badly. ## Part 3: Structural rethinking of Section 2 (Floridi / Abduction) This is the most consequential part of the discussion. Enrico argues that the Floridi section as currently written is too weak an objection to warrant four pages. The problem: the current framing presents Floridi's objection as "abduction is a psychological process, it happens in the mind, not in text." But the paper's section 1 has already established that philosophy's value is in the text. So the reply to Floridi is trivially available: "We care about the text, not the mind. Who cares whether the LLM isn't doing psychological abduction?" If that's all Floridi amounts to, four pages is excessive. Enrico proposes a stronger reconstruction of Floridi — two interpretations: Interpretation 1 (weak, currently in the draft): Abduction is not in the text, it's in the mind. This is easily dismissed by section 1. Interpretation 2 (strong, what the draft should adopt): Even if value is in the text, you cannot have valuable abduction *in the text* without abduction *in the mind*. The claim isn't that abduction lives in the mind instead of the text. It's that the psychological process is a necessary route to the textual product. Without human-style abduction upstream, the text cannot exhibit genuine abductive virtues. Enrico draws a parallel to the zombie argument: the zombie has no problem accepting that value is in the text — it just denies that you can produce a valuable text without the right mental processes. Floridi, reconstructed strongly, makes the same structural move about abduction specifically. This reconstruction also connects better to the paradigm-shift point: the issue becomes not just whether there is abduction, but whether there is *good*, genuinely creative abduction. Nick asks Enrico to repeat the reframing so he can be sure he has it right for later. Enrico does: the interesting reading of Floridi is "even if value is in the text, you cannot have valuable abduction in the text without abduction in the mind." Section 2 should construct Floridi's argument in this strongest form rather than knocking down the weaker version. Nick then suggests that the weaker Floridi point could be folded into section 1: section 1 talks about philosophy being in the text, and along the way notes that Floridi is right that the mechanism is entirely different from what we normally mean by abduction — but that doesn't matter if the text has the right features. This would free section 2 to deal exclusively with the stronger version. Enrico says the PowerPoint presentation had a better framing: first the continental point (value is in the text), then the harder move — even accepting that, someone can still say the text can't have value because the *process* is wrong. ## Part 4: Structural rethinking of Section 3 (Zahavi / Thought experiments / Phenomenology) ### Chinese Room placement Enrico says the Chinese Room reference at the end of the opening paragraph of section 3 — "rule-following without comprehension" — anticipates the section's argument too early and should be cut or moved. The Chinese Room's force isn't just about rule-following; it's about Searle's feeling of not understanding anything while executing the rules. Reducing it to "rule-following" misses this. The passage from "the same holds for the Chinese room…" up to "Einstein's imagination, perhaps" should go. Nick suggests putting the Chinese Room after the Austin paragraph ("philosophy does not usually begin from raw encounter; it begins from what has already been articulated"). Enrico is cautious: Searle is also a thought experiment relying on first-person experience, so it faces the same problem as Einstein — which leads into his deeper concern. ### The Austin paragraph — praised Enrico says the paragraph ending "philosophy does not usually begin from raw encounter; it begins from what has already been articulated" is very good. ### The philosophical problem: why is physics different from philosophy? This is the other consequential structural concern. Enrico says the draft doesn't give a strong enough reason to think physics and philosophy are relevantly different with respect to phenomenology. The argument as Enrico understands it: Zahavi says Einstein needed embodied sensory imagination (imagining himself in a falling elevator) to arrive at the equivalence principle. The paper wants to say this challenge bites harder for physics than for philosophy, because philosophy already has descriptions of experience in its corpus. But Enrico doesn't see why the distinction holds: - Einstein's elevator experience can also be described in text. If the model can use descriptions instead of direct experience for philosophy, why can't it do the same for physics? - Searle imagining himself in the Chinese Room, Jackson imagining Mary seeing colour for the first time, Einstein imagining himself in an elevator — these are all structurally similar. They all rely on introspection and first-person experience being turned into a thought experiment. - Maybe Einstein is more *embodied* (physical sensation of falling) whereas Searle or the trolley problem are less so. But Merleau-Ponty would be just as embodied as Einstein. So the domain lines don't cut cleanly. - The point of philosophical thought experiments is to find cases that don't already exist in the corpus. Mary's black-and-white room isn't described in prior experience reports. So it looks like exactly the Einstein problem — the novelty isn't in the corpus. - Unless you say: philosophy draws on descriptions, physics draws on numbers. But that seems too simple. Enrico presents two options: Option A: Accept that the paper is also challenging Zahavi in the end — arguing that even physics could get by on descriptions, and Zahavi is wrong across the board. Option B: Find reasons why physics genuinely works differently — perhaps because physics training sets are data-based rather than description-based. But even this is shaky, because physics notebooks (like Einstein's reflections) do contain descriptions. Nick mentions an author (name unclear in transcript) who is specifically about the difference between science and philosophy, and who is a scientist — "philosophy is a vocation of concepts." He says the physical world comes in differently for physics versus philosophy. He'd sent this to Enrico already. Enrico says he hasn't read it yet but will, and thinks it might help. ### The Dummett example Enrico says the Dummett case as presented doesn't seem like a significant philosophical discovery. On one reading it's just: the meaning of a sub-sentence contributes to the meaning of the larger sentence. That seems obvious — a ten-year-old could say it. Probably it's more complex, but the draft doesn't convey that complexity. Either elaborate to make the depth visible, or use a different example. ### The Islamic philosophy example Enrico says he'd never heard of this distinction before and it looked like an obvious distinction that doesn't require experience. Nick admits it came from the LLM in the most recent version — it's not his idea. Needs fixing. ### Too many examples Enrico flags that section 3 has too many examples overall. Needs distillation. ## Part 5: The unified paper structure Enrico sketches the architecture that emerges from all the above: Section 1: Philosophy's value is in the text. (Against the practitioner-focused/continental view. The text, not the biography or mental life of the philosopher, is where philosophical value lives.) Section 2: Can you have good philosophy in the text without abduction in the mind? (Floridi, reconstructed in the strong form. Not "abduction is psychological" but "without psychological abduction, the text can't exhibit genuine abductive virtue.") Section 3: Can you have good philosophy in the text without embodied phenomenology in the mind? (Zahavi. Without sensory imagination and first-person experience, certain texts — involving thought experiments, introspection, qualia — allegedly can't be produced.) The framing that unifies sections 2 and 3: both are instances of the claim "you cannot have a good text without the proper psychological process." Both accept that value is in the text (they grant section 1) but insist on a necessary psychological route to the text. The replies have the same general shape but differ in content: Reply to Floridi (abduction): The forms of reasoning — comparisons between possibilities, weighing theories against each other, inferential structures — are already *at work* in the philosophical corpus. Not just described, but performed. The statistical process over this corpus can yield textually abductive results because it's tracking patterns that are themselves abductive. The model doesn't need to literally compare theories; the comparison patterns are already in the distribution. Enrico calls this "abduction-star." Reply to Zahavi (phenomenology): Phenomenological experience is *described* in the corpus. Einstein turned his elevator intuition into a description. Jackson turned the Mary thought experiment into a description. Searle turned the Chinese Room experience into a description. What enters the philosophical argument is always the description, not the raw experience itself. If the descriptions are already in the training data, the model has what it needs. Enrico calls this "phenomenology-star." The distinction between the two replies: for Floridi, it's about the *mechanism* — abductive patterns are at work in the corpus and the model picks them up through statistics. For Zahavi, it's about the *content* — descriptions of phenomenological experience are in the corpus and the model has access to them as data. Nick confirms: "for one of them we're saying something about the training data, specifically descriptions in the data. And for the other we're saying something more about the mechanism itself." Enrico says the conclusion can then unify: there is really one objection with two faces — a functional/inferential face (Floridi) and a phenomenological face (Zahavi). The corpus concept is what allows both replies to work, though in different ways. Enrico's diagnosis of the current draft: it reads as "a patchwork of interesting ideas without a clear structure tying these two objections together." The unified framing should be stated in the introduction and maintained throughout. Nick says he'll work on section 3 first, since it's closer to ready. He mentions colour experience as a good example — he's been working on notes where the LLM makes fine-grained distinctions about specific shades ("if you use this shade it won't pop as much as so-and-so"). Those fine-grained perceptual distinctions are a nice case of phenomenology being sedimented in text. Enrico agrees and connects it to the Austin quotation. But Enrico immediately raises the problem again: colour experience seems to apply to physics too. And the genuinely innovative philosophical cases (Mary, zombies, Chinese Room) don't look like colour experience — they look like Einstein: novel scenarios requiring first-person imagination that isn't already in the corpus. The section still needs to deal with this. The paper is currently under 7,000 words. Enrico thinks with this restructuring and a bit more work on the physics/philosophy distinction, the paper is nearly there. ## Part 6: The aesthetics of functioning (second paper idea) Nick raises the aesthetics of functioning paper. Enrico says he's presenting related material at the Politecnico di Torino (invited by Verity, who works on ethics of technology at the engineering school). He'll cite the forthcoming paper — not the beauty paper but the engineering aesthetics paper. Enrico's sketch of the expanded framework: the existing paper is about the aesthetics of engineering. But the real concept is broader — the aesthetics of *functioning*, of which engineering is one instance. The function/functioning distinction: function is the outcome of a behaviour (what a thing does, what it's for). Functioning is the process by which it does it (how it operates, the dynamics of its operation). Philosophy of technology talks about function all the time but conflates it with functioning. They're different, and aesthetically they matter differently. Two machines can perform the same function but have different functioning. You can appreciate the design of a machine as looking fit for its function. But you can also appreciate the way it operates — the process, the movement, the rhythm. Enrico proposes extending beyond technology to nature: in biology, functioning is just as relevant. Organisms have functions (adapting, reproducing), and organs have functions (the heart pumps blood, the eye sees). But the heart can be beautiful for its rhythm, the way it moves — not just for being fit for purpose. Wings are beautiful for the way they function, not just because they enable flight. The proposed paper would compare three domains of functioning, each with a different relationship to agency: 1. Human functioning: Conscious activity is constitutive of the process. A person using a utensil with grace and skill — Montero's expertise and bodily knowledge. The agent is doing the functioning. 2. Technological functioning: The process itself doesn't have agency, but there is an intentional source (the designer) who shaped how it functions. The clock, the bridge — designed processes. 3. Natural functioning: Neither the process nor its source involves intentional agency. The heart, the wings, biological processes. Nick adds: there could be contrast cases where the same job is done by a person with a utensil, by a machine, and by a natural process. Three kinds of beauty in the same functional neighbourhood. Nick also mentions the aesthetics of causality as a related thread. Enrico raises an additional question: the black-box problem. Can we appreciate the functioning of something we can't see inside (like a computer)? His thought: we can have a kind of phenomenology of it — we can imagine the functioning even without looking inside. Enrico thinks the paper wouldn't be expensive to produce — they could take the existing beauty-in-use paper, rework it with the broader framing (function vs functioning across engineering, biology, nature), and it would be substantive. Nick says he could feed the transcripts into the LLM and see what it generates. ## Part 7: Practical/logistical matters ### Hong Kong conference They discuss dropping the Hong Kong conference. Enrico says it's a pity — nice room, nice people, and it would have been a stepping stone to his honeymoon. But with Nick's back issues, presenting online is probably wiser. The online system apparently works well. ### Finland Enrico didn't get into Finland. But if they drop Hong Kong, they'll still have funding for Finland. ### Japan trip Nick and Natalie want to go to Japan. They need to work out dates that work for both of them. Natalie would like to go before summer. ### Administrative frustration There's an annoying email about some risk issue. Enrico asks Nick to handle the complaint, since Enrico "always complains" and it would carry more weight from someone else. Nick doesn't mind. He mentions a related frustration: his contract extension was organised last year with a hands-up vote, then nothing happened for two and a half months, and then when he was in hospital someone suddenly demanded he sign urgently. ### Enrico's week Working this weekend, going to Turin next week for the Politecnico talk (the engineers, Verity's invitation). ### People mentioned in passing - Maria Silvia — told Enrico something (unclear) - Verity — works on ethics of technology at the engineering school in Torino, invited Enrico, was the chair of something (unclear) in Turin - Maria, Michelle, Christina — possibly attending something in person --- *Due ore dense di lavoro — il paper adesso ha un'architettura, anche se i mattoni restano da posare.*