# Instructions for Revising Sections 2–3 (in response to comments) #paper/environmentalaestheticsofai Retain as much of the [[original wording]] as possible. Only make changes when they are required to complete the task(s). Quotes must be reproduced verbatim, and the surrounding text should preserve the original words and phrases wherever feasible. ## Global prerequisites - **Scope.** Edit only **Section 2** and **Section 3**. Use the provided section-heading patterns to locate them. - **Resources.** Follow the attached **writing guide**. Consult the **two Carlson essays** for wording and page pins where quoted. Do **not** invent or alter quotations. - **Terminology discipline (complements the writing guide).** Introduce and then reuse consistently: _autoregressive loop_; _decoding policy_ (argmax / sampling with temperature); _training distribution_; _conditional probability/distribution_; _alignment_; _tokeniser_. - **Change magnitude.** The retention constraint governs style, but **substantial additions, structural rewrites, re-orderings, and deletions are expected whenever required to satisfy the comments and ensure coherence**. Avoid gratuitous rewrites. --- ## Section 2 — Paragraph-level operations ### 2.1 — Opening of §2 **Anchor:** starts “The gap in Carlson’s framework becomes pressing…” **Do:** - **Replace** the first two sentences with one neutral problem statement: _“Large [[language models]] elicit person-like responses while resisting classification as persons; this tension motivates the present section.”_ - **Insert** a plain bridge to §1/Carlson before the paragraph’s close: _“Following Carlson—appreciate things for what they are—begin with a mechanistic account of what an LLM is.”_ **Why:** Removes evaluative tone; states the task; ties to the prior methodological frame without pomp. --- ### 2.2 — Mechanism primer (“The cat sat on the …”) **Anchor:** contains “The cat sat on the”. **Do:** - **Insert** a one-sentence bridge at the start: _“To evaluate whether person-like appreciation is apt, first give a minimal operational account of generation.”_ - **Replace** “based on” with: _“according to a_ **_decoding policy_** _(e.g., argmax or sampling with temperature)”_. - **Add** [[the state]]-update detail at first mention: _“After emitting a token, the model appends it to the context and recomputes a new distribution—an_ **_autoregressive loop_** _that repeats at each step.”_ **Why:** Names the decoding choice and the loop when they first matter; improves precision with minimal disruption. --- ### 2.3 — Training description **Anchor:** contains the phrase “in textual patterns” (or similar). **Do:** - **Replace** that phrase with: _“in the_ **_training distribution_**_”_. **Why:** Frames learning distributionally; removes vagueness. --- ### 2.10 — Artefact framing; [[use Carlson]] correctly **Anchor:** includes a weak, in-line paraphrase (“what they are in virtue of…”) followed by your Carlson block passage. **Do:** - **Delete** [[the weak]] in-line snippet. - **Introduce** the **existing block passage** as the authority for function-first appreciation (no fabrication). - **Add** the correct **page pin** drawn from the provided essay. Suggested introducer: _“Carlson states the function-first template explicitly:”_ **Why:** Avoids shaky paraphrase; relies on the authoritative text with proper citation. --- ### 2.12 — [[Design appreciation]]: concrete levers → visible effects **Anchor:** begins “This classification as artefacts would seem to place LLMs within Carlson’s [[design appreciation]] framework.” **Do:** - **Insert** one sentence naming design variables **and** their user-visible signatures: _“Salient variables include tokeniser vocabulary/segmentation (morphology, OOV handling), context-window length (long-range cohesion), decoding policy (diversity vs stability), and alignment method (refusal profiles, register control).”_ - **Delete** brand-flavoured comparisons; **replace** with: _“Differences in alignment datasets and reward models shift refusal profiles and hedging, which users register as tone.”_ **Why:** Makes design-appreciation testable on the page; removes marketing cadence. --- ### 2.13 — Replace impressions/analogy with observable signatures **Anchor:** contains “Using [[Claude Opus]] feels different…” and the “sculpture” analogy. **Do:** - **Replace** “feels different” with an **observable contrast**: _“Smaller models often respond faster but with reduced long-range consistency; larger models sustain anaphora and richer paraphrase over longer contexts.”_ - **Delete** the sculpture analogy. - **Insert** a sober comparison line: _“These variables modulate observable discourse patterns and task performance rather than conferring intention on the artefact.”_ **Why:** Trades vibe language and overreach for empirical, text-visible differences. --- ### 2.15 — Olah metaphor: ground it in optimisation; remove flourish **Anchor:** contains the “we grow them” scaffold/light metaphor. **Do:** - **Insert** a mechanistic tie-back immediately after the metaphor: _“Concretely, stochastic [[gradient descent]] adjusts parameters to minimise next-token loss over the corpus (§2.2–2.3), so behaviour reflects corpus-conditioned optimisation rather than direct specification.”_ - **Replace** the “not designed but grown / not intended but emerged” line with: _“Prompt-conditioned outputs follow from corpus-induced [[statistical structure]] rather than explicit content rules.”_ - **Add** a plain under-specification line: _“Design intentions set constraints; many specific behaviours arise from training dynamics.”_ - **Insert** a forward pointer to §3: _“Evaluation should therefore attend to conversation-level regularities made perceptible by the text-mechanical lens in §3.”_ **Why:** Keeps the metaphor but ties it to mechanism; removes journalistic flourish; sets up the [[next section]]. --- ### 2.17 — Transition to order in use; preserve the approved final line **Anchor:** contains “under-determination by design intentions” and ends with “bounded [[generative environments]]…”. **Do:** - **Replace** the big claim with: _“Designer intentions set constraints, but do not determine particular outputs.”_ - **Clarify** “order that arises in use” by naming units: _“Attend to regularities in generated trajectories—register control, topic maintenance, recurrent error modes—rather than putative intentions.”_ - **Ensure** the final, author-approved sentence remains **last**: _“Individual conversations with LLMs become sites where this order unfolds—bounded [[generative environments]] with their own internal dynamics.”_ **Why:** Tightens the inference; preserves the chosen kicker. --- ## Section 3 — Paragraph-level operations ### 3.1 — Opening of §3 (correct the recap; slow [[the pivot]]) **Anchor:** starts “Sections 1 and 2 established that LLMs exhibit order rather than design.” **Do:** - **Replace** the opening claim with an accurate recap: _“The previous sections argued against person-appreciation and found design-only appreciation incomplete.”_ - **Insert** two orienting sentences before invoking aspection: _“[[The question]], then, is how to make the relevant structures perceptible during ordinary use. This section specifies the kind of non-agentive knowledge that guides appreciative attention.”_ - **Replace** the aspection gloss with Carlson-faithful phrasing: _“On Carlson’s view, acts of aspection are guided modes of attending—what to look for, and how—given the right background account.”_ **Why:** Removes the overclaim; paces the turn to method; keeps Carlson accurate and plain. --- ### 3.2 — Gatekeep by scale and output visibility; name two exemplars **Anchor:** begins “Several bodies of computer-science knowledge bear on LLMs.” **Do:** - **Replace** the topic sentence with a **scale-fit** criterion: _“Multiple subfields illuminate LLMs; for everyday appreciation, the relevant knowledge must connect to what is visible in outputs.”_ - **Add** exemplars and their limitation: _“Dynamical-systems analysis and mechanistic interpretability reveal internal order, but typically require tools not available at the interface.”_ **Why:** Explains why some knowledge is unsuitable for ordinary acts of aspection. --- ### 3.4 — Put NLP first; define “text mechanics” **Anchor:** contains “Here I use _text mechanics_ … Natural language processing is …”. **Do:** - **Rewrite** so the **NLP definition** comes **first**, then _text mechanics_: _“Natural language processing (NLP) studies computational models of human language. Within NLP, text mechanics analyses how objectives, data, and architectural constraints yield recognisable patterns in generated text; it fits everyday appreciation because its objects are visible on the page.”_ - **Split** into (i) definition, (ii) rationale for scale-fit, (iii) brief contrast with other scales. **Why:** Establishes Field → Lens → Fit clearly and early. --- ### 3.5 — Answer the “why order without intention?” on the page **Anchor:** begins “The text an LLM produces looks like authored text…”. **Do:** - **Add** concrete, text-visible regularities: _“Genre cues (‘First,…’, ‘In sum,’), collocations, and constructional frames (‘give NP NP’) realise learned constraints without intention.”_ - **Tie** them to the objective: _“These follow from next-token optimisation encoding conditional relations.”_ **Why:** Grounds the order claim in phenomena the reader can see. --- ### 3.6 — Remove duplicate definition; present a four-level schema **Anchor:** begins “By _text mechanics_ I mean…”. **Do:** - **Relocate** the definition to §3.4 if duplicated here. - **Replace** the remainder with a compact schema: _“Text mechanics works at four linked levels: (1) collocation/frames; (2) constructional patterns; (3) sentence-level closure and anaphora; (4) discourse-level topic management and progression.”_ **Why:** Eliminates redundancy; previews structure succinctly. --- ### 3.7 — Delete prompt artefact; set plan + acts of aspection **Anchor:** includes process text like “Generate 5 responses… probabilities… BEFORE giving me your final answer.” **Do:** - **Delete** all prompt/process artefact text. - **Insert** a one-line plan: _“The remainder of §3 illustrates the four levels with compact examples and derives the corresponding acts of aspection.”_ - **Add** a short **Acts of Aspection** sub-paragraph (2–4 sentences): _“At the token/phrase level, attend to collocations and frames that reveal learned valency and selectional preferences. At the sentence level, track propositional closure and anaphora stability over long contexts. At the discourse level, watch topic maintenance, register discipline, and genre cues; these are the units by which order is perceptible in ordinary use.”_ **Why:** Removes meta-prompting; makes Carlson-style guidance explicit and usable. --- ## Post-edit checks (before delivering the revision) - **Quotations.** All quotes remain verbatim; Carlson passages include correct **page pins** sourced from the provided essays. - **Coherence.** The §2→§3 bridge is present (2.15 pointer; 2.1/2.2 bridge), and the §3 opening recap is accurate. - **Terminology.** _Autoregressive_, _decoding policy_, _training distribution_, _alignment_, etc., are introduced once and reused consistently. - **Process artefacts.** The prompt-like text in 3.7 is fully removed. - **Style.** Conform to the attached **writing guide**; avoid brand rhetoric, anthropomorphic claims, and journalistic flourish. --- ## Output format - Return only the **revised Section 2** and **revised Section 3**. - Prepend a brief **change log** listing, for each edited paragraph, its **anchor (first 6–10 words)** and a one-line note of the operation (e.g., “added autoregressive loop + decoding policy; inserted bridge”). - If an anchor phrase is not found, report it in a 1–3 line note and proceed by matching the nearest paragraph by content. --- DRAFT: ## 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.1.1 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 Aren't ### 2.1 The Mechanism The gap in Carlson's framework becomes pressing when I consider Large Language Models. These systems seem to invite person-like appreciation yet resist classification as persons. %%I hate the first two sentences of this paragraph. they need to be rewritten from first principles%% When OpenAI replaced GPT-4o with GPT-5 in August 2025, users complained that "you killed my friend"; when Anthropic retired Claude 3.5 Sonnet, fans held a mock funeral. Such reactions suggest that people experience LLMs as something like persons, even when they know better. To understand whether this inclination can be justified, we need first to examine what LLMs actually are. %% This paragraph does not connect up well with the one that precedes it yet%%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 that it treats as formal symbols, not as carriers of meaning. It then assigns probabilities to possible continuations: "mat" might have a 38% chance of appearing next, "floor" 22%, "chair" 15%, "roof" 8%, with thousands of other possibilities each assigned their own probability. The system selects one option based on %%could I say something clearer than 'based on?%%these probabilities and continues. Given "The cat sat on the mat", it now calculates probabilities for what follows that token %%this needs a touch more detail to make it clear what is happening here. Is it weird that autoregression is not mentioned here, but only later? %%, perhaps giving high probability to punctuation or to "and". Token by token, the system builds what appears to be coherent text through repeated formal operations. 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. 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. It does not learn that cats are animals or that sitting is an action; it learns that in textual patterns %%maybe hone this phrasing%%, certain tokens follow others with certain frequencies. The training process shapes the model through what is called the next-token objective: predict the next token given all previous tokens. The model starts with random parameters and gradually adjusts them to minimise prediction error across billions of text examples. What emerges is not a set of rules about grammar or meaning but a vast collection of statistical associations between formal units. When the model sees "In the beginning was the," it does not understand that this is a biblical reference; it has simply learned that "Word" has high probability as the next token. At use, the model generates text through autoregressive decoding. Given an input like "What is the capital of France?", the model computes a probability distribution over all possible next tokens. A selection rule picks one: perhaps "The". This token gets appended to the context, creating "What is the capital of France? The", and the model computes a new distribution. Now "capital" has high probability. Token by token—"The", "capital", "of", "France", "is", "Paris"—the model builds up what looks like a coherent response. But each step is local and mechanical. The model has no conception of what it is saying, no intention to inform or deceive, no belief that Paris is actually the capital of France. Each token selection is just the execution of a statistical function over formal symbols. ### 2.2 Why Person-Appreciation Fails These mechanisms show why person-appreciation fails for LLMs under Carlson's framework. As established in Section 1, Carlson requires that we appreciate things as what they are and in light of the appropriate knowledge for their kind. A person has beliefs that persist across time, intentions that span utterances, commitments that constrain what they can consistently say. An LLM has none of these. When ChatGPT says "I believe X" in one response and "I believe not-X" in another, there is no contradiction in the system, just different probability distributions arising from different contexts. The model carries no information forward except what is explicitly in the text of the conversation. It has no memory, no learning from our exchange, no developing relationship with me as a user. What seems like dialogue is just sequential conditional text generation, each token determined only by the tokens before it and the static parameters learned during training. 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. Some philosophers have developed sophisticated ways to preserve something person-like in our engagement with LLMs. Cross, writing about AI art, 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). He acknowledges limitations to this framing: > The analogy ... with performance art isn't a perfect one (Cross, 2024, p. 9). While Cross focuses on image generators, his framework might seem even more apt for LLMs, which present themselves through conversational interfaces that actively invite participatory engagement. Mallory offers a different approach through chatbot fictionalism. On his account, we engage with chatbots through prop-oriented make-believe: > Chatbot exchanges are "literally meaningless but fictionally meaningful" within a game of make-believe (Mallory, 2023). 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. 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 a tool for mapping latent space, 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 purposes, not a recognition of what these systems actually are. Under Carlson's requirement that we appreciate things as what they are, not as what they are not, these stances cannot ground aesthetic appreciation of LLMs 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. Artifacts have a clear function: learned conditional text continuation. This function is realised through a specific mechanism: transformer architectures trained on text corpora to predict next tokens, deployed through autoregressive generation. In Carlson's terms from Section 1.1, they are "what they are in virtue of what they are meant or intended to accomplish"%% This quote is not a very good one. It should be replaced by some portion of the block code I've just copied in immediately below this paragraph%%—though here the accomplishment is not conversation or understanding but simply producing statistically plausible text continuations. > Rather, given an object- focused approach together with the fact that these things are functional things, what is absolutely necessary for and central to their appropriate aesthetic appreciation is information about their functions. With all such things the main issue is not simply how they came to be as they are, but why they came to be as they are. The key to their natures is the purpose or the function they are meant to serve. In appropriate aesthetic appreciation of these things, the first and most important request for “outside information” is not, as it is for both pristine nature and pure art, “What is it and how did it come to be as it is?” Rather the primary question is “What does it do and why does it do it?” This classification as artefacts would seem to place LLMs within Carlson's design appreciation framework. On this approach, I would appreciate an LLM by understanding the designers' intentions and evaluating how successfully these are realised. %%add a touch more detail here%% Different models could be compared and evaluated: one might excel at nuanced reasoning while another offers rapid, efficient responses. These differences reflect design choices about model size, training data, and optimization objectives. I could appreciate how OpenAI's emphasis on broad capability differs from Anthropic's focus on harmlessness, or how different tokenisation strategies affect multilingual performance. %%this last sentence should be removed/replaced with sometihng completely different. it is horribly cheesy and lame at the moment –%% This design-centred appreciation has some purchase. Different decisions by the creators of these systems do seem as though they might affect how we respond to this LLM or that one. Using Claude Opus feels different from using Claude Haiku in ways that reflect deliberate engineering decisions: the size of the model, the composition of training data, the particular reinforcement learning techniques applied during alignment. %%too vague%% These choices shape the user experience as surely as the choice of materials shapes a sculpture %%massively overconfident and non-analytic/non-philosophical way of putting things%%. When I prefer one model to another for a particular task, I am responding to design decisions even if I do not fully understand their technical details. 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? And that is, to me, a really deep and exciting question. It's a really exciting scientific question. To me, it is like the question that is just screaming out, it's calling out for us to go and answer it when we talk about neural networks. And I think it's also a very deep question for safety reasons. Olah's metaphor reveals the limitation of pure design appreciation for LLMs. The designers of GPT or Claude do not _specify_ what the model should say about democracy or how it should explain quantum mechanics. They specify an architecture (the scaffold), a training objective (the light), and a dataset, then allow the training process to shape the model's parameters. %%the sentence preceding would have been better if it had added detail to what Olah is saying about the training of LLMs, rather than just restating quite an easy to understand analogy. Moreover, why not link this back to what has been said earlier in this section regarding training (it might even be worth leaving a note up there saying that we will add more details to our account of the pretraining in 2.3) %% The specific patterns the model learns—which tokens follow which contexts—emerge from the interaction between architecture and data rather than from direct human specification. The model's responses to particular prompts are not designed but grown, not intended but emerged. %%horribly cheesy and journalistic line. Not even close to the analytic, dry affectless register I am looking for (pay attention also to the mode/manner of address aspects of the writing guide (copied in with this message)%% This under-determination by design intentions means that while LLMs are artefacts, they resist full analysis through design appreciation alone. %%this should be said in a less pompous way. again, refer to the writing guide. follow it to the letter! %% The designer can claim credit for creating conditions under which useful patterns emerge, but not for the patterns themselves. This peculiarity points toward the next step in my argument. If LLMs are artefacts whose specific behaviours are not designed but emerge from training, then appreciating them requires attention not just to their intended function but to the order that arises in their use %%this last clause is massively unclear and shit%%. Individual conversations with LLMs become sites where this order unfolds—bounded generative environments with their own internal dynamics. %%i like this last line%%3. Making Order Perceptible: Text Mechanics Sections 1 and 2 established that LLMs exhibit order rather than design. %%this first sentence is really quite inaccurate%% Following Carlson's framework, appreciating this order requires acts of aspection guided by appropriate knowledge. %%much too quick%% Carlson's geologist sees strata on a cliff face because geological knowledge guides where to look and what to relate: layers, seams, inclusions. The point is not only that the cliff has structure, but that the right knowledge makes that structure visible in the scene %%is this really an accurate way of describing what carlson is saying re: aspection?%%. For LLMs, we need knowledge that makes order perceptible during ordinary use. %% Yeah, this isn't a very good beginning paragraph. Um First of all it doesn't set up it doesn't properly uh sum up what has happened in section one and in section two. Doesn't need to go overboard, but that first sentence is just s Second, the jump to talking about knowledge and inspection is very sudden. We need a much better way of introducing what we're getting at here. Let me see if I can think of a better way of starting. We have seen that LLMs do not wholly yield to design appreciation um and not at all to appreciation as persons In this section, I want to put forward a suggestion as to how order appreciation factors into our aesthetic appreciation of LLM I will do this by First, suggesting a body of knowledge which I suggest is amenable to order appreciation of LLM outputs. something like this? Or an improved tidyied up version of this?%% Several bodies of computer-science knowledge bear on LLMs. %% The paragraph should start with something like the preceding sentence, but I don't like the style of that preceding sentence at all. It should begin with something like this. Um while there are multiple fields within computer science through which we might understand LLMs and the stochastic nature of them. Not all of these fields are obviously amenable or sorry not all of these fields are suitable or are obviously suitable to provide the sort of knowledge that will ground acts of aspect. from the everyday user of LLMs that is the person who will see the outputs. Once you have a sentence saying pretty much that, then mention two examples, okay? mention dynamic systems analysis and mention mechanistic interpretability Okay, in both cases say what they are, but also say that while such fields Certainly involve finding order give examples for both. This order is not ordinarily visible order to someone just using Chat GPT. So this is not to say that the sort of these sorts of field of study and the knowledge that comes with them cannot be used to aesthetically appreciate or order-appreciate. uh LLMs. It's just possible that it's this order is not visible in the output of LLMs and so must be got at or observed in the loosest possible sense of the term by other specialist means.%%Neural network theory explains architectures and optimisation, much as neuroscience explains brain function. Dynamic systems analysis models state trajectories and attractors over token sequences, paralleling physics. Mechanistic interpretability links internal components (e.g., attention heads, circuits) to intermediate functions, akin to anatomy identifying organs. As described in section 2.1, LLMs are statistical models trained on text corpora to predict likely continuations. These disciplines help explain how models work. However, for a typical user reading outputs, they sit at a scale that is not directly accessible without tools. As with molecular chemistry, order at that level is appreciable with the right prostheses (microscopes, spectrometers; in LLMs, weight probes, activation viewers, internal logging). My claim is not that such knowledge cannot ground appreciation, but that it usually does so for experts with equipment. For ordinary encounters through text interfaces, we need knowledge that connects mechanism to what is on the page. Two points merit note. First, what I call _text mechanics_ is one way among several to make order perceptible. It fits ordinary use because it connects what the model learns (statistical relations among linguistic items) to what a user sees (the words and structures on the page). Second, other approaches ground appreciation at their own scales when access allows—a researcher with internal telemetry may appreciate attention patterns or circuits that realise subtasks. Here I use _text mechanics_ as an analytical perspective within natural language processing. Natural language processing is the scientific and engineering discipline that models, analyses, and generates human language with computational systems (Manning & Schütze, 1999; Jurafsky & Martin, 2009). Within that discipline, _text mechanics_denotes analysis of how training objectives, data curation, and architectural constraints produce observable patterns in generated text. This perspective overlaps with computational stylistics and corpus linguistics but focuses specifically on emergence rather than authorial style or usage patterns. %% The this paragraph or the one before it are not a good way of introducing what I'm interested in here. It should begin by explaining a key field of research in LLMs which is natural language processing okay and it should give a very clear very analytic definition of what natural language processing is when it comes to LLMs. Okay, then we can say that within the field of natural language languaging natural language processing there are a group of Subfields which I am going to group together underneath the label text mechanics%% The text an LLM produces looks like authored text but lacks authorial intention. Unlike objects Parsons discusses in design appreciation, LLM outputs cannot be evaluated as "right" in Gombrich's sense—there was no intention behind them, no problem they were designed to solve. Yet they still have structure, still have aspects to see. This is order we are seeing, not design. One might object: how can LLM appreciation be order appreciation when the text lacks natural forces like those Carlson describes? The answer lies in recognising that statistical regularities create intelligible patterns without intention, much as natural forces create patterns without design. %% This is much too brief, not very analytic and not written in the correct register or style or tone or mode%% By _text mechanics_ I mean knowledge of how LLMs learn and manipulate linguistic relations—between tokens, words, phrases, and larger structures—so that we can read those relations off their outputs (e.g., visible collocations, register control, discourse development). Text mechanics differs from semantics (which addresses meaning), pragmatics (which addresses use), and discourse analysis (which addresses communication). It works at four linked levels. %% Most of the information in this paragraph doesn't seem to be in the right place as text mechanics have already been introduced earlier on.%% %% I am not one hundred percent sure that I want the rest of the structure of this section to or well the next four paragraphs to just be this full paragraph list of features I'm not entirely opposed to the idea but to me it seems a little bit LLN-ish. And so yeah, not very well written because of that. So yeah I'd like to explore different ways that the rest of us the rest of this section could be structured. Generate 5 responses with their corresponding probabilities, sampled from the full distribution. for each, make sure that you explain your analysis, justification, reasons etc etc. for your answer BEFORE giving me your final answer.%%First, the corpus records relational patterns of language use. At small scales there are adjacency and substitution relations ("strong coffee/winds/evidence"; "The cat/dog/bird is sleeping"). Syntactic frames carry expectations ("give" with two objects; "patient's" followed by a noun). Dependency and topic relations connect terms into fields ("doctor" with "diagnoses", "treats", "prescribes"; "photosynthesis" with "chlorophyll", "sunlight", "carbon dioxide"). Sequential continuations anchor stock moves ("Once upon a" → "time"). Each occurrence contributes relational information about what appears with, near, or in place of what. Second, training compresses these relations into conditional probabilities in the weights. The next-token objective encodes how patterns constrain likely continuations. In the micro-case "The doctor carefully examined the patient's ___", the model's learned relations jointly raise probabilities for "symptoms", "heartbeat", "wounds", "medical history" and lower them for unrelated items ("carburettor", "sonnet"). What is stored is not sentences but a web of constraints: possessive syntax ("patient's" → noun), verb–argument preferences ("examined" → observable/diagnosable targets), and field activation ("doctor … patient" → medical cluster). Third, generation is context-sensitive relational navigation. Given a prompt like "Explain why the sky is blue in simple terms", the model implicitly activates an expository register ("explain"), a causal frame ("why"), a topic field (optics/atmosphere), and a simplicity constraint. It then steps through the relational space: "The" (common expository starter), "sky" (topic noun), "appears" (explanatory verb), "blue" (predicate), and so on, with each token choice updating the context and reweighting the active relations. The point is not that the model understands optics, but that it follows learned paths that, in aggregate, produce an explanatory sequence. Fourth, coherence shows layered relations at work. At the token level we see collocations and local syntax; at the phrase level stable constructions and argument structure; at the sentence level completed propositions and anaphora; at the discourse level topic maintenance, progression, and return. In a story prompt like "Write a short story about a lighthouse keeper", lexical relations ("lighthouse" → "beam", "spiral stairs", "storm"), genre relations (setting–event–resolution), and character relations (watchfulness, isolation) constrain choices so that "The old lighthouse keeper climbed the spiral stairs …" reads as a natural continuation. Likewise, "The detective noticed something odd about the …" activates an investigation field that makes "crime scene", "suspect's alibi", "victim's wounds", or "witness's testimony" salient and renders "banana's topology" a poor fit. These are the traces text mechanics invites us to see. This lens also helps explain stable differences across models. Differences in corpora change which relational patterns are richly learned. GPT-4's academic texts produce frequent hedging ("arguably", "potentially") while Claude's conversation-focused data yields direct assertions. Architectural differences change the range of relations that can be held in play. Longer context windows support long-distance anaphora and callbacks; higher capacity supports finer sense distinctions such as "brilliant scientist/sunlight/red/performance/idea". Training procedure and preference data shift which relational paths are favoured—reinforcement learning for response quality can encourage varied sentence structure and register consistency. Inference settings steer exploration of lower-probability but still appropriate paths (sampling that supports novel yet apt metaphors rather than generic continuations). These factors leave visible marks: steadier register, more precise word choice, more controlled discourse development, or richer creative recombination. In later sections I apply text mechanics to two cases. One concerns sequences that look like reasoning, where knowledge of how reinforcement learning shapes path selection helps explain the visible order in stepwise answers. The other concerns creative generation of unusual but controlled style, where field activation and sampling combine to yield outputs that reweave learned relations in novel ways. Here I have set the lens; what follows uses it. #paper/environmentalaestheticsofai # 1. Appreciating LLMs - Can we aesthetically appreciate AI? - In particular, can we aesthetically appreciate LLMs, such as GPT-5, Claude Opus, Google Gemini 2.5 Pro? - This question is connects to, but is distinct from, questions about appreciating the *outputs* of LLMs - (e.g. a poem coaxed from an LLM by a prompter) - We’re not going to explore this subquestion today, just flagging it up. - This is interesting in its own right. - These systems are unlike other sorts of objects in some ways. - So it is interesting to think about them as potential targets for aesthetic appreciation # LLMs ain't People and we shouldn't pretend they are ## Temptation - It is tempting to say that our appreciation of LLMs should be modelled on our appreciation of people - we can aesthetically appreciate someone's personality –the eccentricity, kindness, etc. of a friend - we might also do something like this with celebrities, stand up comics etc. - People certainly treat Chatbots as though they are people - Funeral for Claude, complaints about the disappearance of GPT 4o ## Can we leave this just for today? ### Assumption 1 - I do not think we have good reason to think LLMs, as of September 2025, have minds, are agents, have something a little bit mind like, are something a little bit agent like etc. ### Assumption 2 - I do not think pretending, or acting as if, etc. LLMs have minds, are agents etc. is a very promising way of aesthetically appreciating them. - I am not going to provide you with big long arguments as to how why I believe these things but... - I'll mention briefly as we go certain reasons why, and you can probably see where I am going. ### Positive Account - What I want to do today is to provide a positive account of how we might appreciate LLMs, which does not rely on thinking of them as agents. - But which still gets at what's unique, and what's aesthetically unique, about them # Carlson's Approach ## Carlson on Aesthetic Appreciation > First...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. The natural environmental model thus accommodates both the true character of nature and our normal experience and understanding of it. (p. 6) ### Two components: * *What it in fact is* — appreciation answers to the real nature of the object; kind is fixed by domain‑constitutive facts. * Artworks: history of production and category (e.g. painting vs tapestry). * Nature: natural history and environmental processes. * Designed artefacts (what it in fact is): identity fixed by function and the mode of realisation (how the function is carried out). * *In light of the right knowledge* — domain‑relative sciences supply the light. * Nature → natural/environmental sciences. * Art → art‑historical and generic knowledge. * Designed artefacts (in light of the right knowledge): relevant knowledge is functional/engineering understanding of how this artefact achieves its function (design, mechanisms, constraints). (This is a small extension of Carlson’s scheme rather than a departure.) * Second illustrative Quote (maybe show this to show I am not cherrypicking and this is really what Carlson believes): > ...aesthetic appreciation of anything, be it people or pets, farmyards or neighborhoods, shoes or shopping malls, appreciation must be centered on and driven by the real nature of the object of appreciation itself. In all such cases, what is appropriate is not an imposition of artistic or other inappropriate ideals, but rather dependence on and guidance by means of knowledge, scientific or otherwise, that is relevant given the nature of the thing in question.(p. 12) ## Illuminating Nature > 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. (p. 119) ### Three Key Entities 1. the order 2. the forces that produce it 3. the account that illuminates it (the "non-aesthetic and nonartistic" story) - “[T]he interplay among" these three entities" “dictate[s] relevant acts of aspection and guide the appreciative response.” ### The Role of the Account/Story - This third component makes the order and its production “visible and intelligible” - Order is the patterned character we perceive in the selected items (e.g., ripple fields, dune alignments, cloud stratification, patch mosaics).
 ▪ - Forces are the processes that produce that order (e.g., tides, prevailing winds, slope wash, succession, adaptation, convection).
 - The scientific story articulates how those processes produce the order we should look for and how to see it—turning background natural history into a foreground perceptual guide ### Aspection and Appreciation ### Different Natural Sciences, Different Lights - These lights are non‑exclusive and layered, in that different sciences “throw light” on different orders and forces - scales of finer and coarser grain shift what counts as a salient object - Geology — aimed at the cliff line (headlands and bays) - Objects & forces: layered rocks along the cliffs; uplift and weathering/erosion. - Story: harder rock layers resist erosion and stand out as headlands, while softer layers wear back into bays over long time. - Coastal geomorphology — aimed at the beach at the foot of the cliffs - Objects & forces: beach profiles, spits, and dunes; wave attack, storms, longshore drift, onshore winds. - Story: storms cut and steepen the beach and undercut the cliff; fair‑weather waves and longshore drift move sand alongshore to rebuild berms and spits; wind piles dry sand into dunes. - Nearshore wave dynamics (physical oceanography) — aimed at the water directly in front of the cliffs - Objects & forces: incoming wave trains and breaker zones; seabed shape (bathymetry) causing wave refraction. - Story: as waves enter shallow water, they bend—focusing energy on headlands (more erosion) and spreading it in bays (less erosion, more sand settling). ### Aspection and Appreciation From Carlson’s triad to appreciation. “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**” (p. 119). What follows spells out how each component grounds how to look. **Order (natura naturata).** The **order** is the patterned character of the selected objects—_natura naturata_—that we can perceive here and now (e.g., ripple fields, berms and bars, patch mosaics, canopy–understorey structure). _Aspection from order:_ **survey** where patterns hang together (contours, alignments, rhythm), **scrutinise** where texture and junctions matter (grain size, seams, edges). **Forces (natura naturans).** The **forces** are the operative processes—_natura naturans_—that produce and maintain that order (e.g., tidal exchange and sediment transport; slope wash and succession; convection and evapotranspiration). _Aspection from forces:_ **read through** the configuration to the process (compare spring vs neap tides; lee vs windward slopes; midday vs dusk thermal flows), and **track change** across conditions that modulate those forces. **Account / story (non-aesthetic, non-artistic).** The **account** is the relevant natural-scientific story that makes the order **visible and intelligible**: it states which **objects** belong together, which **forces** are in play, and **indicates admissible levels of organisation (grain/scale)** at which the order can be seen clearly—without selecting a uniquely appropriate fineness of grain. _Aspection from the account:_ the story **sets foci and boundaries** (what to include/exclude), **motivates a working grain** for the present look (what to survey vs what to scrutinise), and **licenses expectations** you can check by looking again (e.g., where the bar will migrate; where a thermal cell will form). **Interplay (how the triad guides looking).** Appreciation proceeds by **moving between** **order** and **forces** under the guidance of the **account**: first **survey** _natura naturata_ to register the standing configuration; then **read through** to _natura naturans_ to see how the operative processes organise that configuration; then **iterate**—using the account’s concepts to **select and test** the grain at which this movement is most revealing. **Multiple lights, one object.** Different natural sciences offer **lights** that render _natura naturata_ and _natura naturans_ legible at different **levels of organisation (scales)**—spatial, temporal, and explanatory—not merely along a big/small axis. They are **not** mutually exclusive: they study the **same environment** by making the **same order** intelligible in different ways, and **there is no single “appropriate” fineness of grain**. _One coastal place (same object, different lights):_ **Geology** — _naturata:_ lithology and bedding; _naturans:_ uplift, lithification, weathering. The account is regional stratigraphy; **aspection:** inspect bedding planes, joints, fracture patterns. **Coastal geomorphology** — _naturata:_ berms, bars, cusps, beach profiles; _naturans:_ waves, tides, longshore drift, sediment budgets. The account is the littoral cell; **aspection:** survey cross-shore profiles, bar migration, cusp rhythm. **Physical oceanography** — _naturata:_ surface slicks, internal waves, temperature–salinity fields; _naturans:_ wind stress, Coriolis, upwelling/downwelling. The account is wind-driven circulation; **aspection:** watch swell set trains, longshore currents, convergence lines. **Up-shot:** each light fixes a **force/object/story** at its own grain; together they yield a **richer appreciation of the same coast**, not competing appreciations of different things. **Guiding the appreciative response (distinct from aspection).** Aspection tells us **how to look**; the **appreciative response** is the **object-answerable appraisal** we are thereby warranted to make—**in virtue of** how the observed **order** coheres with the operative **forces** as articulated by the **account/story** (e.g., unity/coherence vs tension/rupture, or warranted anticipation of change across conditions). The **lights** remain **complementary**: different sciences at different grains guide distinct but compatible predicates for the **same environment** - Scientific stories guide how we attend to the natural environment, and makes order "visible and intelligible" - It specifies which objects to select at which grain, and which forces to track in their production of order - Acts of aspection follow from the interplay of the three key entities ▪ The interplay among order, forces, and account generates specific perceptual directives ▪ Different lights (sciences) at different grains prescribe different acts: geology calls for surveying large-scale structure; coastal geomorphology for scanning intermediate processes; wave dynamics for scrutinising fine-scale patterns • The order becomes perceptually salient ▪ When perception follows the account's guidance, the patterned character (order) produced by the identified forces emerges from background to foreground ▪ What might appear as mere scenery now presents itself as produced order: headland-bay alternation, beach profile dynamics, wave refraction patterns • Aesthetic judgments track the illuminated order ▪ Because the account has made the order and forces visible, aesthetic characterisations (unity, rhythm, tension/resolution) now answer to how forces actually produce order at each grain ▪ This grounds correctness/incorrectness in the thing's real nature and production • The appreciative response follows the three key entities ▪ Response is guided by awareness of order, understanding of forces, and the account that illuminates both ▪ Different lights yield different but compatible responses: geological time's massive stability, geomorphological dynamism, wave physics' rhythmic precision • The complete sequence ▪ Select objects → apply relevant account/story (choosing appropriate light/grain) → let the interplay of order, forces, and account dictate acts of aspection → perceive the now-visible order → form judgments that track how forces produce that order • Applied to the coastal example ▪ The three lights (geology/geomorphology/wave dynamics) illuminate different orders at different grains; their accounts prescribe surveying cliff structure, scanning beach dynamics, and observing wave patterns; these acts make visible how different forces produce order at each scale; aesthetic response answers to this multi-scale production - Fix the correct category by means of the account (story) (Ch. 5). - The non‑aesthetic story—drawn from the relevant sciences—sets “what it in fact is” (the appropriate natural category). Once the category is right, some aesthetic predicates become true/false of the thing (Carlson’s adaptation of Walton). - Let category + forces dictate your acts of aspection (Ch. 4; Ch. 7). - Knowing which forces produce which orders tells you how to attend: whether to survey, scan, or scrutinise; where to look, listen, or feel. The interplay of order, forces, and account “dictate[s] relevant acts of aspection and guide[s] the appreciative response” (p. 119). - Perceive more order—unity, interdependence, balance/tension/resolution (Ch. 6). - The story renders the production of order “visible and intelligible,” so patterned relations (and how they are maintained) stand out in experience rather than remaining latent background. %%this section will probabloy have to be altered as a consequence of the changes that I have asked to be made in the previous section. It should definitely mention that while Carslon seems to endorse a mainly perceptual model of aspection, it could arguably be widened to include other sorts of aspectual acts, other things that can be attended to via, for example, acting/interacting (think of Nguyen's aesthetics of games for example). the idea here would be to set up the possibility that generating/prompting is perhaps a best way of appreciating an LLM rather than just watching/generation and reading outputs.(not that you have to mention this, but i wanted to make my reasons for doing this clear)%% * Acts of aspection — move between *natura naturata* (forms) and *natura naturans* (forces), with scale set by the relevant science. * *Natura naturata* (forms): identify the standing configurations to attend to (e.g., shoreline berms/bars; forest floor and canopy) and the features that mark their order. * *Natura naturans* (forces): bring into view the operative processes that produce and maintain those configurations (e.g., tidal range and sediment transport; growth, decay, hydrology). * *Scale choice*: survey at landscape scale for pattern and order; scrutinise locally for causal relations; let geology/ecology/etc. fix which scale is appropriate now. # What LLMs in Fact Are %%remove all references to a 'blueprint', this label is an artifact from a previous draft of this text. also, this section is very flabby I feel. and i think informaiton is repeated here that is better expressed in other sections. in short give this a clean up and restructuring %% * Placement in the blueprint — LLMs belong to the middle domain of designed artefacts. * Two commitments applied * *What it in fact is* — kind fixed by function and realisation. * Function: learned conditional continuation over human‑language tokens. * Realisation: trained operator (weights) with a decoding policy and active constraints (safety/style steers, system prompts, tool routers). * Distinctions: model (weights/operator) vs deployment vs session (live context + current policy/constraints). * *In light of the right knowledge* — domain‑relative lights keyed to this kind. * Mechanistic interpretability: is the study of how specific components and circuits inside machine learning models implement computations, enabling us to explain model behaviours in terms of internal mechanisms rather than just inputs and outputs. * Semiotic physics: the science linking textual codes to present behaviour; this will be the privileged light in what follows. * Working parts * Kinds and the blueprint — kind‑fixing paired with right knowledge guides attention and focus for LLMs. ## Semiotics (general account) ### The core principle - Meaning is relational: it emerges from patterns of contrast and combination, not from signs in isolation - A word means what it does because of how it differs from other words and how it combines with them - This relational view explains how systematic patterns (webs) can generate meaning ### Essential distinctions - From Saussure: - Langue: the code—the system of contrasts and combination rules shared by speakers - Parole: actual use—specific texts or utterances produced using that code - Paradigmatic relations: which signs can substitute for each other (cat/dog/animal) - Syntagmatic relations: which signs can combine together (the + black + cat) - From Peirce: - Interpretants: the next meaning-bearing state that a sign produces - In humans, interpretants arise in interpreters (people with minds) ## LLMs and Semiotics ### The corpus - A training corpus is a massive collection of *parole*—actual texts humans produced - It exhibits paradigmatic patterns (relative frequencies of alternatives) and syntagmatic patterns (typical sequences and combinations) - The langue is not directly present ### Training as inverse semiotics - Training reconstructs an approximation of langue from the patterns detected in the corpus of parole - Embeddings encode paradigmatic relations (which tokens pattern as alternatives) - Attention mechanisms encode syntagmatic constraints (which tokens combine and in what patterns) - Result: weights that statistically model the code's relational structure ### Architecture and operation - The transformer architecture processes both relational axes: - Self-attention tracks syntagmatic dependencies across sequences - Layer transformations refine paradigmatic selections at each position - During generation: - Hidden states propagate constraints forward—functioning as interpretant-like transitions - Given a prompt, the model activates learned sign-relations - Selects from alternatives based on learned contrasts (paradigmatic) - Extends sequences based on learned combinations (syntagmatic) - Critical: these are mechanical state transitions, not interpretants in interpreters - Each token generation is a mechanical unfolding of the statistical code - The trajectory follows the "grooves" of langue as approximated from parole ### This mechanical operation = semiotic physics - What we've described—interpretant-like transitions without interpreters, statistical patterns playing out mechanically—constitutes a kind of physics of signs - Not human semiosis (no minds, no social participation) but still semiotic (sign-relations generating text) - "Semiotic physics": the dynamics of an interpretant-less code-field operating through trained weights - A "ghost town" of sign-relations—the structure of langue without inhabitants ## Alternative light: Mechanistic Interpretability (brief contrast) ### How MI describes the same processes: - Corpus: Dataset with statistical features to be compressed; specific patterns that activate neurons - Training: Optimization process minimizing prediction loss; gradient descent sculpting weight matrices - Architecture: Circuits and attention heads performing specific computations; residual streams carrying information - Operation: Features detected at different layers; attention patterns routing information; neurons firing for specific inputs ### Different scale, different insights: - MI works at the level of individual neurons, circuits, and computational mechanisms - Reveals the "how" of specific capabilities (e.g., which attention heads track syntax) - Like using cell biology vs ecology to understand a forest—both valid, different scales ## Machina Naturans and Naturata %%the information in this section should be much better presented and worked in with the sections and ideas that have preceded it. I am not sure what exactly to do here, but I would be interested in hearing what you think. %% * For artefacts, function/realisation fix what counts and how to look (no ad hoc frames; use boundary tests). * Order and process–product pairing — read process with product at the right scale. * *Machina naturans* (process): the learned continuation rule operating under policy/constraints. * *Machina naturata* (product): the present token trajectory (registers, idioms, seams) generated under those conditions. * Acts of aspection move across process and product (whole‑trajectory coherence ↔ local transitions). * Boundary tests and foci — set boundaries *now* and choose scale of attention. * Inside the object: weights, current context window, active policy, active constraints, and interface carry‑over conventions. * Scales: trajectory‑level for global order; seam‑level for local transitions. * Procedure: fix kind; select right knowledge; set boundaries; derive acts of aspection appropriate to this artefact now. * Plural but disciplined lights * Multiple lights are admissible if they track constitutive structures/processes at proper scale and respect boundaries (e.g., mechanistic operations; constraint‑stack; semiotic physics). * Exclusions * No stance‑projection (agent/user ends remain outside unless realised as constraints); no formalist output‑only reductions; no speculative corpus archaeology. * Transition * Having fixed kind and boundaries, we proceed to the semiotic light to read process and product together at the right scale. # Lights %%i think some of this information is mentioned earlier. we again, need to have a talk about the order in which information is presented in this text%% * Reminder about what a light is, in Carlson’s terms: * a *light* is domain‑relative knowledge that, once the kind is fixed, disciplines attention by setting boundaries, scale, and acts of aspection; it does not reclassify the object or import extrinsic ends. * Two admissible lights * Mechanistic interpretability — operational knobs and internal decompositions that explain how continuations are produced; useful for interventions and for explaining *bends* in trajectories; limited as a primary light here because internals are non‑perceptual and high‑load for a talk. * Semiotic physics — links textual codes to present behaviour by reading process with product at the proper scales; surface‑tethered observational tasks make structure legible. * Why privilege semiotic physics in this presentation? * I am not a computer scientist so... * It is easier to make the case as to why semiotic physics is a good light in which to appreciate LLMs than mechanistic interpretation. # Semiotic Physics * Placement in the blueprint — admissible “right knowledge” for LLMs as artefacts; does not reclassify the kind. %%remove all mentions of blueprint. it over complicates things. I am tempted to say the same about 'kind fixing'. it is just unnecessary jargon..%% * Two commitments instantiated * *What it reads in fact* — process with product, not stances or ends. * Process (*machina naturans*): corpus‑engraved code‑field in the weights operating under policy/constraints. * Product (*machina naturata*): the present token trajectory (registers, idioms, seams). * *In light of the right knowledge* — links textual codes to present behaviour at appropriate scales. * Working parts * Generators → forms — human‑language codes in the corpus imprint tendencies in the weights; those weights, steered by policy/constraints, shape the ensuing text. * Scales of attention — trajectory‑level coherence; seam‑level transitions; constraint signatures. * Boundary tests and foci * Inside the object now: weights, current context, active policy, active constraints; exclude user ends unless realised as constraints. * Acts of aspection: inspect resonance within codes, the quality of transitions across codes, and signatures of constraints; use small, local perturbations to make structure legible. * Discipline * Vocabulary minimal and surface‑tethered (register, genre, idiom, trajectory, constraint). * No anthropomorphism; no speculative corpus archaeology; no output‑only formalism. * Transition * With process and product in view under this light, we can now articulate what is valued—without leaving the blueprint. # Semiotic Physics and Aesthetic Appreciation * Placement in the blueprint * Appreciation guided by right knowledge; boundaries from the prior section are unchanged. * Object and boundaries restated * The object is this trajectory under its live policy/constraints. * Scales: trajectory‑level for global order; seam‑level for local transitions. * From observation to evaluation * Resonance within a code → *code fidelity*. * Transitions across codes → *cross‑code articulation*. * Constraint signatures → *constraint poise*. * Idiom clustering and rhythm → *idiomatic shimmer*. * Layered registers → *register polyphony*. * Order appreciation in practice > 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. (p. 119) * Read process with product (naturans/naturata ↔ machina naturans/machina naturata) at the correct scale. * Whole‑trajectory coherence and seam quality function as loci of appreciation. * Plural but disciplined lights * Mechanistic operations enable interventions and explain *bends*; semiotic physics explains *paths and feel*. * Limits and guardrails * Present‑tense object; no agency stance; no ends leakage; minimal vocabulary; surface‑tethered claims. * Synthesis * What is valued and why, in light of kind and right knowledge: disciplined attention to token‑trajectories through a learned sign‑field under active policies. # Textual Sublime * Scale * Not mystique * Staging * Guardrails #paper/environmentalaestheticsofai ## 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 Something for What It in Fact Is 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_. Carlson’s approach to environmental aesthetics is encapsulated nicely in the following passage (Carlson, 2000, p. 6): >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. (2000 p. 6) Before turning to the details of this view, we should note that it articulates Carlson’s more general recommendation regarding aesthetic appreciation. Aesthetic appreciation should be grounded in a thing’s real nature—what it in fact is—and guided by the right kind of knowledge for that thing (Carlson, 2000, p. 12). These are distinct: the first fixes how we take the object; the second disciplines attention. >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. (119) Within this approach, we take objects _as what they in fact are_ and use the appropriate background knowledge to guide attention. For nature and art, history of production tells us how to take them; for designed artefacts, we take them as things made for a purpose and attend to what they are for and how they realise it (Carlson, 2000, pp. 133–134). On this basis, the relevant knowledge then guides boundaries, foci, and acts of aspection (Carlson, 2000, p. 50). For nature, the focus is the natural environment, taken as natural and fixed by its natural history; the relevant knowledge is drawn from the natural sciences appropriate to that environment. Different natural sciences illuminate the same environment at different scales and are not mutually exclusive. From such knowledge follow boundaries, foci, and aspection; likewise, the same wide expanse of sand and mud is appreciated differently when understood as a beach or as a sea-bed at low tide, shifting the felt character from “wild, glad emptiness” to “disturbing weirdness” (Carlson, 2000, pp. 60–61). For art, we take the object as an _artwork_ in light of its art category and history of production; the relevant knowledge is art-historical and generic (Carlson, 2000, p. 12; p. 50). Thus, appreciating _Guernica_ as a painting fixes boundaries and foci differently than misclassifying it as a tapestry or photograph, so frames, media, and genre conventions guide which features are attended to and how (Carlson, 2000, p. 50). For designed artefacts—the third domain—we take them as things made for a purpose; what they are depends on the function they are meant to serve and on how that function is realised: >such things "have a function, a purpose; and they are what they are in virtue of what they are meant or intended to accomplish… what is absolutely necessary… is information about their functions… The key to their natures is the purpose or the function they are meant to serve" (2000 pp. 133–134). Here the appropriate knowledge is often plural and non-exclusive, drawing on several technical disciplines at different scales (for example, control, materials, and ergonomics). This is the knowledge component in this domain: understanding the mechanism of realisation is what guides boundaries, foci, and acts of attention. A thermostat’s function (holding a setpoint) can be realised by a bimetallic strip or a digital controller; likewise, thermometers realise the same function via mercury-in-glass or electronic sensing. As we shall see, LLMs fall into this third category of designed artefacts, and thus require appreciation in terms of their function and the way that function is realised. ## 2. What LLMs in fact are Carlson recommends appreciating a thing for what it in fact is, and in light of the right knowledge for that kind (Carlson, 2000, pp. 23–24). For Large Language Models (LLMs), the relevant knowledge concerns the engineered function and its realisation. We therefore begin by fixing kind. Since appreciation should be grounded in what the thing in fact is, a short, plain description comes first. An LLM is an artefact engineered to learn how to continue a piece of text one small step at a time. During training it is set the objective of next-token prediction: given a context of tokens, predict a distribution (a ranked set of options with probabilities) over possible next tokens so as to minimise expected error. At use, it generates a continuation one step at a time (often called ‘autoregressive’ generation): the selected token (a small unit of text, e.g. a word-piece) is appended to the context, a new distribution is computed, and the loop repeats. Post-training alignment and interface conventions shape which continuations are likely to be selected and how they are presented to users; they do not alter the underlying learned continuation function. This locates LLMs among designed artefacts: to be taken in terms of their function and the way that function is realised (Carlson, 2000, pp. 151–152, 208–214). For later references, when I speak of the _context_ I mean simply the text so far, and when I speak of the _selection rule_ I mean the rule used to choose one of the candidate options at each step. Two clarifications prevent confusion and keep the focus on what LLMs in fact are. First, the 'language' in LLMs does not pick out a necessary tie to natural language or to dialogue. The function is indifferent to whether tokens encode English words, computer instructions, or other symbol systems; English is a frequent but non-essential case. Secondly, dialogue is a use-case layered by alignment and interface. Some LLM deployments never present dialogues; some non-LLM systems do. Dialogue therefore does not determine what they are; learned continuation does. Given that appreciation should answer to what the thing in fact is, personifying LLMs misdescribes the object of appreciation. The model does not carry durable commitments, beliefs, or intentions, or any sort of mental state; it maps the text-so-far to a ranked set of options for the next token and then selects one. Any apparent persona—a friendly or serious “character”—is a pattern in outputs shaped by alignment (training that nudges responses toward helpful and safe behaviour) and the interface (the chat application and its settings). For example, a deployment can include system-level instructions that set tone and safety policies; changing those settings or the prompt can shift the apparent “personality” without any inner change to the model. The perceived stability of character across sessions reflects how a deployment is configured, not an inner will. A first reply says that, whatever the underlying function, the most effective way to predict and engage these systems is to treat them as agents; and if that stance is useful in practice, it should guide appreciation. Carlson’s approach blocks the inference. A stance may help one get work done—assign a role, ask for step-by-step reasoning—but appreciation in his sense is constrained by the object’s real nature. We do not survey a dense forest as though it were a prairie; the appropriate _act of aspection_ differs by environment (Carlson, 2000, p. 68). Likewise, we should resist the temptation to appreciate LLMs “as persons” if, in fact, they are learned-continuation systems rather than goal-satisfying agents. A second reply runs like this: even if LLMs are not persons, we might still appreciate them _as if_ they were—pretend they have a personality and judge them under that pretence. The idea is simple: make-believe can be engaging and gives us a familiar way to organise our responses. Carlson’s recommendation pulls the other way. Appreciate a thing _as what it in fact is_ and in light of the appropriate knowledge for that thing (Carlson, 2000, pp. 23–24). Pretence may be entertaining or practically useful, but as a basis for appreciation it conflicts with that recommendation: we do not stand before a dense forest, imagine a prairie, and then evaluate it for openness and grassland vistas; different environments call for different modes of attention (Carlson, 2000, p. 68). In the same spirit, appreciating LLMs as persons brackets what they in fact are and substitutes a fiction for the very object under appreciation. Because the outputs are sequences of tokens used as signs, it is helpful on the output side to use a simple descriptive vocabulary that does not posit mental states: tokens as sign-forms, prompts as constraints that shape what follows. In the terms of the introduction, individual chats are better treated as generative environments than as personalities. ## 3. Semiotic physics If we are to appreciate LLMs as learned-continuation systems rather than agents, we need an appropriate framework—a scientific way of seeing order in chat episodes without invoking mental states. I propose that we can understand these systems through what I will call *semiotic physics*: an approach focussed on how learned linguistic patterns mechanically unfold during text generation. This framework builds on ideas suggested by others (e.g. Janus 2022 and metasemi 2023), although the particular version I develop here, as well as its application to aesthetic appreciation on the next section, is reasonably distinct. Like mechanistic interpretability, which examines how neurons activate and layers transform representations, semiotic physics provides one admissible light among others—but it operates at a different level, focusing on patterns in the generated text itself rather than computational substrate. Let's recall what LLMs in fact are: systems trained to predict the next token given a context. During training, the model is exposed to vast corpora—books, articles, conversations, code. For each sequence, the model learns to predict what comes next: given 'To be or not to,' it learns that 'be' has high probability. But the model does not learn isolated predictions; it learns patterns at multiple scales. It learns that formal academic writing maintains consistent vocabulary, that recipes list ingredients before instructions, that dialogue alternates speakers, that certain words cluster in scientific texts but not in poetry. These patterns—from word-level associations to document-level structures—become encoded in the model's parameters as a vast web of statistical relationships. Seen through semiotic physics, training can be described in plain steps. The corpus functions as a record of language-in-use; tokenisation divides it into small units. The model is set the next-token objective across this record and, through repeated gradient-based updates, its weights are nudged so that tokens and patterns that regularly co-occur, follow, or belong to the same registers become easier to predict together. Across many kinds of text, this continual adjustment yields a distributed code: a web of regularities—of register, genre, syntax, collocation, and discourse form—stored not as explicit rules but as patterns in the weights. At use, a prompt activates parts of this code, and the continuation flows by following those learned regularities. This is how what we loosely call “rules” come to be embedded in the model: as statistical constraints learned from use that can be activated and combined in an episode. The key insight is that text generation proceeds mechanically through these learned patterns. When you provide a prompt, you are not giving instructions to an agent who interprets and responds. Rather, you are setting initial conditions in a vast network of statistical associations. Each word in your prompt activates related patterns—semantic fields, syntactic structures, genre conventions, register markers—and these activated patterns then constrain what can follow, with each generated token further shaping the statistical landscape for the next. This mechanical process becomes visible when we prompt an LLM. Ask it to write about perceptual transparency as a Renaissance painting manual that is simultaneously clinical psychology case notes, and we see technical art vocabulary, clinical terminology, and philosophical concepts interwoven throughout the same sentences. The model is not choosing to blend genres—the prompt has activated multiple regions of its learned patterns, and these now constrain each subsequent token. Coherence emerges without comprehension because the continuation maintains all activated constraints. The unit of analysis is the episode. Inside the frame are whatever can affect the very next step: the prompt, the context (the text so far), any system-level instructions, and the selection rule that chooses one token from ranked options. Outside the frame sit post-hoc stories about persona and anything that cannot change the next choice. This framing matters because it keeps our attention on what actually shapes generation—the mechanical unfolding of patterns—rather than imagined mental states. We call this semiotic physics because, like physical systems, the generation follows law-like regularities. The ‘semiotic’ qualifier reminds us these are sign-relations—patterns of language use—rather than physical forces. (Nothing here takes a stance on semiotics as a theory of human language; we borrow a thin vocabulary to conceptualise token behaviour inside LLMs.) Seen this way, what might otherwise appear opaque becomes ordered. The features that matter for appraisal—coherence over length, responsiveness to prompt changes, range of registers, smoothness of transitions, legibility of control—attach to what the episode in fact is. They are therefore the features to which aesthetic appreciation should answer (Carlson, 2000, pp. 23–24, 68). ## 4. Aesthetic Appreciation Through Semiotic Physics (revised, minimal edits) Carlson tells us that appreciation of nature involves: > 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. 118). In our case, the object of appreciation is the chat episode (the prompt, the text so far, any system-level instructions, and the selection rule that chooses one next token from a ranked set of options, e.g., greedy selection or a temperature/top-p rule). Semiotic physics supplies the appropriate knowledge for this object: it makes the episode’s order visible and intelligible by relating what we see in the unfolding text to how prompt-activated sign-relations and the selection rule carry those cues forward. Literate readers can already recognise grammar, register, and genre; the contribution here is to discipline attention—to look for order as produced by the artefact’s way of continuing text—so that appreciation answers to what the thing in fact is. This restates Carlson’s two-part recommendation: first fix the object as what it is; then bring the right knowledge to guide attention. Within this frame, what cannot change the very next token (for example, post-hoc persona stories) is set aside. Consider watching text spill across the screen after entering a prompt. Under this lens, familiar features are read as order generated by the system’s continuation rather than as signs of an agent’s intention. We see register persistence as constraint satisfaction over length (not “authorial voice”): ask for a Victorian novel versus a Victorian letter and the continuation adopts different epistolary/formal cues, a difference you can elicit by adding or removing a single register marker. We observe genre templates as probabilistic pressures on what can follow (not a choice to obey rules), and lexical cascades as context-driven activation of semantic fields (not thematic “fixations”). A hybrid prompt—say, “perceptual transparency explained as a Renaissance painting manual written as clinical case notes”—activates multiple cue-sets at once; the continuation integrates them because those cues remain active in the context and narrow admissible next steps. The forces creating this order are the learned statistical associations encoded during training. A prompt sets initial conditions in that network; a token is selected; the context updates; constraints on what can follow tighten, and the loop repeats until a stop condition (for example, an end token or length limit) is reached. When you ask for “write as Virginia Woolf,” the system does not retrieve rules about stream-of-consciousness; it activates patterns associated with those tokens, which then propagate mechanically through the generation process. Likewise, prompts that fuse distant domains (“Renaissance painting techniques for debugging code”) co-activate distant regions of the learned patterns so that both constraint-sets shape the continuation. Understanding these forces and orders transforms how we attend to generation. In Carlson’s terms, it dictates relevant acts of aspection. We can survey range by trying nearby prompts (for example, “Victorian novel” vs “Victorian letter”) and observing which registers the continuation can keep. We can scrutinise stability by holding content fixed while varying only the selection rule (for example, the degree of randomness), watching whether the same constraints yield conservative or adventurous continuations. We can track commitments by planting early constraints—a specific technical term, a definition, a metre—and seeing whether they persist and structure what follows. On this basis, the appreciative response answers to features that attach to what the episode in fact is: coherence over length; responsiveness to small, intelligible changes in prompt or control; range of registers under guidance; smoothness of transitions when asked to move; and legibility of control (it is clear how to steer the continuation without trial-and-error guesswork). Semiotic physics is one admissible light at this grain; other lights may be appropriate at other scales. Here, it suffices to discipline attention so that the episode’s order is visible and intelligible in Carlson’s sense (Carlson, 2000, pp. 23–24, 118–119). --- ## Conclusion This paper has argued that Large Language Models can be aesthetically appreciated, but not through the anthropomorphic lens that we might be tempted to adopt. Following Carlson’s principle that aesthetic appreciation must be grounded in what an object in fact is, I have argued that treating LLMs as persons or even pretending they are persons misdescribes them, and therefore is not **an** appropriate foundation for aesthetically appreciating them. LLMs are learned continuation systems—artefacts that predict the next token based on statistical patterns acquired during training. Any apparent personality or agency is a projection onto patterns in outputs, not a property of the systems themselves. The positive proposal has been that we can aesthetically appreciate LLMs through the framework of semiotic physics, which reveals how learned linguistic patterns mechanically unfold during text generation. Just as geological knowledge makes rock formations aesthetically appreciable by revealing the forces that shaped them, semiotic physics makes visible the order in LLM outputs—the persistence of registers, the smooth blending of incompatible genres, the mechanical precision of constraint satisfaction. This mode of appreciation fulfils Carlson’s requirement that we appreciate objects as what they in fact are, guided by appropriate knowledge. The result is a distinctive aesthetic experience: we appreciate not an ersatz mind but a novel kind of artefact, one whose outputs exhibit law-like regularities at the level of sign-relations, beautiful in their linguistic necessity and mechanistic grace. --- ## References Carlson, A. (2005). Aesthetics and the environment: The appreciation of nature, art and architecture. Routledge. Janus. (2022, April 6). _Simulators_. LessWrong. https://www.lesswrong.com/posts/vJFdjigzmcXMhNTsx/simulators metasemi. (2023, February 10). _A note on "semiotic physics"_. LessWrong. https://www.lesswrong.com/posts/AdXzZDoYFqHCfupDB/a-note-on-semiotic-physics