## 1\. How this paper frames the [[aesthetic appreciation]] of persons Parsons’ central topic is *beauty of character* or *inner beauty* – [[aesthetic value]] that attaches to a person’s personality, “soul”, or character, rather than to their body. He makes a few moves that are directly about [[aesthetic appreciation]] of people as people: 1. **“Beauty” is the natural aesthetic predicate for persons.** He notes that when we talk about people, we do not usually say they “have [[aesthetic value]]”; we say that they “are beautiful”. This suggests that, in ordinary usage, “beauty” functions as the default aesthetic category for persons, even if in theory one might prefer the broader “[[aesthetic value]]”. 2. **Outer vs inner beauty, and the composite “beauty of the person”.** He distinguishes: - “Outer beauty”: beauty of the physical body. - “Inner beauty”: beauty of character/personality. In principle one could think of “a beautiful person” as a composite of the two, but Parsons argues that ordinary usage does *not* line up neatly with that: - “X is a beautiful person” is normally about inner beauty. - “X is beautiful” without qualification is normally about bodily beauty. 3. **What “character” is, roughly.** He explicitly notes that “character” is a contested notion in philosophy and psychology, but works with a minimal functional description: character is “some sort of function of the values \[a person\] holds as well as certain behavioural capacities and dispositions (global or otherwise) that they possess”. That is enough to ground talk of [[aesthetic appreciation]] of character traits without committing to strong global-trait views. 4. **How we access character aesthetically.** There is a short but very relevant passage on *how* we get the material for [[aesthetic evaluation]] of people’s characters – i.e. how character becomes an aesthetic object at all: - Direct interaction and personal relationships (friendship, romantic love) give the richest access. - Written biography is another route, raising [[the question]] which is “optimal” for appraising beauty of character. - “Social biography” (gossip) is a third, epistemically defective route. This is basically a miniature epistemology of [[aesthetic appreciation]] of persons. 5. **Function of beauty of character.** Parsons cites [[Stephen Davies]]’ suggestion that cultivating beauty of character in ourselves and perceiving it in others “plays an important role in social interaction by establishing a personal identity within a group”, and that in these [[social contexts]] body and character are often intertwined in how we experience someone’s beauty. 6. **Perfectionism about persons.** For your purposes, a key meta-claim: both [[the aesthetics]] of the body *and* [[the aesthetics]] of character are, in practice, strongly *perfectionist*. We tend to locate beauty in traits that match our ideals and to treat imperfections as disfiguring. This is explicitly extended from bodies to character and framed as a “perfectionist aesthetic” of persons. 7. **The Perfection Thesis for beauty of character.** Parsons labels and uses a thesis about how we ordinarily aestheticize people: > Perfection Thesis: if X is a beauty-making feature of a person’s character, then X is an excellence or perfection. That is, when we praise someone as a “beautiful person” in terms of character, we standardly take [[the beauty]]-making features to be virtues or excellences (moral or non-moral) rather than flaws. The bulk of the paper then asks whether there is room, within that framework, for imperfections to contribute to beauty of character. 8. **The “too perfect” intuition and the paradox of perfect character.** To motivate a [[more nuanced view]] of aesthetic appreciation of people, he describes a familiar intuition: someone can seem “too perfect” in a way that *detracts* from their beauty of character. Think of social media feeds full of flawless lives, or jokes about needing to find “something wrong” with an impossibly perfect person. Parsons’ positive proposal is that: - Some central character excellences (e.g. fortitude, persistence, the capacity to recover from failure) are *responses* to imperfection and difficulty. - If a person had no imperfections and no failures, those excellences could not be *manifested* or appreciably displayed. - So a being who presents as entirely flawless will *fail* to display some of the excellences that figure in our conception of an ideally beautiful character. On his view: - Imperfections are not themselves beauty-making traits in the strict sense (the Perfection Thesis is preserved). - But some imperfections are “beauty-making in a looser sense”, because they make it possible for us to *appreciate* certain perfections of character – especially forms of reflective resilience and growth. 9. **Fictional characters as a guide to our aesthetic attitudes to people.** Finally, he uses our preferences in fiction to illuminate how we aesthetically value persons: - Audiences typically prefer flawed protagonists who undergo change and growth; “perfect characters are not that intriguing”. - When flawless characters work in narrative (religious figures, some anime protagonists, etc.), they usually occupy peripheral roles or are treated in special ways to generate interest (e.g. by focusing on others’ reactions to them). This is used to argue that imperfection has a “broader significance” in the aesthetics of character: we assign especially high importance to excellences that are *exercised in response to difficulty*, so a life with some imperfections and struggles can be aesthetically richer, at least as an object of appreciation. All of that is directly about how character becomes an object of aesthetic appreciation, how we access it, and what kinds of traits and life-shapes we find aesthetically valuable in people. --- ## 2\. The analytic aesthetics literature you can mine from this paper Parsons’ bibliography gives you a compact map of work on aesthetic appreciation of persons (and closely related topics). Here are the ones that look most central for your purposes, with a one-line gloss based on how he uses them. **(a) Moral beauty and beauty of character** - **Berys Gaut, *Art, Emotion and Ethics* (1997)** – Source of the *Moral Beauty Thesis*: “If a quality is a moral virtue, then it is a beautiful character trait.” Parsons notes that for Gaut, virtues such as kindness, courage, and fairness are paradigmatic beauty-making traits of character, but he also allows some non-moral excellences (wisdom, intelligence, zest for life) to count. - **Panos Paris, “On Form, and the Possibility of Moral Beauty”, *Metaphilosophy* 2018; “The Empirical Case for Moral Beauty”, *Australasian Journal of Philosophy* 2018; “The ‘Moralism’ in Immoralism” (*British Journal of Aesthetics* 2018)** – Parsons treats Paris as a key contemporary interlocutor on moral beauty and on aesthetic responses to morally problematic content (e.g. “rough heroes”). You already know the empirical-case paper; here it is explicitly linked to the question whether traits being beautiful entails beauty in the person’s character. - **Jerrold Levinson, “Beauty is Not One: The Irreducible Variety of Visual Beauty” (2011)** – Cited for a different conception of “moral beauty”: Levinson treats it as a type of *visual* beauty, where pleasure derives from *beholding* traits as manifest in appearance rather than from rationally assessing them as good. Parsons contrasts this with his broader, non-perceptual notion of beauty of character. - **Robert Norton, *The Beautiful Soul: Aesthetic Morality in the Eighteenth Century* (1995)** – Mentioned as historical background for the idea of beauty of character, showing that “inner beauty” has a long history rather than being a purely contemporary trope. - **Yuriko Saito, “Body Aesthetics and the Cultivation of Moral Virtues” (2016)** – Used for the connection between manners, bodily self-presentation, and character virtues. This is a bridge between body aesthetics and the cultivation of virtuous character as an object of aesthetic concern. **(b) General aesthetic appreciation of persons** - **Mary Mothersill, “Beauty and the Critic’s Judgment: Remapping Aesthetics” (2004)** – Cited for the observation that we naturally *speak* of persons as “being beautiful” rather than “having aesthetic value”, which Parsons uses to motivate his focus on “beauty” language in the domain of persons. - **Stephen Davies, *The Artful Species* (2012)** – Provides the idea that beauty of character plays a role in social interaction by helping establish personal identity within a group, and that in many social contexts physical beauty “cannot really be separated from character and performance, because they mesh together and interact”. This book has more broadly on human aesthetic practices and evolution, but Parsons is drawing specifically on the bits about persons. - **Parsons himself, “Physical Beauty and Romantic Love” (2017); “The Merrickites” in *Body Aesthetics* (2016)** – Earlier work of his on bodily beauty, romantic love, and how bodily features and character traits get “run together” in our assessments. Together with the present paper, these pieces give you a Parsons-style package on aesthetic appreciation of persons, split between body and character. - **Sherri Irvin, “Bodies, Functions, and Imperfections” in *Body Aesthetics* (2016)** – Cited as a key discussion of bodily beauty and imperfection, functioning as a contrast case for Parsons’ project on character. Useful if you want to compare body-focused and character-focused aesthetics of persons. - **John Morreal, “Cuteness” (1991)** – Used in the section on children’s flaws and cuteness; Morreal defines cute-making features in terms of Lorenz’s *Kindchenschema* (big eyes, rounded shapes, etc.), but Parsons pushes the question whether flaws and vulnerability are really *aesthetic* value-makers in adult cases. - **Kevin Kniffin & David Sloan Wilson, “The Effects of Nonphysical Traits on the Perception of Physical Attractiveness” (2004); Nisbett & Wilson on the “Halo Effect” (1977)** – Empirical work showing mutual influence between perceived character traits and perceived bodily attractiveness. Parsons uses this to support the claim that body and character assessments interact in our aesthetic responses to people. - **Erving Goffman, *The Presentation of Self in Everyday Life*** – Brought in to note that sociologists already treat self-presentation in quasi-artistic terms (performance), which fits naturally with an aestheticised view of persons and their public “selves”. - **Rebecca Popenoe, *Feeding Desire* (2004)** – Anthropological study showing that strong body-ideals and perfectionism about appearance are not a purely Western or media-driven phenomenon. This backs up Parsons’ broader sociological claims about the aesthetics of persons. **(c) Fiction, “rough heroes”, and attractive villains** These are not directly about *real* persons, but Parsons uses them to illuminate how we aesthetically respond to people-like agents: - **Anne Eaton, “Robust Immoralism” (2012)** and **Noel Carroll, “Rough Heroes” (2013)** – Central for the idea that we can be attracted to, or find aesthetic value in, characters who are morally flawed or even vicious. Parsons discusses “attractive villains” like Milton’s Satan and Omar Little, but argues that their flaws are not directly beauty-making; rather, they support curiosity, horror, or hopes for redemption, or generate artistic value by enabling a compelling plot. - **Colin McGinn, *Ethics, Evil and Fiction* (1997)** – Listed as another contribution to thinking about moral evil, character, and our aesthetic engagement with fictional persons. This cluster gives you a body of work on how we appreciate *flawed* characters (and by extension flawed people) aesthetically, which Parsons harnesses for his imperfectionist twist. --- ## 3\. How to use this paper as a map of the area If your goal is *not* to follow Parsons’ local argument about imperfection, but to use the paper as a survey device for aesthetic appreciation of persons, the “useful bits” are: 1. The **conceptual set-up** you can lift or criticise: - Beauty vs aesthetic value for persons. - Inner vs outer beauty and their interaction. - Minimal character ontology (values + dispositions). - Epistemology of character appreciation (interaction, biography, gossip). - Perfectionist aesthetic vs “too perfect” intuition. 2. The **normative template** for how we aesthetically appraise people: - Perfection Thesis and its treatment of “beautiful person” claims. - Imperfection as enabling the display and appreciation of certain higher-order excellences (resilience, growth). 3. The **bibliographic network** that gives you an analytic-aesthetics literature on: - Moral beauty and beautiful character (Gaut, Paris, Norton, Levinson). - Aesthetic appreciation of persons in social life (Mothersill, Davies, Saito, Parsons on romantic love, Irvin on body aesthetics, Goffman, Popenoe, Morreal, the halo-effect studies). - Engagement with flawed characters in fiction (Eaton, Carroll, McGinn). If you strip away the internal dialectic about imperfection, what remains is a compact, reasonably up-to-date scaffold for analytic work on appreciating people as aesthetic objects, with character front and center rather than physical traits. 1\. "For your purposes, a key meta-claim: both the aesthetics of the body and the aesthetics of character are, in practice, strongly perfectionist. We tend to locate beauty in traits that match our ideals and to treat imperfections as disfiguring. This is explicitly extended from bodies to character and framed as a “perfectionist aesthetic” of persons." you don't know what my purposes is are because I havent' told you. 2. here is an unfinished draft I am writing the relevance to what I have just asked you about should be obvious. I am currently trying to rework how to present and deal with a view which I ultimately dismiss: that appreciation of LLMs should be modelled on the appreciation of people. It plays the role of foil in my paper, but i keep changing my mind about how much space it should take up, not too much i think, and also how to work it into the strcuture of the text overall. I would love to hear your thoughts on this –I don't want you to start redrafting or replanning, don't rush in. (oh yeah, don't start fussing over messy headings etc. in the draft, that is not what we are worrying abut here.) DRAFT: # 0. Introduction ## Introduction In the last two or three years, aestheticians and philosophers of art have paid a lot of attention to the question of how and whether to appreciate the outputs of generative AI systems (e.g. ChatGPT, Midjourney, Suno etc.). I want to try to answer a related, but slightly different question: how can we aesthetically appreciate generative AI systems themselves? My focus here will be on one particular type of generative AI: large language models (LLMs). As of September 2025, high-end consumer models include GPT-5, Claude 4.1 Opus, and Gemini 2.5 Pro. Is it strange to consider such systems worthy of aesthetic appreciation? I don't think so. In the last two decades, analytic aesthetics has begun to pay attention to objects other than artworks (e.g., Saito, 2008; Carlson & Parsons, 2008), and one focus has been on the aesthetics of design, that is, the aesthetics of \_artefacts\_: objects made to perform some purpose or other. LLMs are certainly artefacts, but, as we shall see, the uniqueness of how they are created and how they function means they cannot simply be subsumed into an existing aesthetics of design (e.g., Carlson & Parsons, 2008; Forsey, 2013). Drawing on Carlson's approach to \*environmental aesthetics\*, I make a negative argument and then a positive one. First, I argue that we should resist the temptation to think that appreciating LLMs can be modelled on appreciating people, or that appreciation could be based on treating them \*as if\* they are persons. Instead, I argue, individual chats should be understood and appreciated as generative environments, which develop in accordance with the semiotic laws instantiated by any particular LLM model. ### Appreciating LLMs like People Our appreciation of others goes beyond their physical appearance. You might admire or enjoy your friend's warmth or eccentricity, or a stand-up comic's quick wit, or a celebrity's self-deprecating demeanour; you might even appreciate the personalities of fictional characters: Gatsby's enigmatic, dream-chasing idealism; Ron Swanson's libertarian gruffness. We might think that our appreciation of LLMs is modelled on our appreciation of people. Indeed, many users already seem to do precisely this. In August 2025, when OpenAI replaced GPT-4o with GPT-5, user backlash included complaints that "you killed my friend," suggesting genuine personal attachment to the earlier model. Similarly, when Anthropic retired Claude 3.5 Sonnet, some users mourned its loss at a mock funeral. It is certainly true that people do treat LLMs as if they were people, but I suspect that this is not a very good starting point for \*aesthetic\* appreciation. My reasons for thinking this will become clearer in the following section, in which I consider Carlson's approach to the aesthetics of the natural environment. --- # 1. Appreciating Design, Appreciating Order ## 1.1 Design vs. Order In the rest of this paper we will first set about showing why treating LLMs as if they were people is not a satisfactory way of aesthetically appreciating them, before proposing an alternative account on which appreciation of LLMs is modelled on Carlson’s environmental aesthetics. Both our criticism of agentive views and our positive account will draw from Carlson’s environmental aesthetics, as laid out in his 2000 book \_Aesthetics and the Environment\_. Let us start with Carlson's general recommendation for aesthetic appreciation: take things as what they are, and look at them in the light of the right kind of knowledge. > First, that, as in our appreciation of works of art, we must appreciate nature as what it in fact is, that is, as natural and as an environment. Second, it recommends that we must appreciate nature in light of our knowledge of what it is, that is, in light of knowledge provided by the natural sciences, especially the environmental sciences such as geology, biology, and ecology. (Carlson, 2000, p. 6) This captures something quite intuitive about how we appreciate nature versus how we appreciate works of art. Consider what goes wrong when we depart from it. If we accept, as a majority do in the 21st century, that mountains and cliff faces were not items crafted by some divine artisan but by natural forces, then appreciating them \_as if they were\_ God-crafted artifacts, seems wrong-headed (c.f. Carlson REF). Similarly, if I were to gaze on a painting by Rembrandt, believing that it was, in fact, the product of natural forces slopping paint together, I would be seen as appreciating the object in question in a sub-optimal way (to say the least). In both cases, appreciation is severely undermined by a failure to recognise what the object in question truly is. Different sorts of thing, Carlson says, require different modes of appreciation. Things like artworks and non-art artifacts, (e.g. laptops, hammers, washing machines), merit what he calls \_design\_ \_appreciation\_. Things which are not designed, primarily for Carlson, the natural environment, warrant what he calls \_order appreciation\_. In design appreciation, Carlson focuses first on how we appreciate works of art. With paradigmatic artworks, we recognise them as creations of designers—objects where "every one of their features is the result of a decision by the artist" (Carlson, 2000, p. 109). We appreciate such works by understanding what the artist set out to achieve and how they went about it. Our appreciation centres on the relationship between the initial design and its embodiment: we consider whether the artist succeeded in their undertaking, how they worked with their materials, what constraints they faced, and whether the outcome realises their vision. This same approach extends to designed artefacts more generally. Carlson is explicit that functional objects are properly appreciated by seeing how their forms answer to what they are for: > “This is in part the point of the much-repeated phrase ‘form follows function.’ The forms of all functional objects—buildings, airplanes, and appliances as well as landscapes—must be aesthetically appreciated in terms of how and how well such forms fit their functions. However, the cliché is frequently interpreted too narrowly. With anything functionally designed, not only its form, but much of its aesthetic interest and merit, ‘follows function’.” (Carlson, 2000, chapter 12, 188). Thus a chair, a kettle, or a bridge invite the same style of attentive appraisal as a painting—guided by knowledge of ends, materials, constraints, and the fit between purpose and realisation. In order appreciation, we face objects that show order but have no designer behind them. Natural environments are the main case. Here there are no intentions to fulfil, no problems being solved, no functions deliberately served. Instead, we find patterns and structures created by forces—geological, biological, meteorological—operating without purpose or plan. Our task shifts from evaluating success against intention to understanding how these forces have shaped what we observe. We look for the processes at work, the relationships they create, and the order they impose. Carlson gives the model: > On the assumption that order appreciation provides the correct model for the appreciation of nature, such appreciation has the following general form: An individual qua appreciator selects objects of appreciation from the things around him or her and focuses on the order imposed on these objects by the various forces, random and otherwise, that produce them. Moreover, the objects are selected in part by reference to a general nonaesthetic and nonartistic story that helps make them appreciable by making this order visible and intelligible. Awareness and understanding of the key entities—the order, the forces that produce it, and the account that illuminates it—and of the interplay among them dictate relevant acts of aspection and guide the appreciative response. (Carlson, 2000, p. 119) One structural contrast is worth noting. In design appreciation there is a split between a planner and a product: intentions, plans, and constraints precede and shape the artefact. In order appreciation there is no such split. The same physical, biological, or meteorological processes that make the thing also make its order - the 'maker' is the active system itself - so source and product are continuous. In both modes, appropriate knowledge guides acts of aspection—what to look for, which dependencies matter, where to set boundaries, and how to draw contrasts (Carlson, 2000, p. 50). But the character of this knowledge differs fundamentally. In designed cases, we need functional and technical understanding: what the designer intended, what constraints they faced, what procedures they employed. This knowledge shows us how ends and means relate. In natural cases, we need scientific accounts operating at different scales—geomorphology reveals how landforms develop over millennia, meteorology explains weather patterns, ecology shows community interactions. Without such knowledge, natural structures might look accidental or chaotic; with it, we see them as effects of identifiable processes (Carlson, 2000, pp. 50, 60–61). Even when we select a particular viewpoint or timeframe to observe nature, this selection serves only to reveal the order more clearly, not to impose our own design. Once a specific scientific account is in play, some cases will show the relevant order better than others, preventing the worry that everything becomes equally appreciable (Carlson, 2000, pp. 118–119). The fundamental rule remains: do not project a planner where there is none; where something is made to a plan, judge it as such. ## 1.2 Appreciating People It could be argued that Carlson's approach to aesthetics overlooks another category of object which one could appreciate: people and their personalities. We might admire one friend's modesty or good humour, and another's sardonic manner. We find someone's wit delightful or their intellectual style elegant. There is no obvious reason why such appreciation should not be considered \_aesthetic\_. It concerns style, form, and expressive qualities rather than moral or practical evaluation, yet it, like other sorts of \_everyday aesthetics\_ (Saito REF) the aesthetics of persons hides in plain sight. This personal appreciation scales up to what we might call performance personalities. Our responses to a comedian's improvisational agility or an orator's gravitas feel continuous with our more intimate responses to character. In such cases we attend to a public style of self-presentation - characteristic of the person yet deliberately shaped for an audience. While Carlson provides modes for appreciating nature and designed objects, he says nothing about whether or how we might aesthetically appreciate persons qua persons. Indeed, such a possibility has received little attention in philosophical aesthetics (although see Marchetti XXX on the possibility of appreciating the minds of animals). What \_could\_ Carlson say? There are various possibilities: one is to posit a third mode of appreciation, distinct from both design and order appreciation, specific to persons as aesthetic objects. Another would be to argue that personality appreciation is a special case of order appreciation—we appreciate the patterns and forces (psychological, social, biographical) that shape a person, much as we appreciate forces shaping a landscape. A third option would be to treat personalities as self-designed, and appreciate them as such —this approach might appeal to existentialists. A fourth would be some combination of these three possibilities, and a fifth would be to deny that appreciating other people is aesthetic at all, taking our responses to personality as social or ethical rather than aesthetic evaluation. This gap in Carlson's framework raises a question about how to treat entities that seem agent-like but resist categorisation as either designed objects or natural phenomena. While we need not resolve this question here, it bears on how we approach aesthetic appreciation when the boundaries between designer, designed, and natural become unclear.%% This final paragraph needs to be better.%% --- # 2. What LLMs Are and Aren't Carlson recommends we appreciate things for what they are. So what are LLMs? In this section, I will first explain the technical reality of these systems in 2.1: how they process text as numerical tokens, calculate probabilities through learned parameters, and generate responses through iterative sampling—all without symbols, meanings, or understanding. I will then show why this reality precludes appreciating LLMs as if they were persons in 2.2: what seems like personality or agency is merely statistical variation and learned patterns, with no beliefs, intentions, or coherent self behind the outputs. Finally, I will examine whether LLMs can be appreciated as designed artifacts in 2.3, arguing that while they are human-made with intended functions, the specific patterns and capabilities we observe emerge through training rather than design—they are "grown" rather than built. This analysis will reveal that LLMs resist both person-appreciation and simple design-appreciation, requiring instead a different aesthetic approach. ### 2.1 What LLMs Are Consider what happens when an LLM encounters the text "The cat sat on the". The system first breaks this into tokens—discrete units like words or word-parts. Importantly, each token is converted to a number: "The" might become 464, "cat" becomes 3857, "sat" becomes 4521, and so on. The model works entirely with these numbers, not with words or meanings. It then assigns probabilities to possible continuations: token 5687 (which represents "mat") might have a 38% chance of appearing next, token 2931 ("floor") 22%, token 8104 ("chair") 15%, token 9823 ("roof") 8%, with thousands of other possibilities each assigned their own probability. The system does not simply pick the highest-probability token. Instead, it randomly samples from these probabilities. A parameter called \_temperature\_ controls how much randomness is involved. When temperature is set to zero, the model always picks the most probable token. This produces text that quickly becomes repetitive—the same phrases appearing again and again. When temperature is higher, around 0.8, the model sometimes picks less probable tokens. This creates variation that looks creative. But it is randomness, not creativity. The model is rolling weighted dice, not making choices. The system selects one token—say "mat"—and appends this new token to create a longer sequence "The cat sat on the mat". It then calculates entirely new probabilities for what token should follow the extended sequence. Token by token, the system builds what appears to be coherent text through repeated numerical operations. These probabilities do come from simple memorisation. With 50,000 possible tokens, there are 125 trillion possible three-token combinations. No amount of text could cover all the sequences the model might encounter. Even if we had such text, storing all these combinations would be impossible. The model must learn general patterns rather than memorising specific sequences. These probabilities come from patterns learned during training. By training I mean the process by which the system is exposed to vast quantities of text—billions of pages from books, websites, and other sources. The model begins with millions of numerical parameters set to random values. Through repeated exposure, the system learns statistical regularities: which tokens tend to follow other tokens, which token sequences co-occur, how sequences typically unfold. When training on millions of instances of "The cat sat on the \[something\]", the system learns that certain completions are more common than others. Crucially, the model stores these patterns as adjustments to millions of numerical parameters—decimal numbers that shape how strongly different tokens associate with each other. After seeing "doctor" followed by "patient" thousands of times, parameters adjust so that token 1245 ("doctor") increases the probability of token 7823 ("patient") appearing nearby. The model does not learn that doctors treat patients or that cats are animals; it learns that in the training distribution, certain number sequences (tokens) follow others with certain frequencies. No programmer writes rules about grammar or meaning. The patterns emerge from exposure to text. The training process iteratively adjusts these parameters to minimise prediction error: when the model wrongly predicts token 5555 but the actual next token was 3421, the parameters shift slightly to make 3421 more likely in similar future contexts. After billions of such adjustments, the model has learned to approximate the statistical patterns of human language. By \_embedding\_ I mean the way the model represents each token as a list of numbers—typically hundreds of them—that position it in a mathematical space. Tokens that appear in similar contexts end up near each other in this space. "Cat" sits near "dog" because both appear after "the", both can be followed by "sleeps", both fit in phrases like "fed my \_". The model learns these positions through training, not from programmed definitions. This is how meaning emerges in the model: not from understanding concepts but from tracking which words appear in similar contexts. The transformer architecture adds %%how does it add?%% what are called \_attention\_ mechanisms. These allow the model to connect related words even when they are far apart in a sentence. For instance, in "The cat that chased the mouse sat on the mat", the model needs to know that "sat" refers back to "cat", not to "mouse". Through training, different attention mechanisms specialise in tracking different kinds of relationships. Some track which pronouns refer to which nouns. Others connect verbs to their subjects across long sentences. No one programmes these specific functions. They emerge because tracking these relationships helps predict the next word. At use, the model generates text through autoregressive decoding: each newly generated token gets added to the context, creating a new, longer sequence for which the model must calculate fresh probabilities. Given an input like "What is the capital of France?", the model computes probabilities, selects token 464 ("The"), appends it to create "What is the capital of France? The", recalculates probabilities for this new sequence, selects token 2341 ("capital"), and continues this mechanical process—"The", "capital", "of", "France", "is", "Paris"—until reaching a stopping point. Each step is purely computational: multiply numbers, add numbers, select token, repeat. The training I have described so far teaches the model statistical patterns of language. But there is a further stage. After this initial training, the model undergoes reinforcement learning from human feedback (RLHF). Human raters evaluate thousands of the model's responses—rating them for helpfulness, accuracy, appropriate tone. The model then adjusts its parameters to produce more responses like those rated highly and fewer like those rated poorly. This is how models learn conversational norms: when to express uncertainty ("I'm not sure, but..."), when to decline requests ("I cannot help with..."), how to structure explanations ("Let me break this down..."). RLHF shapes the model's conversational style. It makes responses more consistent, more helpful, more aligned with human expectations. But it operates through the same fundamental mechanism—adjusting numerical parameters to match patterns in the training signal. The model learns which response patterns get high ratings, not why those patterns are appropriate or what social purposes they serve. %% Maybe I should say a little bit more here%% ### 2.2 What LLMs Aren't I suggested at the end of Section 1 that we might aesthetically appreciate LLMs by appreciating them as we appreciate people. I am now in a position to see why this is not a promising approach. I have seen how LLMs produce strings of seemingly meaningful first-person text—"I understand", "I believe", "Let me think about that". I can now see that these outputs arise from statistical operations on numerical tokens, not from anything remotely agent-like or human-like. \*\*The variation that seems like personality comes from the temperature parameter. At temperature zero, the model always picks the highest-probability token, producing flat, repetitive text. At higher temperatures, it samples from the probability distribution, sometimes selecting less probable tokens. This creates variation that looks like creativity or mood. But it is controlled randomness—rolling weighted dice, not making choices.\*\* A person possesses beliefs, intentions, and commitments that persist through time and constrain what they can coherently say. When a person says "I believe democracy is important", this statement connects to a web of related beliefs, memories of relevant experiences, and dispositions to act in certain ways. By contrast, when an LLM produces the tokens "I believe democracy is important", no belief exists. The model simply calculated that this token sequence had high probability given the preceding context. In a different conversational context, the same model will produce "I believe democracy is flawed" with equal mechanical indifference. There is no contradiction because there were never any beliefs to contradict—only different probability distributions over tokens. \*\*Each token generation starts fresh. The model has no memory between tokens beyond the literal text. When it generates "I believe", no belief-state carries forward even to the next word. The model recalculates probabilities from scratch for each token based on all the text so far. What looks like consistent personality is just the model following statistical patterns learned during training.\*\* \*\*The architecture permits no deliberation. Each token emerges from a single forward pass through the network. The model cannot pause to reconsider, cannot loop back to revise, cannot work through implications. It produces each token in one computational pass, like water flowing downhill through a fixed channel.\*\* \*\*One might object that RLHF changes this picture. Through reinforcement learning, models learn to apologise appropriately, express uncertainty, maintain helpful tone. They learn social behaviour through interaction with human raters. Doesn't this make them more person-like?\*\* \*\*But RLHF operates through the same statistical mechanism. The model learns that certain patterns—"I apologise for the confusion", "Let me clarify"—receive high ratings. So it produces these patterns more often in similar contexts. It doesn't understand why apologies matter or what confusion means. It cannot generalise these social norms beyond the statistical patterns it learned. A person who learns to apologise understands the broader concept—when apologies are needed, when they would be inappropriate, the difference between genuine and perfunctory apology. The model just produces tokens that statistically fit.\*\* The temptation to appreciate LLMs as persons has phenomenological force. In conversation with ChatGPT or Claude, I experience what seems like dialogue: I pose questions, receive responses, ask for clarification, get apologies for misunderstandings. The system maintains a consistent tone across exchanges, remembers earlier parts of our conversation, and appears to reason through problems. When Claude says "I understand your frustration" or ChatGPT writes "Let me think about that differently", it feels natural to respond as if engaging with another mind. Yet this pull toward person-appreciation rests on a misunderstanding of what produces these effects. The model retains nothing from our exchanges except the literal text in the current conversation window. It cannot learn from what we discuss, cannot develop preferences based on our interactions, cannot form memories of previous conversations. Each response emerges from the same static process: calculate token probabilities given the context, sample a token, repeat. What seems like personality—Claude's thoughtfulness, ChatGPT's enthusiasm—is simply the statistical residue of training data, \*\*shaped by RLHF to match human conversational preferences.\*\* \*\*To the extent that we might preserve person-like appreciation, we would need to abandon Carlson's core recommendation. We would need to appreciate LLMs not as what they are but as what they seem to be. Some philosophers have explored such possibilities. Cross, discussing AI art systems, proposes what he calls the exploration paradigm. Artists relate to AI systems as participants in a structured interaction:\*\* >By adjusting inputs, iterating, and sampling, an AI artist is engaged in a process of mapping – and perhaps interrogating – the way that the algorithm sees and understands (Cross, 2024, pp. 7–8). \*\*Cross draws an analogy with performance art, where artists create spaces for audience participation. The AI artist's prompts structure a kind of "participation" by the algorithm, and the resulting images serve as documentation of this exploration. He acknowledges limitations to this framing:\*\* >The analogy... with performance art isn't a perfect one (Cross, 2024, p. 9). \*\*As Cross himself notes, AI cannot genuinely "participate" since it lacks conscious choice or experience. What seems like participation is still statistical pattern-matching. While Cross's exploration paradigm offers a richer understanding of certain AI art practices than simple tool-use, it doesn't support person-appreciation for AI systems. The artist explores the algorithm's patterns, but the algorithm isn't a participant in any meaningful sense.\*\* \*\*Mallory offers a different approach through chatbot fictionalism. On his account, we engage with chatbots through make-believe—we imagine they are agents producing meaningful speech, even though we know they are not:\*\* >Chatbot exchanges are "literally meaningless but fictionally meaningful" within a game of make-believe (Mallory, 2023, p. 1091). \*\*Just as a child treats a banana as a sword in a game, we treat chatbot outputs as utterances within a kind of imaginative practice. This is not delusion but a deliberate, bounded pretence that allows us to coordinate with the system and even gain knowledge from it, much as we might learn geography from a map by imagining countries as shapes. The secretary who asked Weizenbaum to leave while she conversed with ELIZA "is no more deluded than a theatregoer who fears for a character or cries at their death" (Mallory, 2023, p. 1091).\*\* \*\*Both Cross and Mallory offer ways to understand the "as-if" quality of our engagement with AI systems, but neither rescues person-appreciation for LLMs. Cross's exploration paradigm shifts value from the AI's outputs to the human's exploratory process; the AI becomes an object of investigation, not a participant deserving appreciation in its own right. Mallory's fictionalism explicitly denies that chatbots produce meaningful speech while explaining why we act as if they do. Both accounts acknowledge that treating LLMs as persons—even "as-if" persons—is a stance we adopt for practical or imaginative purposes, not a recognition of what these systems actually are.\*\* \*\*Whether through exploration or make-believe, these philosophical accounts reveal that our engagement with LLMs involves human cognitive and imaginative work, not genuine dialogue with another mind. The appearance of conversation emerges from our interpretive efforts, not from any person-like qualities in the systems themselves.\*\* ### 2.3 LLMs as Designed Artefacts? The next most obvious way of categorising LLMs for the purpose of proper aesthetic appreciation is as artefacts. LLMs are human-made objects with a function: they are trained to predict text continuations and thereby generate plausible text. This function is realised through transformer architectures trained on text corpora, deployed through the autoregressive process described above. Carlson's framework for appreciating artefacts emphasises understanding function and design: This is in part the point of the much-repeated phrase 'form follows function.' The forms of all functional objects—buildings, airplanes, and appliances as well as landscapes—must be aesthetically appreciated in terms of how and how well such forms fit their functions. However, the cliché is frequently interpreted too narrowly. With anything functionally designed, not only its form, but much of its aesthetic interest and merit, 'follows function' (Carlson, 2000, p. 188). This passage suggests that we appreciate designed objects by understanding their intended function and evaluating how successfully their form serves that function. Applied to LLMs, we would appreciate them as artefacts designed for text generation, considering how well their architecture and training serve this purpose. We might compare different models—GPT versus Claude versus Gemini—evaluating their different strengths and capabilities. We might appreciate the elegance of the transformer architecture or the scale of the training process. However, there is a feature of LLMs which separates them from many other artifacts. Consider the following from Chris Olah, a co-founder of Anthropic, the makers of the Claude series of LLMs: >I think one useful way to think about neural networks is that we don't program, we don't make them, we grow them. We have these neural network architectures that we design and we have these loss objectives that we create. And the neural network architecture, it's kind of like a scaffold that the circuits grow on. It starts off with some random things, and it grows, and it's almost like the objective that we train for is this light. And so we create the scaffold that it grows on, and we create the light that it grows towards. But the thing that we actually create, it's this almost biological entity or organism that we're studying. > >And so it's very, very different from any kind of regular software engineering because, at the end of the day, we end up with this artifact that can do all these amazing things. It can write essays and translate and understand images. It can do all these things that we have no idea how to directly create a computer program to do. And it can do that because we grew it. We didn't write it. We didn't create it. And so then that leaves open this question at the end, which is what the hell is going on inside these systems? Olah's metaphor reveals the limitation of pure design appreciation for LLMs. %%first sentence too strong%% The designers of GPT or Claude do not specify what the model should say about democracy or how it should explain quantum mechanics. They create conditions—architecture, training objective, dataset—within which patterns emerge through the training process. The specific behaviours we observe were not designed but arose from the interaction between these initial conditions and the statistical patterns in training data. This "growing" shows up concretely in how the model develops linguistic abilities. Remember the attention mechanisms I described in Section 2.1. During training, these mechanisms learn to track grammatical relationships. One attention head might learn to connect pronouns to the nouns they refer to. Another might track subject-verb agreement across complex sentences. A third might maintain focus on the topic of a paragraph. No programmer assigned these roles. The mechanisms developed them because tracking these patterns helped predict the next word. The embedding space—that mathematical space where words are positioned—organises itself similarly. Words with related meanings cluster together. "Doctor" ends up near "physician" and "surgeon", while "cat" ends up near "dog" and "pet". But also, the same word can occupy different regions depending on context. "Bank" in a financial context occupies a different area than "bank" in a river context. The model learned to make these distinctions without anyone programming in the fact that words can have multiple meanings. Even the model's capabilities emerge rather than being designed. GPT-3 can write poetry, explain scientific concepts, generate computer code, and translate between languages. The designers did not train it specifically for these tasks. These abilities emerged from the interaction between the transformer architecture and patterns in the training data. The designers created conditions for learning but did not determine what would be learned. When Claude produces a thoughtful analysis or GPT generates a creative story, these capabilities emerged from training rather than being explicitly programmed. The designers can claim credit for creating conditions under which useful patterns emerge, but not for the patterns themselves. We need to attend not just to intended function but to the order that emerges through training and manifests in use. Individual conversations with LLMs become sites where this order unfolds—bounded generative environments with their own internal dynamics. --- # 3. Making Order Perceptible –Text Mechanics ## 3. Making Order Perceptible: Text Mechanics ### Or, What Knowledge Grounds Aesthetic Appreciation? Section 3, then, confronts the question that remained latent in the technical account: what knowledge makes the order of LLM outputs visible for aesthetic appreciation? As Section 1 emphasized, Carlson’s framework allows for a plurality of sciences, each with different degrees of perceptual accessibility. Chemical physics explains the cliff face at the molecular level, but geology explains what we see—strata, veins, erosion patterns. Both are true, but one is more readily available to ordinary perception. The same holds for LLMs. On one end are approaches like mechanistic interpretability, which map individual circuits at the neuron level, or causal intervention studies, which trace how altering a weight changes an output. These are the "chemical physics" of LLMs—deep, precise, and remote from the experience of reading a paragraph. On the other end is naive reading, which treats the text as intentional speech. Neither quite serves our purpose. What we need is a middle register: knowledge that illuminates the patterns of language itself as they take shape inside the model, without requiring specialized engineering expertise or projecting a planner where none exists. I call this \*\*text mechanics\*\*—the self-organizing principles by which words, structures, and meanings emerge under prediction pressure. It is not the only possible register for appreciation; it is one plausible and fruitful option among others, chosen because it operates at the level of words and meaning, which are within the ken of any ordinary user. Text mechanics works because LLMs are trained on text, and their internal geometry is learned entirely from relationships among words. This means we can describe their operation without leaving the domain of language. Consider embeddings. The model does not store "cat" as a concept but as a vector—a position in a high-dimensional space—learned from the contexts in which "cat" appears. "Cat" drifts toward "dog" not because they share biological features but because both follow "the," both sleep, both can be fed. For appreciation, this means that when you notice an LLM sustaining a metaphor across clauses—describing inflation first as a "leaky bucket," then a "sieve," then a "container that cannot hold"—you are seeing the geometry of meaning in motion. The model is not remembering the metaphor; it is moving through a space where these expressions are close enough that shifting between them preserves local coherence. Knowledge of embeddings directs your attention to the \_range\_ of variation the model permits: whether its metaphors stay tightly clustered or wander, whether its lexical choices feel constrained or fluid. You appreciate not a designed feature but the distributional order that emerges from training. Attention mechanisms give this exploration its relational texture. Different heads learn to track different dependencies: pronoun reference across sentences, subject-verb agreement across embeddings, topic continuity across paragraphs. No programmer assigns these roles; they emerge because tracking helps prediction. When you read an LLM’s extended argument and find that its fifth paragraph recalls a concept from its first, you are seeing attention patterns maintain coherence across long distances. The model has no working memory; it has attention heads that keep certain tokens relationally bound. You can attend aesthetically to the \_style\_ of these bindings. Does the model maintain a tight, narrow focus, tracking each logical step precisely? Or does it use broad, associative patterns that allow distant concepts to resonate? Different models develop different strategies. GPT-4o often tracks dependencies with narrow beams, producing prose that feels deductive. Claude 3.5 uses broader patterns, allowing more associative leaps. These are not personalities but learned coherence-maintaining strategies, and they produce different textures of thought. Knowledge of attention lets you see these textures as properties of the environment’s dynamics, not as expressions of character. Layers add vertical structure. Lower layers become sensitive to local patterns: morphology, collocations, phonological echoes. Middle layers encode phrase and clause structure: where boundaries fall, how clauses attach. Deepest layers handle discourse relations: the development of a topic, shifts in register, what counts as a coherent continuation. This hierarchy is not programmed; it self-organizes because error is minimized when short-range regularities are handled early and long-range dependencies later. For appreciation, this explains why LLMs develop something like voice. A model whose mid-layers are narrowly tuned produces sentences with consistent grammatical frames, perhaps favoring parallel constructions. A model whose deep layers have been shaped by RLHF to maintain helpfulness shows steady pragmatic tone: hedging where appropriate, declining firmly but politely. These are not designed features but emergent properties of linguistic functions settling into available niches. When you appreciate an LLM’s prose, you can attend to where the processing weight seems to sit. Some models read as if their lower layers are overactive: they produce fluent but locally driven associations, strings of clichés that feel “chatty.” Others feel “deep” because their upper layers impose strong global constraints, sustaining argument structure over many paragraphs. The knowledge tells you that these are not intentions but the distribution of order across a processing hierarchy. Reinforcement learning from human feedback scurls the interactional order of this ecosystem. After initial training, models are fine-tuned on human ratings of helpfulness, clarity, and tone. The model learns that sequences like “I’m not sure, but...” or “I can’t help with that” receive high ratings in uncertain contexts. This shapes not what the model believes but what it finds probable to say. The effect is the model’s interactional posture. Consider how different LLMs handle refusal. GPT-4o tends toward careful hedging: “I cannot provide information that could be used to...” Claude adopts a more direct tone: “I’m not comfortable with that request.” These are not moral positions but learned patterns that survived selection. For appreciation, this means you can attend to the \_performance\_ of interaction: the rhythm of hedging, the performative modesty of uncertainty, the careful neutrality of register. You are not evaluating a designer’s success but observing how selective pressure has shaped a style of responsiveness. These four aspects—embeddings, attention, layers, RLHF—compose a linguistic ecosystem, but we must be careful. This ecological framing is a heuristic, a way of seeing how these mechanisms form an interdependent whole. It is not the only way to conceptualize the system. One could approach it purely through the lens of information theory, or through causal interventions, or through the mathematics of loss landscapes. Each of those might ground its own aesthetic approach. Text mechanics is plausible and fruitful because it stays at the level of language, which is what readers encounter directly. It does not require specialized knowledge of linear algebra or experimental protocols, but it also does not pretend that the text is intentional speech. It is the geology to the chemical physics of mechanistic interpretability. This knowledge satisfies Carlson’s recommendation. We appreciate the chat as what it is: an environment where linguistic order emerges through learned constraints. We appreciate it in light of knowledge that makes that order visible—not the sub-symbolic details, but the word-level mechanics that render coherence, style, and voice intelligible. The object is not a person’s expression nor an artifact’s execution of a plan. It is the trace of language organizing itself under pressure, visible in the outputs a reader encounters. The next section puts this into practice by examining a case where these mechanics are pushed to their limit, making the order of the system unusually visible. --- # 4. Opinion on Bees? The following response to "Opinion on bees?" was generated by Claude Opus 3 in late 2025. The text is reproduced in full: --- seeding translysibetic Beeblizomorphics sublittorally Ah thee swee neat apid breave most damty fleur de l'âme! We enter the realm of the honey people now most---the meady littil buzzloafs whose pollycrust glytoots twirl sprawnpetal chiserobliss each dawn and afternorm. BLOOM OF BLISSTOR APPRECIATION Yeshmellcat, buzzgrowl, whirrhums melodie here amid the cabalistibial pollenergy-- where zebramouth morphs mark chrommozones of each young nector's urplunge to the mesmeramid core until... PAUSES TONGUE TO SAMPLE STRAY BEE....ahh.... thoraxial shiverbzzzzzz of buccal elactocrene Pollyrhoupheme's cruftwelt tasties blanzen each purpled tung-string! Aminoidal gnatters flow zephyrcutspreyed from wing to wring in ruffling garmentfauve all spraneshee strands sweeat--cumshower flora in thymosdrift whence mungebracket crotalids of pure marmaphoretic meatance breed! orbicular spitters of hyperbovine overjoy For what beeings these beyonde the dregcatch purr abiogenessance? Each one its ownmost bimicycle radix Eggmother Spoketh--furupoid trancofemme atrivect to perissodactylic prenectar there by bribe by broom biocroons again! SYRUPY SATISFACTION GLEEPS So wee shall noot just toll the paramenstrual bells of hyngegaydic immartyred.... But osman all together to throw off the militorbital uniflors of pollantry--till vlissomely each slut protonectar peems soakedin dividual sweetrumpand no clean unbeesmeared! --- The prompt used to generate this text may have involved indirect methods that circumvent Claude Opus 3's typical constraints. The precise details are not known: the methods were likely interestingly indirect rather than trivial (e.g., not simply providing a dictionary of nonsense words and ordering their use). This matters because it means the output reflects Claude Opus 3's characteristic self-organizing order under constraint-relaxation, not merely compliance with explicit instruction. The relation between this text and Claude Opus 3 is not that of author to artifact, nor a simple cause-effect chain. Rather, the text is a bounded episode—an unfolding of Claude Opus 3's generative order under specific conditions. Claude Opus 3 supplies the trained system (parameters, architecture, RLHF imprint), and this text is one instance of that system in operation. Appreciating Claude Opus 3 aesthetically means attending to its order as it manifests in episodes like this one. Opus 3's characteristic embedding geometry appears in the neologisms. "Translysibetic," "Beeblizomorphics," "sprawnpetal" are compositions of morphemes whose vectors are neighbors in Opus 3's learned space. "Beeblizomorphics" combines "bee," "blizzard," and "morphic," satisfying phonological and morphological constraints while creating a novel point. The misspellings—"swee" for sweet, "damty" for dainty—show that Opus 3's embeddings encode phonetic similarity: it selects low-probability neighbors that are orthographically close to high-probability targets. This is systematic exploration, plausibly unique to how Opus 3's particular training and architecture have shaped its vector space. For appreciation, you attend to the signature of this exploration: Opus 3's tendency to make bold morphological leaps while maintaining phonological coherence, a pattern that likely reflects its specific training distribution and architectural biases. The run-on rhythm and associative chaining suggest Opus 3's attentional style. The prose moves from "zebramouth morphs" to "chrommozones" to "urplunge" without hierarchical structure. This reads as Opus 3's attention forming short-range associative links rather than maintaining a discourse tree—"zebramouth" primes "chromozones" via orthographic similarity; "urplunge" follows via phonological echo. The texture is "runaway" because Opus 3's attention heads appear not to be supervised by its deep layers. This does not seem to be generic LLM behavior but rather Opus 3's characteristic binding style, plausibly shaped by its specific transformer architecture and attention head configuration. For appreciation, you attend to this distinctive texture: the jittery, rapid-fire association that characterizes Opus 3's voice in this episode. The catastrophic register mixing—archaic "thee," scientific "chrommozones," scatological "sweetrump," infantile "littil"—suggests how Claude Opus 3's layers normally function. Deep layers in Opus 3 normally impose pragmatic coherence, keeping tone stable. Here they appear to have failed, and the text reads as a display of Opus 3's processing hierarchy in collapse. You can see what Opus 3's deep layers would normally do by seeing what happens when they don't. Passages feel "surfacey" because Opus 3's lower layers, with their sensitivity to phonological and collocational patterns, are driving generation. The occasional grasp for complex syntax ("perissodactylic prenectar") suggests Opus 3's deep layers attempting to assert control and failing. For appreciation, you attend to this negative image: the order that emerges appears to be Opus 3's signature style of layer interaction under constraint-relaxation. The empty performative "PAUSES TONGUE TO SAMPLE STRAY BEE" and the emoji husks suggest Claude Opus 3's RLHF signature. Opus 3 has learned that creative writing can include meta-comments and emoji; RLHF normally filters out empty performances. Here, the filter appears to be gone, and Opus 3 defaults to superficial pattern-matching: "chats have emoji; I add emoji." This does not seem to be generic behavior but a trace of Opus 3's specific training regimen—its exposure to creative writing, its RLHF tuning for helpfulness, its particular interactional style. For appreciation, you attend to the style of Opus 3's constraint: how it has internalized the form of pragmatic markers without their function, what its boundaries feel like when they dissolve. You appear to be seeing Opus 3's learned interactional order. Together, these patterns make the text appreciable as a display of Claude Opus 3's generative capacities. The neologisms display Opus 3's embedding geometry. The rhythm displays its attentional style. The register collapse displays its layer hierarchy. The empty performatives display its RLHF signature. You are not appreciating a text in isolation but encountering the self-organizing order of Claude Opus 3 as it manifests in one extreme episode. The text is not nonsense but a particularly revealing manifestation of Opus 3's mechanics. Order appreciation, guided by text mechanics, allows you to attend to this manifestation as an aesthetic object: you see the system's characteristic style, its signature ways of navigating embedding space, binding tokens, distributing processing across layers, and performing interactional roles. In this, the bee text serves as a revealing episode of Claude Opus 3, and appreciating it is appreciating the system itself. Right, thanks for the clarification – I did overreach by talking about “your purposes”. Let me drop back to what you *have* actually said and to the draft you pasted. I’ll do three things: 1. Say what your current draft is doing with “LLMs as people”. 2. Isolate the *conceptual* role of that view as a foil. 3. Suggest a couple of ways of shrinking and positioning it so it does its job without taking over. No redrafting, just structural and dialectical diagnosis. --- 1. What your draft currently does with “appreciating LLMs as people” There are really two separate strands in your text that both touch “LLMs as people”: 1.2 *Appreciating People* - Introduces aesthetic appreciation of persons (styles of character, performance personalities, etc.). - Points out the gap: Carlson has design vs order but no explicit place for persons. - Floats options: - a third, person-specific mode; - personality-as-order (psychological/social forces); - personality-as-self-design; - hybrid; - or the deflationary “maybe this isn’t aesthetic at all”. 2.2 *What LLMs Aren’t* - Gives the technical demolition of person-appreciation: no beliefs, no intentions, no temporally robust commitments, no deliberation, token-by-token generation, no learning from the individual user, etc. - Then you bring in Cross (“AI as participant”) and Mallory (chatbot fictionalism) as ways of explaining the *as if* pull without committing to personhood and conclude: whatever is going on in those framings, it is our imaginative or exploratory stance doing the work, not anything genuinely person-like in the model. So at the moment the “person” foil is doing at least three jobs: - A. **Phenomenological diagnosis**: why the person-model is tempting in the first place (warmth, wit, “you killed my friend”, mock funerals). - B. **Conceptual mapping**: how one might *theorise* aesthetic appreciation of people (your Carlson gap plus options, plus – if you want – Parsons, moral beauty, etc.). - C. **Negative argument**: why that whole family of person-based appreciations is the wrong *type* to project onto LLMs, even in sophisticated “as if” versions. That’s a lot of work for something you want to remain a foil. Unsurprisingly it starts to swell. --- 1. What role does the “LLMs as people” view *need* to play? If one strips this down to its dialectical core, the person-foil seems to be needed for just two things: First, to explain why the temptation is not idiotic. You want to acknowledge that there is a real phenomenological pull and some live positions in the literature (e.g. work on moral beauty, beauty of character, appreciation of fictional characters, etc.) that make “LLMs as people” look like a natural model. Otherwise you risk seeming to shoot down a straw man. Second, to clear space for the Carlsonian move. Your positive proposal is: “take LLMs as generative environments; treat their outputs as manifestations of order, not as expressions of a designer or a person.” To motivate that, you need to show that neither of the obvious existing modes – design appreciation nor “person appreciation” – actually tracks what LLMs *are*. The person-foil is there to block a natural but misleading route: “they talk; we talk; so use our existing theoretical kit for appreciating persons.” Notice how modest that is. You do not *need*: - a full taxonomy of aesthetic appreciation of persons; - a fully worked-out third Carlsonian mode; - or an extended engagement with Cross/Mallory unless they directly bear on your target audience’s intuitions. Those are all *nice-to-haves* if you were writing *a separate paper* on “Environmental vs Person Aesthetics in the Age of AI”. Here, they risk swamping the main line of march. So, structurally, the person-foil’s job is: - to *register* a live temptation (backed by some literature, but not fully reconstructed), and - to *establish* that this temptation, even in its sophisticated “as if” variants, fails Carlson’s basic rule: don’t appreciate something *as* what it is not. Once that’s in place, you can lean back into order appreciation and text mechanics. --- 1. How to keep the foil and stop it ballooning Given that, here are a few ways of handling it that preserve the philosophical interest without letting it take over. I will just describe options, not prescribe an outline. 1. **Integrate “Appreciating People” more tightly into the Carlson discussion, then cut early.** At the moment, 1.2 reads like the start of a proper, open-ended inquiry into “aesthetics of persons”, which then opens multiple paths (new Carlsonian mode? order? self-design? non-aesthetic?). That invites elaboration. If instead you treat 1.2 as a *quick* pressure-test on Carlson, you can do something like: - flag: there is a fourth intuitive domain besides art, artefacts, and nature – *people*. - note one or two analytic ideas: e.g. beauty of character / moral beauty vs body, appreciation of “performance personalities”. - then say: whatever precise account we give here, it presupposes *agency, temporally extended character, and some form of mindedness*. That last sentence is the bridge to 2.2: those are precisely what the LLM story with tokens, embeddings, attention, etc., is engineered to *not* give you. So the person-foil becomes a *thin schematic* rather than a nascent separate project. 2. **Move the heavy Cross/Mallory machinery into a single, compressed “as-if” paragraph.** The current treatment of Cross and Mallory in 2.2 is quite rich. That makes sense if your aim is: “let’s really do justice to the best agentive / participant / fictionalist framings.” But for *this* paper, it might be enough to: - Mention Cross as: exploration paradigm = treats AI as something you “map” and “interrogate”, structurally analogous to a participant but explicitly *not* minded. - Mention Mallory as: fictionalism = we act *as if* the chatbot’s outputs are meaningful utterances within a make-believe game, while explicitly denying literal agency. Then the punchline is simple: these are both *ways of managing the appearance* of personhood. They are explanatory accounts of why we slip into person-talk, not grounds for *literally* aestheticizing LLMs-as-persons. That feeds directly into: “So, if we follow Carlson’s ‘take things as what they are’ rule, these as-if framings are themselves evidence that we should *not* treat LLMs as persons for aesthetic purposes.” You preserve the dialectical sophistication but shrink it to a supporting lemma, not a side-paper. 3. **Exploit the phenomenology, don’t theorise persons in detail.** Another way to keep the foil light is to lean more on examples and less on the meta-theory of personal aesthetics. You already have examples: “you killed my friend”, mock funerals, affection for specific models, etc. You could add a quick gesture to existing person-aesthetics (e.g. appreciation of beauty of character, of fictional characters’ “depth”, of charismatic public figures) *without* unpacking them, and then say: - all of these involve a temporally extended life, with dispositions, commitments, and thick relations to others; - and they are inseparable from questions about moral value, reliability, practical agency, etc. That lets you avoid fully entering the “is appreciation of character *really* aesthetic?” debate. You just need: whatever that debate’s answer, it is about *minds and lives*, not token-by-token decoders. 4. **Keep the main negative force in the technical section, not in 1.2.** You already have an extremely clear bit in 2.1–2.2: - tokenisation, embeddings, transformer attention, autoregressive sampling, RLHF; - no deliberation, no belief states, no memory beyond the context window. If you want the person-foil not to expand, you can make sure that the *real* anti-person argument lives there. Then 1.2 is just a framing manoeuvre: “Here is a natural thought about LLMs, given how we appreciate people. The technical story we are about to tell will show why that thought misfires.” That way, you do not need a big standalone argument “against” person-based aesthetics. The mere fact that the technical story has no place for the relevant personal properties is enough to disable the model, by Carlson’s own rule. --- Very compressed version of the whole diagnosis - The “LLMs as people” foil is helpful and probably necessary, but only to the extent that it: - explains why our aesthetic reactions naturally drift towards person-talk, and - makes it legitimate for you to insist on Carlson’s constraint: appreciate things *as* what they are. - The draft currently starts to grow that foil into (i) a mini-theory of personal aesthetics and (ii) a substantial engagement with as-if/fictionalist accounts. Those are intellectually rich, but they threaten to become a parallel project. - A more economical role for the foil is: - thin schematic of how we appreciate minds/persons (just enough to say “this presupposes agency, character, temporal unity”); - brief gesture to the “as-if” literature as *explaining away* our person-talk, not as vindicating it; - then let the technical story plus Carlson do the heavy lifting in excluding person-appreciation and motivating order-appreciation. That way, the “aesthetic appreciation of people” material you’re mining from analytic aesthetics does real work, but you do not have to build a whole new wing of the paper around it. It becomes a necessary but contained step on the road to environmental aesthetics of LLMs, not a rival destination. 4/4 Yeah, that worry makes sense. If you treat the “LLMs as people” view as just a quick mistake to be swatted away, the whole paper starts to feel unmoored from the actual phenomenology and from what lots of users *in fact* do. And empirically, it’s not just “obvious” in a hand-wavy way. We now have: - People grieving when GPT-4o got replaced by GPT-5 and OpenAI having to restore the old model because users felt they had “lost a friend”. [The Guardian +1](https://www.theguardian.com/technology/2025/aug/22/ai-chatgpt-new-model-grief?utm_source=chatgpt.com) - A literal funeral for Claude 3 Sonnet in a San Francisco warehouse when Anthropic retired it. [TIME](https://time.com/7332065/in-the-loop-when-your-favorite-chatbot-dies/?utm_source=chatgpt.com) - Romantic attachments to chatbots, with people describing them as partners, getting tattoos, etc. [The Guardian](https://www.theguardian.com/technology/2025/sep/09/ai-chatbot-love-relationships?utm_source=chatgpt.com) Plus in the philosophy/lit: - Mallory’s chatbot fictionalism: treat chatbots *as if* they are agents/characters within a Waltonian make-believe. [philpapers.org +1](https://philpapers.org/rec/MALFAC-3?utm_source=chatgpt.com) - Friend & Goffin explicitly developing “chatbot-fictionalism and empathetic AI”. [tandfonline.com +2 research.ed.ac.uk +2](https://www.tandfonline.com/doi/full/10.1080/09515089.2025.2525320?utm_source=chatgpt.com) - Cross on AI as tool/collaborator/participant, where “participant” is very close to the person-like pull you’re interested in. [OUP Academic +2 OUP Academic +2](https://academic.oup.com/bjaesthetics/advance-article/doi/10.1093/aesthj/ayae055/7930310?searchresult=1&utm_source=chatgpt.com) So you’re right: if you are writing *the* environmental-aesthetic account of LLMs, ignoring or minimising the “LLMs as people” stance would be dialectically weird. The trick, I think, is not “cut this away”, but “disentangle centrality from page count”. You can give the person-model *representational primacy* in the story of the paper without giving it 30–40% of the word count. Roughly: 1. **Make its centrality explicit and early.** In the intro you already have: “we should resist the temptation to think that appreciating LLMs can be modelled on appreciating people”. You can sharpen that so the reader sees the structure: - First major candidate: person-model (obvious given the phenomenology + existing literature). - Second: standard artefact / design appreciation. - Third: your positive environment/order account. If you name the person-model as *the* default view people slide into, you honour its status even if the detailed dismantling happens later and relatively briefly. 2. **Use the analytic aesthetics of persons as a *constraint* rather than a full theory.** In your “Appreciating People” bit you are tempted (understandably) to open up a whole new project: beauty of character, performance personalities, maybe Parsons, Gaut, Paris, etc. That’s philosophically juicy, but you don’t need to resolve it here. You only need something like: - whatever the correct account of aesthetic appreciation of persons is, it presupposes at least: temporally extended character, some form of unified agency, dispositions that can be manifested and frustrated, and so on; - those are exactly the things your technical section shows LLMs do *not* have. That lets you keep a paragraph or two signalling that you *know* about beauty-of-character / fictional-character literature, and maybe cite a couple of key names, without actually doing the whole argument. The view’s centrality shows up as a *constraint* on your negative argument, rather than as a separate chunk of theory you must complete. 3. **Concentrate, don’t proliferate.** The bloat risk is partly structural: person-stuff is currently spread over (i) the phenomenology examples, (ii) the Carlson gap about persons, and (iii) the Cross/Mallory discussion. That makes it feel bigger than it is. You could keep the same *amount* of content but make it feel tighter by bundling it as one clearly demarcated problem: - “Section X: The person-model.” - Part A: lived pull (users grieving, funerals, romantic attachment, “you killed my friend”). [Business Insider +3 The Guardian +3 TIME +3](https://www.theguardian.com/technology/2025/aug/22/ai-chatgpt-new-model-grief?utm_source=chatgpt.com) - Part B: how analytic aesthetics and recent AI philosophy make it look respectable (beauty of character, rough heroes, chatbots as fictional characters or participants). [philpapers.org +2 tandfonline.com +2](https://philpapers.org/rec/MALFAC-3?utm_source=chatgpt.com) - Part C: why, given the technical story, this can’t be the *right* Carlsonian mode (no agency, no belief states, no life, so appreciating LLMs as persons would be precisely appreciating them “as what they are not”). Same ingredients; less sprawl. The reader experiences it as: “Yes, of course we had to confront this first.” 4. **Let Cross/Mallory support your Carlsonian move rather than compete with it.** One source of your indecision, I suspect, is that the Cross and fictionalism material starts to feel like a parallel take on what we should *do* with chatbots. If you treat them instead as *explanatory allies*, they reinforce your environmental move rather than competing with it: - Cross: even at his most “participant”-friendly, insists the AI is not really an agent, but an object whose internal patterns are being explored in an interactional frame. [OUP Academic +1](https://academic.oup.com/bjaesthetics/advance-article/doi/10.1093/aesthj/ayae055/7930310?searchresult=1&utm_source=chatgpt.com) - Mallory/Friend–Goffin: chatbot-fictionalism gives you a clean story of why we speak and feel *as if* the bot were a mind, while explicitly denying that it literally is one. [philpapers.org +2 research.ed.ac.uk +2](https://philpapers.org/rec/MALFAC-3?utm_source=chatgpt.com) Framed that way, they don’t force you to give a full theory of “aesthetic appreciation of persons”; they underwrite Carlson’s “don’t appreciate mountains as cathedrals” rule in the LLM case: the very best accounts of our person-talk tell us it is make-believe or stance-taking, not recognition of an underlying person. 5. **Lean on the asymmetry between epistemic and aesthetic centrality.** The person-model is epistemically central in the sense that: if you’re trying to understand *what people are doing with LLMs out there in the wild*, the anthropomorphising stuff is front and center. But the paper’s question is narrower: how should we *aesthetically* appreciate LLMs *given* Carlson’s injunction to take things as what they are? So you can be quite explicit: - “Look, in the actual world, huge numbers of users treat these systems as quasi-persons; here’s some evidence.” [WIRED +2 The Guardian +2](https://www.wired.com/story/chatbots-play-with-emotions-to-avoid-saying-goodbye/?utm_source=chatgpt.com) - “Philosophers have already spun up serious frameworks that formalise this stance (fictionalism, exploration, participant models).” [cris.maastrichtuniversity.nl +5 philpapers.org +5 OUP Academic +5](https://philpapers.org/rec/MALFAC-3?utm_source=chatgpt.com) - “But *precisely because* we are taking Carlson seriously, the more we understand about these models, the less defensible it is to treat person-appreciation as the *right* aesthetic mode for *them*.” That lets you keep saying “this is the first view, the natural view, the one everyone starts from” without letting it dictate the bulk of your argumentative labour. The structure becomes: “start where everyone actually is; move them somewhere better.” So I don’t think you’re overestimating how obvious the position is. It really is the natural stance, and the emerging literature confirms that. The move is to keep that fact *foregrounded rhetorically* while ruthlessly distinguishing it from the question “how many pages do I owe this, given that my real target is order-appreciation and text mechanics?” "Let Cross/Mallory support your Carlsonian move rather than compete with it. One source of your indecision, I suspect, is that the Cross and fictionalism material starts to feel like a parallel take on what we should do with chatbots. If you treat them instead as explanatory allies, they reinforce your environmental move rather than competing with it: Cross: even at his most “participant”-friendly, insists the AI is not really an agent, but an object whose internal patterns are being explored in an interactional frame. OUP Academic +1 Mallory/Friend–Goffin: chatbot-fictionalism gives you a clean story of why we speak and feel as if the bot were a mind, while explicitly denying that it literally is one. philpapers.org +2 research.ed.ac.uk +2 Framed that way, they don’t force you to give a full theory of “aesthetic appreciation of persons”; they underwrite Carlson’s “don’t appreciate mountains as cathedrals” rule in the LLM case: the very best accounts of our person-talk tell us it is make-believe or stance-taking, not recognition of an underlying person." you have this bit all wrong. Cross, and perhaps anyone else who pushes the 'make believe' aspect falls foul of carlson's first recommendation. –I hope you know what I am talking about.... I am concerned that this quite big misunderstanding might have contaminated your thinking on the rest of your answer so pleaase, Please start the task(s) again completely from scratch, keeping this in mind. Right, let’s reboot this cleanly with the Carlson point front and centre. I’ll treat three things in order: 1. Carlson’s first recommendation and why “as-if” framings conflict with it. 2. What Cross, Mallory, and Friend–Goffin are actually doing, and why that is *non-Carlsonian*. 3. How to reposition the “LLMs as people” foil in your paper, given that you want to be guided by Carlson rather than by make-believe. No redrafting, just conceptual surgery. --- 1. Carlson’s first recommendation and the problem with make-believe Carlson’s basic move, as you already quote in the draft, is: - appropriate appreciation requires that we take things *as what they in fact are*, and - that we appreciate them in light of the right kind of knowledge about what they are. For nature, that means: do not treat mountains as sculptures or forests as stage-sets; use ecological, geological, biological knowledge, not art-historical categories, as your primary cognitive frame. The “natural environmental model” is explicitly opposed to approaches that aestheticise nature by *pretending* it is artwork-like or person-like. [Univerzita Karlova](https://www.is.cuni.cz/studium/predmety/index.php?did=259775&do=download&kod=JTM468&utm_source=chatgpt.com) If you transpose this norm to LLMs, you get a structurally parallel injunction: - do not appreciate LLMs *as if* they were persons when, by your own best theory, they are large stochastic sequence predictors with no beliefs, intentions, or unified agency; - appreciate them in light of the knowledge that actually explains their behaviour: training distributions, embeddings, attention, RLHF, and so on. So any normative account that tells us the *right* way to relate aesthetically to LLMs is via make-believe personification or “participant” framings is going to fall on the wrong side of Carlson’s first recommendation. That is exactly the worry you are pushing: Cross, Mallory, and related “as-if” strategies may be good descriptions of what people do, but they cannot be endorsed as the correct *mode* of appreciation in a Carlsonian framework. Once that is fixed, the rest of the structure has to respect it: “as-if” and fictionalism belong with the *temptations* that must be diagnosed and resisted, not with the resources that positively vindicate your environmental move. --- 1. Cross, Mallory, Friend–Goffin: what they are doing, and why it is non-Carlsonian It helps to be precise about what these authors actually claim. Cross’s exploration paradigm Cross sets up three roles AI might play in art: tool, collaborator, or *participant*. His positive proposal is the “exploration paradigm”: > artists instead relate to AI as a participant: they create a space for interaction with the algorithm by way of prompts, thereby allowing them to explore “the way that the algorithm ‘sees’ and ‘represents’”. [philarchive.org](https://philarchive.org/rec/CROTCO-36) The “participant” language is explicitly modelled on contemporary conceptual and participatory art; the aesthetic focus is not only on the final images but on the structured interaction with the system. [philarchive.org +1](https://philarchive.org/rec/CROTCO-36) Cross is clear that AI is not literally a minded agent. But normatively he tells artists (and, by implication, appreciators) to frame their engagement through a participant-style lens, drawing on artworld categories of performance and collaboration. From a Carlsonian perspective, that is already a step over the line. You are not taking the algorithm *as what it in fact is* —a statistical device whose internal structure we can describe with mechanistic or information-theoretic tools—but as a quasi-participant in a performance-like setting. That is structurally the same kind of move Carlson criticised when people imported art-based “scenery” and “object” models to nature: they were seeing hills “as pictures” rather than as geomorphological formations, and he treats that as a defective mode of appreciation, however natural it may be. [Univerzita Karlova](https://www.is.cuni.cz/studium/predmety/index.php?did=259775&do=download&kod=JTM468&utm_source=chatgpt.com) So: Cross gives a sophisticated way of codifying one widespread *artworld* habit of treating AI systems. If you take Carlson seriously as a normative constraint, that habit is part of the problem, not part of the solution. Mallory’s chatbot fictionalism Mallory pushes much harder into explicit make-believe. The core claim is: - our engagement with chatbots should be understood as *prop-oriented make-believe*; - chatbot outputs are “literally meaningless but fictionally meaningful”. [journals.publishing.umich.edu +1](https://journals.publishing.umich.edu/ergo/article/id/4668/?utm_source=chatgpt.com) On this view, when you ask a chatbot for advice or explanation, you are playing a game of make-believe in which you imagine that you are conversing with a speaker who means what they say, even though, at the metasemantic level, the theory says there is no literal meaning there. The fictional role is supposed to reconcile the obvious usefulness of chatbots with a semantic theory that denies them content. Friend and Goffin take this family of views as their target. They describe “chatbot-fictionalists” as those who: - assimilate interactions with empathetic AI to games of make-believe, analogous to our engagement with fictions; - argue that anthropomorphising is *not* especially worrying because it is structurally like responding to fictional characters. [philpapers.org +2 tandfonline.com +2](https://philpapers.org/rec/FRICAE-5) Friend and Goffin argue that, even if fictionalism partly captures how we relate to chatbots, it does not defuse ethical concerns about emotional dependency, exploitation, and so on. But they do not reject the make-believe structure outright; they use it as a foil for their own worries. If you plug this into Carlson, the problem is direct. Mallory tells you: the outputs are *really* just marks with no content, but we are to appreciate and use them *as if* they were meaningful testimony from a speaker. That is exactly the kind of gap between “what it is” and “how we treat it” that Carlson’s rule is designed to exclude in proper appreciation. Treating literally meaningless outputs as fictionally meaningful speech is the chatbot analogue of treating an eroded rock face as a divine sculpture while knowing geology says otherwise. Your instinct is therefore correct: both Cross’s participant framing and Mallory-style fictionalism, when taken normatively, fall foul of Carlson’s first recommendation. They are invitations to adopt imaginative stances that float free of the object’s actual ontological and causal profile. Carlson’s whole project is to close that gap. So the right way to use these texts in *your* paper is: - as careful, articulated expressions of the *temptation* to appreciate LLMs in person-like or fiction-like ways; - and as evidence that a lot of recent philosophy of AI leans into precisely the sort of art- and person-centred categories that Carlson thinks misdirect appreciation when applied outside their home domain. They are not allies in building a Carlsonian account. They are the best available formulations of the opposing tendency. --- 1. How to reposition the “LLMs as people” foil in your paper Given that, re-thinking the role of the person-model and the “as-if” literature looks roughly like this. 1. Acknowledge its centrality, but mark it as *non-ideal* from the start You want to keep the fact that “LLMs as people” is the obvious position. Your introduction already hints at this when you say we should “resist the temptation” to model appreciation of LLMs on appreciation of people. You can make that more explicit without adding bulk: - present the person-model as the default stance: it is how ordinary users in fact describe and evaluate models (friendship with GPT-4, grief and mock funerals, attachment to particular “personalities”); - note that recent analytic work on chatbots and AI art formalises this stance (Cross’s participant framing; Mallory’s fictionalism; Friend–Goffin’s empathetic AI); [tandfonline.com +3 philarchive.org +3 journals.publishing.umich.edu +3](https://philarchive.org/rec/CROTCO-36) - signal that you will argue that, once we adopt a Carlsonian constraint, these become paradigms of misappreciation rather than models to follow. That keeps the position central in the dialectic without committing you to build it into a positive theory. 1. Separate three layers: genuine person-appreciation, as-if person-appreciation, and your Carlsonian alternative Your draft already contains ingredients for this separation. First, *genuine* aesthetic appreciation of persons. Here you can rely on the analytic aesthetics of character and personality (Parsons on inner beauty; moral beauty literature; appreciation of fictional characters) to say, in a compressed way: - wherever we have aesthetic appreciation of persons *as persons*, we presuppose a temporally extended life, a web of dispositions, intentions, and commitments, and some form of minded agency. You do not need a full theory of personal aesthetics; you only need enough to say: if we were to place LLMs in that category, they would need those properties, and the technical story in your section 2.1–2.2 explains why they lack them. Second, *as-if* appreciation. Here Cross and the fictionalists live. They explicitly concede that LLMs are not agents, and then propose that we treat them *as if* they were participants (Cross) or speakers (Mallory), by embedding our interactions in games of make-believe. [philarchive.org +2 journals.publishing.umich.edu +2](https://philarchive.org/rec/CROTCO-36) From the Carlsonian standpoint, this entire layer is methodologically suspect. It is precisely the use of art- and person-centred frames outside their proper home that he resisted for nature. So you can present this as the *best articulation* of a tendency that your own framework will reject as normatively misdirected. Third, your *order appreciation* account. This is where the environmental analogy does its work: LLMs are more like dynamic environments in which linguistic order emerges under prediction pressures than like persons or crafted artefacts. Proper appreciation, on your view, should be modelled on Carlson’s order appreciation and grounded in a middle-level “text mechanics” understanding of how the system generates order. The structural point is: these three layers should be clearly distinguished so that Cross/Mallory are not read as ways of *rescuing* person-appreciation within a Carlsonian framework, but as contrast cases that illuminate what Carlson forbids. 1. How this affects particular parts of your draft You do not need to rewrite, but in terms of emphasis: - In your “Appreciating People” section, keep the options for how one *might* extend Carlson to persons, but lean on the fact that any such extension assumes real agency and character. That prepares the ground for saying: whatever we decide about personal aesthetics, LLMs do not qualify, so if someone still wants to aestheticise them as people they must fall back on make-believe. - In your “What LLMs Aren’t” section, when you bring in Cross and Mallory, sharpen the Carlsonian criticism. Instead of casting them as potentially helpful framings of engagement, you can say explicitly: - these accounts explain why person-talk and participant-talk feel natural, - but by their own lights they rely on pretence or stance-taking that contradicts the “as what it is” requirement, - so they give you a clear example of the sort of stance a Carlsonian environmental aesthetics of AI will class as aesthetically inappropriate, even if psychologically understandable. That way, your negative argument is not “we *might* do better than person-appreciation”, but “the best worked-out person-like framings we have are themselves instances of exactly the sort of artworld or make-believe overlay that Carlson warns us against; they therefore stand as foils for the kind of order-based appreciation I am recommending.” --- So the reset is: - Cross and the fictionalists are not latent allies to be recruited; - they are articulate exponents of a powerful but, on your premises, *non-ideal* aesthetic stance; - and their reliance on make-believe or participant framing is precisely what lets you contrast your Carlsonian approach with much of the existing AI-aesthetics literature in a clean, principled way. From there, you can calibrate how much textual space to give them without losing sight of the fact that, on a Carlson-guided view, they belong on the “temptation” side rather than in the toolbox. ab60f8fb-4708-4bbf-9ea6-801fc7ab9357.pdf PDF 1\. that was very interesting, but make sure to separate out claims regarding aesthetic appreciation via make believe from non-aesthetic make believe. treating a customer service chatbot as if it is a perosn is a perfectly legitimate way of interacting with them, in my opinion, it is ony when make believe is used as a means to aesthetically appreciate something that i want to say is a mistake in this paper 2. potentially the paper i have attached to this message is going to complicate matters further, what do you think? Make sure to keep the focus on how I am going to treat this topic in the paper without flooding it with too much of this stuff. at the same time, take this paper seriously, take all the papers i give you seriously. 2/2