```dataview LIST WHERE paper = "environmental aesthetics of generative ai" ``` ```dataview LIST WHERE paper = "the environmental aesthetics of generative ai" ``` # Old 2nd March Version ## 1. Appreciating Natura Carlson’s _Natural Environmental Model_ for [[environmental aesthetics]] rests on two ideas: > "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) Regarding his first point—that we must appreciate nature as what it in fact is—Carlson argues that the environment is not merely a collection of objects or scenes but a system of interconnected elements shaped by various processes and forces (ibid. p. 44). Environments thus “come about ‘naturally,’ [in that] they change, grow, and develop by means of [[natural processes]].” From this perspective, [[aesthetic appreciation]] involves recognising what one is encountering—a [[natural environment]] in which components are interrelated—and understanding it in light of scientific or other relevant knowledge that illuminates its composition and development. Just as an informed grasp of artistic traditions can enrich one’s appreciation of an artwork, Carlson argues that familiarity with geology, biology, or ecology can guide our attention to patterns and processes that might otherwise remain unnoticed. For instance, one might begin by noticing only the colours or shapes of a coastal cliff’s sedimentary layers. However, discovering that these layers formed over thousands of years of deposition and compaction reveals changes how one regards it aesthetically. If we see a forest or reef as subject to diverse forces and processes, an appropriate aesthetic engagement will centre on how those forces have shaped what we observe. Carlson considers an understanding of the *unity* of the [[natural environment]] to be a relevant to aesthetic **understanding**: > “natural objects possess [...] an [[organic unity]] with their environments of creation: such objects are a part of and have developed out of the elements of their environments by means of [[the forces]] at work within those environments. Thus the environments of creation are aesthetically relevant to natural objects.” (p. 44). [[Organic unity]] between a system and its constituents can be understood in terms of Spinoza’s concepts *[[natura naturans]]* and *[[natura naturata]]*. [[Natura naturans]] refers to the active, generative aspect of nature—such as seeds sprouting from soil—whereas [[natura naturata]] designates the relatively stable entities that result, including trees, soil, and pebbles. Appreciating the forest is informed by an understanding of the flora, fauna, and substrata of which it consists; appreciating a particular tree in a forest will be enhanced by an understanding of [[the forces]] and elements which have led to its creation and qualities. - Thus, nature can be understood both as the self-causing, active process ([[natura naturans]]) and as the passive, resultant structure ([[natura naturata]]). ## 2. Generative Environments I suggest that Carlson’s recommendations also hold for the aesthetic appreciation of generative AI: not only should we appreciate these systems for what they are, in light of our scientific knowledge of what they are, but they are also fruitfully characterised as a type of environment. What then, are generative AI systems? Consider first the remarks of Chris Olah, Co-founder of Anthropic: > I think one useful way to think about neural networks is that we don't program and we don't make them. We kind of, *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 [...] 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, you know, the light that it grows towards. But the thing that we actually create, it's this almost biological, you know, entity or organism that we're studying. > > (Lex Fridman, [Chris Olah](app://obsidian.md/Dario%20Amodei): Anthropic CEO on Claude, AGI & the Future of AI & Humanity | Lex Fridman Podcast #452) my emphasis. LLMs are 'grown' in so much as they are not coded line‑by‑line but developed through an iterative training process in which their internal parameters adapt over time in ways that cannot be entirely predicted. Factors which influence how such a system grows include the specific architecture of the model, the nature and extent of the data used to train it, and the particular training goals set by the researchers. We can describe this evolving process as an instance of “machina naturans.” While human developers determine the overall architecture and set the training parameters, the system’s eventual capabilities emerge from patterns that are not explicitly scripted. - **this part is weak and needs to be rewritten - This process parallels how natura naturans can shape an environment over time. In nature, geological and ecological factors steer gradual changes that culminate in the formation of particular features, without direct step-by-step intervention. Similarly, in a machina naturans system, the interplay of training data, architectural constraints, and iterative updates drives the emergence of new capabilities in ways not fully determined by its human creators.** ## 3. Gardeners and Natura One might might wonder how the preceding discussion is going to help us understand how AI art is aesthetically appreciated. So far it would seem that at most we have suggested a way in which generative AI systems themselves can be the object of aesthetic appreciation. Moreover, we might think that the knowledge required to aesthetically appreciate AI systems is out of reach of almost everybody. Chris Olah, might be in a position to aesthetically appreciate generative AI systems in virtue of what he knows about them, but the vast majority of the users of these systems do not know their backpropagation from their gradient descent. Does this not mean that aesthetic appreciation of AI is out of reach of the population anyway? Here, I argue that a comparison between prompting and *gardening* can help answer these questions. A gardener directly engages with natura naturans with the aim of influencing the resulting naturata. For example, disbudding a dahlia plant redirects growth resources from multiple small florets to a single, symmetrical bloom. The gardener is not creating a dahlia bloom ex nihilo, nor does she impose form on inert material as would a sculptor; rather, she guides its inherent potential towards the form she wants. We might also think that, over time, the gardener gains knowledge of her environment, such as knowing to avoid watering dahlias too frequently after observing how excess moisture weakens their stems. Though she would not use terms like **_cortical cell lysis_**, her practical knowledge – spacing plants and rotating beds based on wilt patterns – is that the dahlia's tubers perform better when the soil is well-drained. While she may not be able to express her knowledge in scientific terms she _understands_ her garden. In engaging with natura through her actions she comes to know it, as 'what it in fact is'. It is also not hard to see how this practical understanding grounds her aesthetic appreciation of gardens, others' as well as her own, and the natural environment in general. Her knowledge of what dahlias are, the natural processes which impact them, and how one might guide these processes will inform her aesthetic appreciation of dahlias. ## 4. Prompters and Natura - Similar to the gardener, the prompter does not produce AI Art _ex nihilo_, nor impose on matter some fixed plan on as a painter might paint a picture; the prompter, like the gardener, *draws* naturata *out* from naturans. - While natura naturata is a phrase for a rather fuzzy sort of thing –all organic matter is in some sort of naturans flux, given that all living things are at some level or other in a state of flux - that is natura is working, doing something - they are you know bodies are slowly decaying rocks are slowly being eroded - whereas if we think of makina natura we find that these things are much more fine grained and focused - given that we can think of a passage of text from chat gpt or an image from mid journey as finished crystallisatios of makina natura - ### claude version which i don't love of 4. Similar to the gardener, the prompter does not produce AI Art _ex nihilo_, nor impose a fixed plan on matter as a painter might approach a canvas; the prompter, like the gardener, _draws_ naturata _out_ from naturans. When a user interacts with Midjourney or ChatGPT, they are not creating something from nothing, but rather guiding the generative potential of the system toward particular manifestations. Just as the gardener redirects the growth potential of a plant, the prompter steers the generative capabilities of an AI system toward specific outputs. Through this iterative interaction, prompters develop a unique position from which to appreciate both the system and its outputs. A person who regularly uses Midjourney, for instance, begins to recognize how certain terms in a prompt influence the style, composition, and elements of the resulting image. They come to understand, not through technical knowledge of neural networks, but through practical engagement, how the system interprets and responds to different guidance. This parallels how gardeners develop practical knowledge about soil conditions and plant responses without necessarily understanding the cellular biology involved. The prompter's direct engagement with the system creates a form of understanding that positions them to appreciate both the generative capabilities of the system itself (machina naturans) and the specific outputs they help create (machina naturata). This practical knowledge grounds their aesthetic appreciation in a way that differs from more passive or theoretical engagement. While natura naturata is a phrase for a rather fuzzy category of things—all organic matter exists in some sort of naturans flux, given that all living things are at some level or other in a state of change—we can still meaningfully apply this distinction. Natural entities are continuously subject to processes: bodies slowly decay, rocks slowly erode, forests gradually transform. The distinction between process and product in nature is never absolute but exists along a continuum where stability is always relative. When we turn to machina natura, however, we find that these processes and their results are much more fine-grained and focused. The machina naturans of a generative AI system produces distinct "crystallizations" in the form of machina naturata—completed texts from ChatGPT or images from Midjourney. Unlike the gradual, ongoing transformations in nature, these outputs can be understood as finished products, specific manifestations of the generative process that have reached a state of completion. This difference in the relative stability of outputs makes the appreciation of AI-generated artifacts distinct from environmental appreciation in important ways, while still maintaining the fundamental parallel in how understanding of process informs appreciation of product. - - - - - And, similar to the gardener, the prompter is uniquely placed to appreciate the synthetic unity between machina naturans and and machina naturata. - While a computer scientist may understand the architecture, training data, and iterative procedures of a generative model, the prompter can gain practical knowledge by observing how small changes in prompting redirect outcomes. - In learning how the system responds, the prompter acquires a partial yet valuable understanding of what the system in fact is, without requiring comprehensive technical expertise. - Architectural constraints and patterns emerge through iterative trial, mirroring the gardener’s observations of how watering practices or planting configurations steer growth. - The user who engages with these patterns can develop familiarity with the interplay of encoded artefacts, data, and refinements within the latent space, even without formal terminology. - Such engagement places the prompter in direct contact with the system’s emergent capabilities, much as the gardener’s familiarity with soil and seasonal factors informs her appreciation of a single bloom. - In this sense, the prompter and gardener are in a special position to aesthetically appreciate the generative AI system or garden environment because they stand within the unity (perhaps even as part of that unity) between individual elements and the processes and forces that shape them, paralleling Carlson’s notion of “organic unity.” - Instead of merely viewing a static artefact, the prompter interacts with a system that evolves over repeated refinements. - Recognising that the system’s final outputs result from a dynamic interplay of constraints and training goals addresses the suggestion that only experts can appreciate generative AI. - Prompting, like gardening, becomes a hands-on practice that fosters aesthetic appreciation through guiding, observing, and discovering how new capabilities appear over time. - This partial but practical understanding demonstrates that one need not possess specialist scientific knowledge to appreciate AI art in the sense of understanding how it emerges from a process resembling _natura naturans_. - Consider now the prompter. Like a gardener she can be thought - The prompter, like a gardener, does not need to impose a complete design on inert material. ### old version of this section - Instead, a can guide an ongoing process of machina naturans through repeated practical engagement, much as the gardener influences natura naturans by disbudding a plant or adjusting moisture to cultivate the resulting naturata. - While a computer scientist understands the model’s architecture, the nature and extent of the training data, and the iterative process through which a system’s internal parameters adapt, the prompter can still gain practical knowledge by observing how small changes in prompting redirect the output. - In this way, the prompter learns what the system in fact is, without requiring explicit command of every parameter. - The architectural constraints and patterns are revealed through trial, in parallel with the gardener who notices how stems respond to watering or how different arrangements guide latent energy in a particular direction. - The user’s familiarity with the interplay of encoded artefacts, iterative updates, and data within the latent space can illuminate the processes that shape AI art in ways not fully determined by scripting. - Even if one lacks formal terminology, hands-on engagement can make the underlying processes visible. - Such engagement places the prompter in direct contact with emergent capabilities, much as the gardener gains an informed appreciation of her environment by understanding how soil or season influences a single bloom. - Rather than merely witnessing a static artefact, the prompter interacts with a system that changes, grows, and develops through iterative refinements. - Recognising that the system’s eventual outputs result from a dynamic interplay of constraints and training goals addresses the concern that knowledge of AI is inaccessible to most people. - Prompting can thus be understood as a practice comparable to gardening, enabling users to appreciate AI art by refining prompts, observing how the system generates specific patterns or stylistic features, and discovering how it develops new capabilities over time. - This partial but practical understanding shows that one does not require scientific expertise to recognise how AI art emerges from a process akin to natura naturans, thereby clarifying how AI art can be appreciated in light of what these systems in fact are. ## 5. Unity and Appreciation Unity plays a crucial role in Carlson's approach to aesthetic appreciation. His environmental model emphasises how understanding the relationships between parts and processes enhances our ability to appreciate natural environments. The unity between natura naturans and natura naturata—between the active processes of nature and their resulting structures—provides a foundation for meaningful environmental aesthetics. When we see a forest not merely as a collection of trees but as an integrated system shaped by countless interactions, our appreciation deepens. This concept of unity takes on special significance when applied to generative AI systems. These systems are not merely collections of code or weighted parameters—they represent something fundamentally different. The critical insight here is that generative AI systems constitute the _encoding and decomposition of a vast corpus of experiential artefacts_. A system like ChatGPT or Midjourney does not simply contain abstract rules for generating text or images; it encodes patterns derived from millions of examples of human creative expression. This encoded corpus of experiential artefacts fundamentally shapes the system's generative capacity. Given this understanding, we can see that prompters engage with a form of unity that goes beyond simple "code-to-output" relationships. When working with generative AI, they participate in a relationship between the specific experiential artefacts they elicit through prompting and the broader corpus of encoded experiential artefacts within the system. A text generated by ChatGPT bears a relationship not just to the prompting process that created it, but to the vast body of text that informed the system's parameters. Similarly, an image from Midjourney connects to a broad visual tradition encoded in its training data. This perspective aligns with Carlson's emphasis on understanding things "in light of" relevant knowledge. Users with extensive knowledge of experiential artefacts—literature, art, music, or other creative domains—can better recognise the stylistic influences and references present in AI-generated outputs. Someone familiar with Impressionist painting, for example, might better appreciate how a Midjourney image adapts Impressionist techniques when prompted to do so. This knowledge enhances appreciation not through technical understanding of the system's architecture, but through recognition of cultural patterns and references. The development of practical knowledge through hands-on engagement further enhances this form of appreciation. Just as gardeners develop intuition about plant responses through direct interaction, prompters gain understanding of how generative systems respond to different inputs. They learn which phrases elicit particular styles, tones, or content, without necessarily understanding the computational processes involved. This practical knowledge grounds aesthetic appreciation without requiring technical expertise, making meaningful engagement with AI-generated art broadly accessible. Ultimately, prompters become part of the unity they appreciate. Their engagement bridges individual artifacts and the broader cultural corpus encoded in the system. When a user prompts ChatGPT to write in the style of Jane Austen, they participate in a relationship that connects the resulting text, Austen's literary corpus as encoded in the system, and their own understanding of Austen's style. This completes the parallel with environmental aesthetics, where appreciation involves recognizing one's place within the appreciated system. Just as a hiker's appreciation of a forest is enhanced by understanding their role in the ecosystem, a prompter's appreciation of AI-generated content is deepened by recognizing their participation in the generative process that connects cultural history to new artifacts. ### brainstorm idea for section five unity of what idea here unity is important in Carlson's view in appreciation okay unity between and naturata the interesting thing about machina naturata and therefore machina naturata that generative AI systems simply um weights or code of any old type they are the encoding vast corpus of experiential artifacts so chat gpt text uh mid-journey images so as an AI artist draws skilfully from the rounds they are not simply drawing it from a machine they're drawing it the imposed experiential artifact # Idea for section five - unity is important in Carlson's view of appreciation - unity between and naturata the interesting thing about machina naturata and therefore machina naturata is that generative AI systems simply are not just weights or code of any old type > [!IMPORTANT] > - they are the *encoding and decompostion of a vast corpus of experiential artifacts* - so in working with the generative system a prompter is not simply in a position to appreciate the unity between text or image being created and a load of code - They are in a position to appreciate the unity between the experiential artifacts (image or text) that they are able to elicit from the system and the corpus of experiential artifacts that have been encoded. - Notice how well this fits with understanding a thing 'in light of'. A person with a vast knowledge of experiential artifacts (or at least some set of them) will likely enhance their appreciation of the system and its outputs ### LLM TEXT ON THIS IDEA - We have explored _machina naturans_ and _machina naturata_ and noted that Carlson stresses unity for understanding a system. Recognising how the parts and processes fit together helps us appreciate it. - In the context of generative AI, this unity links the visible outputs (images or text) with the large body of encoded experiential artefacts that produce them. - A prompter can see how a single generated artefact connects to the wider collection of cultural and textual data within the system. - This reflects Carlson’s idea that understanding something “in light of” relevant knowledge matters. Someone who knows many experiential artefacts is more likely to see how certain styles or themes appear in particular outputs. - Gaining practical knowledge of how the system’s encoded data works is similar to learning about an environment’s geology or biology. Greater insight into the factors that shape it leads to a deeper sense of unity. - Generative models predict the most likely next word or image. A user who can guess these probabilities shows an awareness of the system’s cultural environment. - A broad “vocabulary” of prompts and familiarity with cultural references in the system helps the user interact with it more effectively. - As noted in earlier sections, gardeners and prompters both guide and observe a process. Just as a gardener understands soil and plants, a prompter learns how AI adapts to different prompts. - The unity of a generative system comes from combining stored artefacts with ongoing output. Appreciation involves noticing how each new text or image grows out of a larger corpus. - The prompter does not only see the final result but also senses the traces of a broader cultural background behind the system’s outputs. - By working directly with both the immediate artefact and the wider collection of data, the prompter becomes part of that unity, in a way that parallels how a gardener is connected to an evolving environment. - This brings Carlson’s framework of environmental appreciation into generative AI. Awareness, hands-on engagement, and recognising unity form the basis for appreciating these systems. ### oriand Ultimately, prompters become part of the unity they appreciate. Their engagement bridges individual artefacts and the broader cultural corpus encoded in the system. When a user prompts ChatGPT to write in the style of Jane Austen, they participate in a relationship that connects the resulting text, Austen's literary corpus as encoded in the system, and their own understanding of Austen's style. This completes the parallel with environmental aesthetics, where appreciation involves recognising one's place within the appreciated system. Just as a hiker's appreciation of a forest is enhanced by understanding their role in the ecosystem, a prompter's appreciation of AI-generated content is deepened by recognising their participation in the generative process that connects cultural history to new artefacts. # idea for a possible section 5? - Okay, so it seems like I'm going to have enough space to go into more detail about generative AI systems. - This is an important and interesting direction. - To really know a model and understand what a particular model does - similar to knowing what a particular section of the environment is like - one not only needs to have a wide vocabulary to use these systems effectively, but also needs to have enough cultural knowledge to understand what a particular system is. - This is what these models are doing anyway. - They're working around the *most likely* average next predicted word or most likely image that fits this prompt. - So if a user of these systems understands enough about the cultural environment, the environment of experiential artifacts, then they can better anticipate how the model will respond. # scraps about ai art from the 3000 word draft ### transcript In this section I shall argue that an analogy with gardening can provide the answers to these questions. - Gardening can be thought of as a direct engagement with natura naturans. - At the very least a gardener learns how to garden that particular plot, what grows well there, what doesn't, what needs to be done to make things grow there well. - These possible problems are all part of direct interaction with the naturans processes that we talked about earlier. - The gardener learns these things - about these processive forces and how they feed into the naturate in that particular garden, having a direct appreciation of the unity between the system processes and the more stable elements of that system. - This is a way somebody can understand and thereby appreciate the natural environment without having to have a very scientific theoretical background. - Let's return to the prompter. - It seems that we can also say that the prompter, as they're prompting and iterating on the images that Midjourney produces or producing and iterating on the text that an LLM produces, they are also experiencing some naturans aspects - the machina naturans aspects of the system with which they are working. - This knowledge of naturans, whether arrived at via science or practical engagement, might also allow the gardener to aesthetically appreciate the outputs of gardening. - Earlier, we saw that Carlson wants to say our appreciation of the environment depends on, or is at least enhanced significantly by, scientific understanding of the environment - how to look at it like an act of inspection. - We can also say that a related, slightly dissimilar version of this would be a gardener looking at the fruits (if you'll pardon the pun) of their fellow gardeners' work. - They will not only be able to look at the pretty flowers but also know something about the ease with which that type of plant is grown, and the naturans forces that went into the flower being as it is. - We can now say something similar is going on in the case of the prompter. - Someone who is knowledgeable about producing things with generative AI systems will look at the objects that are produced by prompters and such systems with a more discerning eye than someone with little interest in AI art. - This is exactly the same for real art as well, of course. - One might still object that this analogy is not doing all of the work that it needs to. - Prompting AI is still not something a lot of people know very much about, and so you might think "Well, maybe I've explained how AI artists can appreciate AI art, but I haven't explained how non-AI artists can appreciate AI art." - Here I'd say, well, this is what somebody would need to understand AI art. - Think of this final analogy: it's quite easy to imagine a group of people who aesthetically appreciate gardens without gardening themselves, nor having any particularly sophisticated scientific understanding of gardens. - However, they might still learn, perhaps in a more practical way but not practically learned, about natura naturans and natura naturata. - The people I have in mind here are the people who enjoy Gardener's World, which is a beloved and long-running BBC TV show. - Just as a small digression and analogy, think about the amount of pleasure people get from watching cooking shows and seeing how good food is made, and understanding something about the systems of how food is made without cooking very much themselves or having anything like a sophisticated understanding of the chemistry behind creating food. - Perhaps though, there is nothing wrong with saying that if we want to understand how someone could appreciate AI art, we might think that for that person to appreciate AI art, a certain amount of learning is required, and the more one knows, the more one is in a position to appreciate AI art. - Such a perspective offers a new lens for engaging with the outputs of generative AI. - Rather than evaluating an AI-generated painting or poem merely by human-art standards, one can reflect on how the outputs exhibit the system’s generative processes and constraints. - A comparison might be made with horticulture: a gardener prepares the soil, selects seeds, waters the plants, and arranges conditions. - Yet the ultimate shapes and features of the resulting growth are not entirely dictated by the gardener. - Likewise, a user enters prompts, sets parameters, and selects from model options, but the final text or image arises from the interplay between those user actions and the model’s internally “grown” capacities. - On this analogy, one may find aesthetic value in recognising both the user’s partial influence and the system’s partially autonomous unfolding. - Carlson’s approach to natural environments shifts the focus away from function and towards an understanding of how ongoing processes generate the scene before us. - Similarly, a generative model’s capabilities and aesthetic qualities can be illuminated by knowledge of how the system was trained, which data it encountered, which architecture channels that data, and how prompts guide the generation. - By paying attention to that interplay, one can see the outputs as emergent “fruits” of a dynamic environment. - One might object that these AI systems are still human-designed. - How, then, can they be approached like natural environments? - While there is indeed a profound difference between a rainforest and a neural network, Carlson’s own account acknowledges that even natural environments can be subject to human intervention, and that artifactual or partly modified landscapes can also be appreciated in an environmental mode if they exhibit dynamic processes not entirely dictated by an overarching design. - Generative AI meets that requirement insofar as the training process yields unexpected emergent properties. - Another potential objection is that Carlson’s framework seems to require scientific or specialist knowledge, whereas many AI users lack a formal background in machine learning. - However, Carlson emphasises that there is a continuum of knowledge: a professional geologist may bring deeper insights than a casual hiker, but both can appreciate a canyon’s layered rock formations to some degree. - The same holds for AI. - One need not be an expert in backpropagation or gradient descent to notice consistent patterns or design constraints in a model’s outputs. - Users who experiment extensively with prompts can acquire an intuitive sense of where the model excels or flounders, which styles it can mimic convincingly, and how it reverts to certain rhetorical habits. - This practical familiarity provides a basis for a more informed aesthetic appreciation, much as a local walker can notice the changing seasons or visible ecological patterns in a forest, even without formal scientific training. - Another challenge is that generative AI changes rapidly, with new versions, updates, and entirely new models appearing. - Whatever knowledge one acquires about a particular system might become quickly outdated. - Yet it is also true that natural environments evolve and that acquiring knowledge about them is an ongoing process. - A forest may undergo transformations due to climate change or invasive species, requiring an observer to update their understanding. - Carlson’s emphasis on how knowledge guides attention allows for this dynamism. - One can remain open to new manifestations of generative AI in the same way that an ecologist stays alert to environmental changes. - Indeed, for many users, it is precisely this sense of continual discovery—of finding new uses or encountering surprising behaviours—that fuels aesthetic wonder. - Seeing the outputs themselves as belonging to an environment points toward a shift away from certain categories typical of art criticism. - If we evaluate an AI-generated poem as though an individual human wrote every line, we may feel uneasy about the apparent absence of personal intention. - When, however, we treat that poem as an instance of a larger system’s emergent capacities, the result can be appreciated in a manner that acknowledges both the user’s role and the system’s algorithmic processes. - Instead of searching for authorial sincerity or purely functional design, we notice how the patterns of training data, the structure of the model, and the user’s guiding prompts converge in a particular textual or visual form. - The horticultural parallel again holds: the arrangement in a cultivated garden may be valued not as a purely wild or purely engineered creation but as a hybrid that calls attention to the interplay between design and emergent growth. - A further extension applies to the user communities that have developed around generative AI. - While not formal “ecologists,” many enthusiasts share prompt strategies, test the model’s limits, and exchange observations about how it evolves in new versions. - This communal practice generates a shared body of anecdotal or informal knowledge, analogous to how hikers share trail conditions or local knowledge about certain ecosystems. - These collaborative efforts enrich the aesthetic dimension by revealing previously unknown “paths” through the model’s latent space or documenting novel emergent behaviours. - Such knowledge helps users attend to aspects of outputs they might otherwise overlook and fosters the development of a more reflective engagement with what these systems produce. - Some may argue that this approach fails to consider the moral or societal implications of generative AI, such as biases baked into the training data or the possible erosion of human creativity. - Yet acknowledging ethical or social concerns does not preclude aesthetic appreciation, just as one can appreciate an ecosystem’s complexity while still worrying about pollution or habitat destruction. - Indeed, knowledge of these broader issues may further inform the aesthetic experience by highlighting tensions within the system’s development or revealing how certain outputs reflect cultural distortions or omissions. - Appreciating something aesthetically does not entail endorsing every dimension of its operation, but understanding how it works can deepen one’s perspective on its aesthetic and broader significance. - It might also be questioned whether equating an AI system with an environment overemphasises the emergent aspects. - After all, these systems are designed and tested for commercial or research purposes, and they remain dependent on human-made infrastructures. - Unlike forests or coral reefs, they do not exist autonomously in the wild, and they require vast amounts of energy and compute resources to function. - Nonetheless, Carlson’s logic does not require that an environment be fully free of human influence in order for it to be appreciated in an environmental manner. - Urban green spaces or managed woodlands can still be appreciated as “environments” in Carlson’s sense, provided that the observer attends to the array of forces shaping them and appreciates the way these forces operate together. - In the case of generative AI, the “forces” include training algorithms, data distributions, user prompts, and ongoing updates—none of which is purely random. - Still, their confluence often yields outputs and capabilities that developers did not plan in detail. - By directing attention to the open-ended, partially unforeseeable character of these systems, we can begin to appreciate them in a manner akin to how Carlson suggests we appreciate natural ecosystems. - Where, then, does this leave the aesthetic appreciation of individual AI-generated works, such as images or poems? - We may still examine them as stand-alone artefacts, noticing compositional or stylistic features. - However, this traditional approach sits awkwardly with how such works arise. - A painting that takes a human artist months is distinct in origin from an AI-generated image produced in seconds via latent space sampling. - The conceptual framework offered by Carlson’s environmental aesthetics invites a supplementary way of understanding these outputs, not exclusively as fully formed artworks but as manifestations of a larger, processual system. - One can note the emergent patterns that shape them, the training corpus that influences them, and the user’s guiding input. - This perspective reduces confusion over authorship or authenticity and instead locates the aesthetic object in the relationship between system and user, data and architecture, design and unplanned growth. - An additional concern relates to the degree of knowledge that fosters appreciation. - Carlson is clear that one need not be a scientist to benefit from science-informed insights. - Similarly, AI appreciation does not require being a professional machine learning researcher. - A minimal grasp of the training process and the significance of learned latent representations can suffice to shape one’s aesthetic attention. - For instance, if a user knows that Midjourney’s training emphasised certain art movements or that it had incomplete coverage of other visual traditions, this contextual information might clarify why the system produces some styles more readily than others. - One might look at an image and say, “It resembles a pastiche of nineteenth-century Romantic paintings, which may result from the prevalence of that tradition in the training data.” - That observation parallels noting that a mountainside’s colours and contours reflect underlying geological strata or historical glacial activity. - The more knowledge one acquires—whether scientific, technical, or experiential—the richer the aesthetic engagement becomes. - It is worth stressing that in Carlson’s view, informed attention does not merely accumulate facts but also offers a structured way to look at or listen to something. - A geologist does not simply know more facts about a rock formation but is trained to see patterns and processes in that formation. - Similarly, an AI researcher or an enthusiast with substantial practice might “see” the tell-tale signs of overfitting, biases in generated text, or the influence of certain architectural choices. - They are directed to features and patterns that a novice may miss. - As a result, they can develop a more layered appreciation of the model’s outputs and of the system’s generative character. - Carlson’s analogy between knowledge of an artistic tradition and knowledge of environmental science applies here: just as understanding the conventions of a genre can deepen one’s appreciation of a painting, some familiarity with the generative process can deepen one’s engagement with AI outputs. - One might also mention the community dimension: the knowledge base for generative AI is partly distributed across many hobbyists, developers, and casual users. - They exchange tips about prompting, share unusual responses or images, and track changes between model versions. - This public conversation allows the collective refinement of a system’s “ecological map,” so to speak. - Everyone contributes incremental insights into how the AI “behaves,” how it might have been trained, and how new versions differ from older ones. - Such knowledge sharing strengthens the possibility of an environmental mode of appreciation by foregrounding the processes and conditions that shape outputs. - Instead of seeing each image or text in isolation, the community begins to recognise the patterns of emergence behind them, creating a shared interpretive framework that parallels how amateur naturalists share observations about a local woodland or coastline. - Several objections still remain. - One is that generative AI is so thoroughly artifactual that it cannot be analysed using a framework developed for nature. - Yet Carlson’s fundamental insight is less about the “naturalness” of an environment and more about the fact that an environment is shaped by ongoing processes that one can partially understand through structured knowledge. - We can appreciate farmland or other managed landscapes on an environmental model if we see how they result from the interplay of climate, soils, human decisions, and so forth. - AI is similarly shaped by an interplay of data curation, architecture design, training constraints, and user interactions—factors that can be studied, at least in principle, and related to aesthetic experience. - Another objection is that knowledge of AI systems can be difficult to acquire, not only for laypeople but even for experts, given the complexity of deep learning. - Yet Carlson’s model only requires that one has enough knowledge to direct one’s attention. - Increasing one’s knowledge can enrich the appreciation, but even partial knowledge may suffice to begin noticing relevant features. - A further concern is the rapid pace of AI development. - Models evolve, new architectures emerge, and user communities discover inventive applications. - Some might worry that an environmental approach presupposes more stability than generative AI can offer. - However, natural environments are far from static: they too evolve, sometimes dramatically, requiring ongoing engagement by those who seek to understand them. - AI’s changing character can likewise become part of the aesthetic intrigue, much as one might track a volcanic landscape’s transformations over time. - Rather than a one-time standard, knowledge becomes a continuous process of learning how the system is evolving, which intensifies the sense of ongoing exploration central to an environmental perspective. - Finally, generative AI calls for a novel kind of attention that does not evaluate each artefact solely on the grounds of traditional artistic intention. - While a painting by an individual artist prompts us to consider the vision and skill of that artist, an AI-generated image prompts us to consider the role of training data, user input, and algorithmic processes. - Focusing on the emergent interplay of these factors might enrich our response to AI outputs, avoiding simplistic dismissals of them as “merely machine-produced” or naive claims that they are “equivalent to human artworks.” - Instead, one can see them as distinct phenomena, worthy of aesthetic consideration precisely because they arise from a complex environment shaped by human and computational activities together. - In sum, one may appreciate generative AI by acknowledging its environment-like character and drawing from Carlson’s environmental aesthetics, which views appreciation as grounded in understanding of underlying processes and forces. - These AI systems, which can be described as machina naturans, are grown rather than fully programmed in the traditional sense, and their functions are multifarious rather than singular. - Adopting an environmental perspective clarifies how knowledge of data distributions, model architectures, and emergent patterns can inform aesthetic engagement in a way that parallels scientific understanding of natural ecosystems. - While significant differences remain between actual natural environments and any software-based creation, the points of contact—particularly around ongoing processes and partial autonomy—warrant taking Carlson’s framework seriously. - If a primary lesson of environmental aesthetics is that informed awareness of processes enriches our aesthetic experience, then the same lesson can be applied to generative AI. - By seeing it not as a mere tool or as an entirely human-like creator but as a dynamic system shaped by multiple forces, we gain a more nuanced and instructive way of engaging aesthetically with these new artificial landscapes and the images or texts they produce. - By foregrounding this process-based stance, we can appreciate how generative AI resists simple function-based classifications while providing opportunities for aesthetic wonder and critical reflection. - The parallel with nature illuminates that appreciation may involve an evolving familiarity, ongoing curiosity, and openness to what emerges, grounded in the partial but growing knowledge we bring to bear. - Like an environmentalist who tracks subtle changes in a rainforest’s flora and fauna, a thoughtful user of generative AI can learn to see patterns, anticipate certain emergent behaviours, and reflect on the interplay between design intention and open-ended development. - This shared perspective, drawn from Carlson, offers an intellectually coherent framework that relocates AI appreciation away from purely anthropocentric or purely functional approaches, instead emphasising the manifold processes through which these systems generate novel forms. - In doing so, it helps to make sense of why and how we might find aesthetic value in the ongoing interplay of data, algorithm, user input, and emergent output that characterises the evolving field of AI art. # old version of section 3 In this passage, Olah argues that neural networks develop through a process of growth rather than by a mechanical assembly. He observes that while designers establish the overall architecture and set the training objectives, the system “grows” in a way that is analogous to how a living organism evolves. In this account, the network’s capabilities are not built one step at a time; instead, they emerge iteratively as the network processes large volumes of data. Consequently, the final patterns and functions arise without being explicitly programmed. Olah’s discussion parallels Carlson’s ecological perspective, which holds that geological and ecological forces produce formations without direct human orchestration. In both cases, the resulting configurations and functionalities evolve in a manner that is, to some degree, independent of any single engineer’s blueprint. This observation leads to the suggestion that generative AI may be viewed as a kind of “machina naturans.” In other words, even though people collect, clean, and curate the data, the system ultimately exhibits configurations and behaviors that no designer can fully predict—a phenomenon reminiscent of Spinoza’s distinction between \(\natura\ naturans\) (nature in the act of producing) and \(\natura\ naturata\) (nature as already produced). Conceiving of a large model as an evolving environment further illuminates its multifaceted utility. Olah’s comparison to a living organism highlights the existence of many “microhabitats” within the network. Users, by applying different prompts, effectively traverse these internal regions. Much like Carlson’s notion of learning about varied natural terrains, one cannot reduce such a model to a single function. Instead, it resembles a dynamic landscape in which one may elicit creative stories, moral arguments, translations, or computer code. The particular path taken through the model’s terrain depends on the underlying data distribution, the internal weighting, and the architecture. By navigating these routes, users can observe recurring themes, habits, and characteristic biases in the system similarly to how a botanist might explore a rainforest and note how plant life responds to different conditions of sunlight or moisture. The system thus invites a form of open-ended exploration. Repeatedly prompting the model reveals new areas of competence as well as intricate patterns of error, akin to discovering unexpected species in a rainforest. These emergent properties appear to arise from the model’s own “ecology”—the interplay between data, architecture, and training aims. Following Carlson’s insight, one might argue that achieving an informed aesthetic appreciation of the system requires familiarity with these algorithmic forces. Even for those who are not experts in machine learning, noticing characteristic modes of response and testing prompts that yield consistent results fosters an understanding of the model’s internal order. In this way, as with the study of natural landscapes, a deeper familiarity with the processes shaping its outputs enriches our encounters with this generative environment. # AI as a Designed Object type:: draft Generative AI is sometimes described as a 'new type of tool' (REF). If this were correct then existing accounts of design aesthetics could be extended to cover these such systems. However, this may not work for all existing accounts of design aesthetics. In design aesthetics, one frequently assesses an object’s function, examining the relationship between a designed artefact’s intended purpose and its form. On Parsons and Carlson's 'Functional Beauty' approach, for example, an understanding an object’s function helps clarify which of features contribute to its aesthetic qualities. A similar perspective is found in Forsey’s view that design is to be judged according to how well a product realises its purpose within a specific category or intended use. Yet these function-based approaches look less promising for generative AI, a This is because Functional Beauty theory relies on seeing what something looks like to understand its design and function. But generative AI systems don't really have a visual appearance that we can easily see and judge. We don't know what the 'form' of a generative AI looks like in a way that we can aesthetically appreciate. For Forsey's theory, you need to understand how the design (form) helps the object do its job (function). But generative AI systems are often 'black boxes', meaning we don't really know how they work inside. Because we can't see the 'form' of the AI and how it creates its function, Forsey's way of thinking about design aesthetics also becomes difficult to apply. %%not happy with some of the language above%% # Old version of the disucsion on naturata and naturans By recognising the distinct individuals that constitute a forest and how each is shaped by the forest’s conditions, we enhance our appreciation of both the individual tree and the wider environment. Carlson’s views as to what the natural environment is can be understood in terms of the Spinozian concepts of natura naturans and naturata naturans. Natura naturans refers to the active, generative aspect of nature –trees drawn upwards to sunlight, seeds spouting from soil– whereas natura naturata refers to the (relatively) stable *things*, that naturans generates: the trees, the soil and pebbles on the forest floor. - What would this mean in terms of appreciation? - This helps us understand what carlson is talking about with unity. > natural objects possess what we might call an organic unity with their environments of creation: such objects are a part of and have developed out of the elements of their environments by means of the forces at work within those environments. Thus the environments of creation are aesthetically relevant to natural objects. (p. 44) > When nature is aesthetically appreciated in virtue of the natural and environmental sciences, positive aesthetic appreciation is singularly appropriate, for, on the one hand, pristine nature—nature in its natural state—is an aesthetic ideal and, on the other, as science increasingly finds, or at least appears to find, unity, order, and harmony in nature, nature itself, when appreciated in light of such knowledge, appears more fully beautiful. (p. 11) - all things into existence. In contrast, natura naturata designates nature as the passive product generated by natura naturans; it consists of the created universe of phenomena, namely the individual entities that arise as modes or expressions of nature’s essence. For example, consider a forest ecosystem. One may distinguish between the following: - **Natura naturans:** This includes the innate drive of trees to grow towards sunlight, the generative power of soil to support plant life, and the evolutionary adaptations that shape species. - **Natura naturata:** This encompasses the individual trees, plants, animals, and fungi as well as the overall structure of the forest canopy and understory, along with the food webs and nutrient cycles. - In Carlson’s view, the patterns we see in a forest—such as the seasonal variations, the regenerative growth, and the continual interaction of climatic factors—reveal natura naturans. - In these cyclical processes, the forest is perceived as a living system in perpetual flux. - By contrast, natura naturata refers to the finite, concrete outcomes of these active processes. - - Natura naturata encompasses nature as a produced, relatively stable system, denoting the established elements—the trees which make up the forest, the animals that live in it– trees of various species—that appear settled. - It encompasses the ordered structures—like the established tree trunks, the fixed species composition, and the spatial distribution of flora—that result from nature’s self-sustaining operations. - While natura naturans speaks to the inherent dynamism running through the forest, natura naturata reflects the stable, enduring features that define its visible form. - This distinction supports Carlson’s central claim: we achieve true aesthetic appreciation of the forest by recognizing not only its apparent stillness, but also the underlying processes that have brought it into being and continue to maintain it. - By contrast, natura naturans refers to nature’s active, self-sustaining power that generates and transforms these structures over time, for example through ongoing regenerative cycles that continually shape the forest’s composition. - Carlson observes that natural objects share an “organic unity” with their surroundings, an idea encapsulated in [QUOTE] (Carlson, [placeholder for citation]). - A single rock illustrates this principle: as natura naturata, it is a realised outcome of geological forces, shaped over millennia by tectonic shifts and erosion. - The broader environment—natura naturans—remains active and continually reshaping those very conditions that formed the rock. - Temporally, a rock’s present form contrasts with the ongoing processes of soil accumulation or water flow; causally, its fixed composition reflects completed actions, whereas its environment generates new formations; epistemologically, the rock is immediately observable, while the deeper environmental ground that produced it is only partially knowable. - Through this lens, Spinoza’s duality becomes concrete: natura naturans refers to the ever-active, generative backdrop, whereas natura naturata manifests as discrete objects within that dynamic unity. # Old version of section 3. [type::scrap] [paper:: environmental aesthetics of ai] ### old version We suggest that Carlson's recommendations also hold for the aesthetic appreciation of generative AI: we should appreciate these system for what they are, and we should appreciate them in light of our scientific knowledge of what they are. Moreover, I suggest that generative AI systems themselves might be thought of as a type of environment. First, consider the remarks of Chris Olah, Co-founder of Anthropic: > I think one useful way to think about neural networks is that we don't program and we don't make them. We kind of, we grow them. You know, 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 [...] 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, you know, the light that it grows towards. But the thing that we actually create, it's this almost biological, you know, entity or organism that we're studying. And so it's very, very different from any kind of regular software engineering... > > (Lex Fridman, [Chris Olah](app://obsidian.md/Dario%20Amodei): Anthropic CEO on Claude, AGI & the Future of AI & Humanity | Lex Fridman Podcast #452) my emphasis. - who describes neural networks in terms that echo processes of growth rather than mechanical assembly. - He notes that designers can specify an overall architecture and a training objective, but what actually emerges is more akin to a grown organism than to a conventional piece of software. - The system “develops” capabilities iteratively as it encounters large amounts of data, so the final model exhibits patterns and functions that the developers did not write out line by line. - This dynamic resonates with Carlson’s emphasis on processes that shape a natural environment: geological or ecological forces bring about certain formations without direct human orchestration. - Although generative AI is unquestionably a human invention, and the data are collected and curated by people, the training process still exhibits partial autonomy and yields outcomes not fully anticipated by any designer. - For that reason, one might treat a large model as “machina naturans,” drawing a parallel with Spinoza’s notion of natura naturans. - Spinoza distinguishes natura naturans (nature as active, self-producing) from natura naturata (nature as already produced). - In an AI context, machina naturans highlights the ongoing, generative dimensions of these systems, whose complex internal transformations and emergent capabilities can be likened to a system that “grows” forms rather than simply retrieves them. - Seeing generative AI as an evolving environment has distinct advantages over thinking of it as a single-function tool. - One might prompt a text-based model for different types of outputs—stories, explanations, translations, or code. - The multiplicity of uses suggests that the model is not built with one function in mind. - Instead, it is more like a dynamic terrain, which users can navigate with prompts that direct the generation in various ways. - This “terrain” is determined by the large-scale data distribution underlying the model, as well as by internal weighting and architecture. - In effect, the user is touring a learned manifold of latent representations, with each prompt sampling or activating a different region. - Much like an environment, the system supports open-ended exploration. - One can sample again and again, discovering patterns, biases, or surprising areas of competence and incompetence. - Analogously, a botanist might walk through a rainforest to observe how plant life emerges in different microhabitats, shaped by light, soil, and rainfall. - In generative AI, the data distribution, architecture, and training constraints collectively shape the model’s “ecology.” - Where Carlson insists that aesthetic appreciation of an environment benefits from at least some knowledge of geology or ecology, an informed appreciation of an AI system arises when one understands how these algorithmic and data-driven “forces” shape the final outputs. - One need not be a machine learning expert to notice recurring themes, discover characteristic “habits” of the model, or discern how certain prompts reliably lead to specific results. - This experiential familiarity fosters what Carlson calls the “making intelligible” of a system’s order. - # Scrap on Function and Chatgpt [type:: scrap] t least in the form of large language models (LLMs) or multimodal models used for text and images, because such systems do not have a single, clearly defined function. - A conventional tool like a hammer has a straightforward purpose, and we can assess its design by seeing how well it drives nails. - In contrast, what is ChatGPT “for”? - It can be used to produce study aids, generate short stories, compose code, create outlines for essays, or support unanticipated applications discovered only after its release. - There is no fixed, singular role for ChatGPT or other LLM-based systems, so the link between form and function that underpins conventional design aesthetics becomes tenuous. - If one cannot specify a central function in advance, it is not obvious how to assess generative AI on a function-based standard. - Complicating matters further, many developers of generative AI candidly concede that they do not fully understand how their models perform various tasks. - Modern deep learning involves complex, large-scale data-driven processes, sometimes described as “opaque,” in which the model is trained to minimise a loss function but emerges with capacities that designers did not explicitly plan. - The open-endedness and partial inscrutability of these systems suggest that function-based approaches, reliant on a stable definition of purpose, do not map well onto them. # Notes on Multi-Function Artifacts and LLMs [type:: note] The idea that LLMs are grown as opposed to programmed or made is evocative, but before considering that, it is interesting to note that if we are picky about Amodei's use of the term 'artifact,' we might wonder whether LLMs really are artifacts. On the one hand, they are intentionally brought into existence, but on the other, it is hard to understand them as having a proper function. A chair is for sitting, a car is for driving, an LLM is for...? Arguably, one ends up saying its function is something very general like producing text, producing human-like text etc, or ascribing it an extremely large number of functions. Neither is it obviously a sort of container for other artifacts, like a Swiss Army knife or a mobile phone (the function of which is arguably as an environment in which to store and run apps), as users do not install applications on an LLM, but rather they discover them (examples include creative writing, code generation, language tutoring, and various forms of research assistance). Many of these user innovations are quite technical, but they underscore the difficulty of assigning a single function to an LLM.** # Notes on Naturans and Carlson [type:: notes] [paper:: environmental aesthetics of artificial intelligence] On Carlson's account, understanding the enviroment in terms of the elements and processes is central to the appreciation in the same way that knowledge of artistic traditions and styles is necessary to appreciate artworks (p. 50, **and the ziff reference**) > “if to aesthetically appreciate art we must have knowledge of artistic traditions and styles within those traditions, then to aesthetically appreciate nature we must have knowledge of the different environments of nature and of the systems and elements within those environments.” (p. 50) - Furthermore, these natural entities “are a part of and have developed out of the elements of their environments by means of the forces at work within those environments.” (ibid. p. 44) - This statement indicates that natural objects and their environments are fundamentally connected, evolving through natural processes rather than through deliberate design. Turning to Carlson's second idea——we see that he regards environments as entities that “are not the products of designers and typically have no design. Rather they” emerge as outcomes of natural processes. He further points out that “in the way in which the art critic and the art historian are well equipped to aesthetically appreciate art, the naturalist and the ecologist are well equipped to aesthetically appreciate nature.” (ibid.) Accordingly, recognising these processes and the interrelation of the systems and elements provides the groundwork for aesthetic appreciation. By acknowledging nature’s unity, order, and coherence in this way, we can regard it not only as an evolving physical environment but also as an aesthetically significant one. ### 2.2 Appreciating Naturans - If we turn to Carlson's second idea, that appreciation of nature takes place "in light of knowledge provided by the natural science", we can see that he considers > environments [...] are not the products of designers and typically have no design. Rather they - The natural processes he has in mind are those studied by biologists, ecologists, geologists etc. This way of conceptualising nature fits well with an older idea: Spinoza's concept of *natura naturans*. Natura naturans refers to the active, self-causing dimension of nature, which he equates with "God or Nature".[^2] - In his view, knowledge of geology, biology, and ecology enriches our aesthetic engagement by revealing how landscapes and ecosystems emerge. - This understanding goes beyond attending merely to surface-level properties such as shapes or colours; it extends to recognising the mechanisms that shape environments. - He writes that we must “appreciate nature as what it in fact is, that is, as natural and as an environment” and “in light of knowledge provided by the natural sciences.” - This contrasts with models of nature appreciation that treat natural settings as if they were static artworks, or that reduce appreciation to an emotional response without understanding. - Instead, Carlson underscores a holistic engagement, informed by at least some grasp of ecological and evolutionary forces. - For instance, an old-growth forest might be appreciated differently once one learns how mature trees, understory plants, and fungi interact to sustain growth, resource exchange, and biodiversity. - The point is that scientific or at least practical knowledge helps the appreciator see and hear aspects of the forest that might otherwise go unnoticed. - With this knowledge, we attend more thoughtfully to the movements of species or the structure of the canopy. ### 2.2 Appreciating Naturans Generative AI DRAFT 27th February 2025 # Quotes I will use paper:: the environmental aesthetics of ai type:: quotes ## Carlson > "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." (p. 6) “if to aesthetically appreciate art we must have knowledge of artistic traditions and styles within those traditions, then to aesthetically appreciate nature we must have knowledge of the different environments of nature and of the systems and elements within those environments.” (p. 50) He further points out that “in the way in which the art critic and the art historian are well equipped to aesthetically appreciate art, the naturalist and the ecologist are well equipped to aesthetically appreciate nature.” (ibid.) Accordingly, recognising these processes and the interrelation of the systems and elements provides the groundwork for aesthetic appreciation. By acknowledging nature’s unity, order, and coherence in this way, we can regard it not only as an evolving physical environment but also as an aesthetically significant one. ## Dario Amodei > I think one useful way to think about neural networks is that we don't program and we don't make them. We kind of, we grow them. You know, 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, and they sort of, you know, it starts off with some kind of random, you know, 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, you know, the light that it grows towards. But the thing that we actually create, it's this almost biological, you know, 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, you know, write essays and translate and, you know, understand images." (Lex Fridman, Dario Amodei: Anthropic CEO on Claude, AGI & the Future of AI & Humanity | Lex Fridman Podcast #452) my emphasis.** # Text from an LLM that can be used in a future draft - In light of these flexible, open-ended capabilities, aesthetic appreciation of AI faces a conceptual question: _By what standard do we judge systems whose “function” is so fluid?_ Function-based approaches, which normally tie aesthetic virtue to an artefact’s fulfilling its purpose with elegance or efficiency, struggle here. One might say a large language model’s purpose is “to generate text,” but the possible uses of that text are so diverse that it becomes nearly impossible to define a stable “ideal form” the system must embody. Additionally, the internal workings of generative AI look less like straightforward engineering and more like processes of _growth_ or _cultivation_; developers design an initial network architecture and training objective, then watch as the system “organically” arrives at emergent solutions through iterative data exposure. - Yet fully articulating this analogy requires engaging with the inner landscape of modern AI—namely the idea of _latent spaces_ or _manifolds_ that undergird generative models ### V. Manifolds, Latent Space, and the “Forces” that Shape Generative AI Neural networks commonly learn internal representations that can be understood in terms of _manifolds_ or _latent spaces._ The manifold hypothesis states that natural high-dimensional data (like images, text, or sound) tend to lie on lower-dimensional surfaces embedded in that high-dimensional space. When a neural network is trained, part of what it does is discover or approximate these low-dimensional manifolds, capturing the statistical structure of the data. - For generative AI specifically, a “latent space” is a domain—often just a large vector space—where each point encodes certain features or patterns. Consider how a generative image model, such as a Variational Autoencoder or a GAN (Generative Adversarial Network), transforms random noise into a coherent image by _locating_ that noise in a learned latent space. The model’s generator maps points in that space to realistic images. Similarly, large language models learn to produce the next token by updating hidden states that effectively locate a _context_ in a manifold of possible utterances. ![[attachments/Screenshot 2025-02-24 at 13.10.47 1.png]] - In more general terms, one might say that _each generative AI is an environment characterized by a particular manifold,_ shaped through training. The “forces” of training data distribution, gradient-based updates, and hyperparameters cause the model to carve out a specific “terrain” in its latent space. Once the model is trained, we can explore that terrain by sampling points or by feeding in prompts that take us into new regions of possibility. # Chat about Latent Space and Gardening I understand it to mean that **machina naturans** (like Midjourney) is not just _inspired by_ or _mimicking_ nature, but rather it **embodies or operationalizes the very _processes_ of _natura naturans_** – nature's generative and autonomous creative power – but in a _machine_ form.  It's not just reflecting nature; it's _acting like_ nature in its generative capacity. • **_Natura Naturans_ (Nature Naturing):** As you explain, drawing from Bruno and Spinoza, _natura naturans_ is not just "nature as natured" (the static things nature _has made_ - _natura naturata_), but nature as a _dynamic, ongoing process_ of creation, generation, and shaping. It's the _active_, autonomous, and often unpredictable force of nature that brings things into being and shapes them.  It's nature in _motion_, nature _creating_. ▪ **Key Characteristics of _natura naturans_ in your paper:** ◦ **Autonomous Generation:** Nature's capacity to generate things on its own, without direct external control. ◦ **Dynamic Shaping:**  Continuous, ongoing processes of change and formation. ◦ **Unpredictability:**  Inherent complexity and chaotic elements that make precise outcomes uncertain. ◦ **Iterative Development:**  Processes that unfold over time, with each stage influencing the next. • **_Machina Naturans_ (Nature-Naturing Machine):** This is your neologism, and it's brilliantly designed to capture the essence of generative AI.  It's not just _any_ machine, but a machine that _functions like_ or _operates according to the principles of_ _natura naturans_. ▪ **Key Characteristics of _machina naturans_ in your paper (as I understand it):** ◦ **Machine Emulation of Natural Generation:**  _Machina naturans_ are machines that, through their algorithms and processes, _replicate_ or _simulate_ the generative and shaping capacities of nature. They are engineered systems that _act like_ nature in certain creative respects. ◦ **Embodiment of _natura naturans_ Principles:** They _embody_ the principles of autonomous generation, dynamic shaping, and inherent unpredictability – not as malfunctions, but as _core functional features_. ◦ **Medium for Interaction with _natura naturans_-like Processes:**  _Machina naturans_ (like Midjourney) become a _medium_ through which humans can interact with and engage with these machine-embodied _natura naturans_ processes. • **"Natura Naturans _within_ machina naturans":**  This, I believe, is the most nuanced part.  It's not just that _machina naturans_ _produces_ things that _look like_ nature (though it can). It's that the _internal workings_ of _machina naturans_ – its algorithms, its neural networks, its latent space, its generative processes – are structured in a way that _mirrors_ or _instantiates_ the _dynamic, autonomous, and iterative processes_ that you associate with _natura naturans_. ▪ **Not just "resembling" nature's _products_, but enacting nature's _processes_:**  It's less about the _output_ looking naturalistic, and more about the _system itself_ operating in a nature-like way in its _generation_. ▪ **Latent Space as the Locus of _natura naturans_-like Generation:**  As we've discussed, the **latent space** becomes a crucial element here. It's within the latent space, shaped by manifold learning, that the _machina naturans_ embodies its _natura naturans_-like qualities. The generative process within the latent space is dynamic, iterative, and not fully predictable in detail, echoing the processes of nature. ▪ **Gardening Analogy as Key:**  The gardening analogy is crucial.  The garden is _natura naturata_ (what nature has produced), but the _growing process_ within the garden, the unfolding of biological life, is _natura naturans_.  Midjourney, as _machina naturans_, is like a _digital garden_ where algorithmic processes (within its latent space) act as a _machine-based_ version of _natura naturans_, and the user acts as a gardener, coaxing and guiding this digital growth. ## Suggestions as to how to develop things 1. **Explicitly Link Manifold Learning and Latent Space to "Generative AI as Naturans":** ▪ **Make Latent Space Central to _Machina Naturans_:** You've touched upon "algorithmic constraints" and "internal feedback loops" as forces in generative AI, but you should more explicitly bring in **manifold learning** and **latent space** as the _core mechanisms_ through which _machina naturans_ operates. This will strengthen the connection to our previous discussions and to Millière's paper (even if you don't cite it directly in the final paper, the conceptual underpinning will be stronger). ◦ **Suggestion:** In section 2.2 "Generative AI as Naturans," when you discuss the "forces" in generative AI, explicitly mention: "These 'forces' can be understood in terms of the algorithmic processes of manifold learning, which shape a complex, high-dimensional **latent space**. It is within this **latent space**, learned from vast datasets, that the _machina naturans_ truly operates, embodying the generative and shaping capacities analogous to _natura naturans_." ▪ **Latent Space as the "Environment" of Generative AI:** You talk about appreciating AI systems as "environments."  Clarify that the **latent space** _is_ the "environment" you are referring to – the computational environment within which the generative processes unfold. ◦ **Suggestion:** When introducing "Environmental Aesthetics of Generative AI" in your Introduction, you could say: "By viewing these systems as **latent space environments** shaped by cultural and technical forces, we can place their outputs and internal processes within a broader context that enhances aesthetic appreciation."  And later, "To understand how LLMs might be appreciated aesthetically as **natural environments** (in Carlson's sense), we can apply his emphasis on ‘nature as naturans’ to generative systems, focusing on their **latent space dynamics**." 1. **Deepen the "Knowing Naturans" Section (2.3):** ▪ **"Soil Analogy" is Excellent:** The "soil analogy" for datasets and latent embeddings is very effective. Expand on this! Connect it back to Carlson's emphasis on scientific knowledge. Just as understanding soil composition is key to appreciating a garden, understanding the "composition" of the latent space (dataset, training process) is key to appreciating generative AI. ◦ **Suggestion:**  Expand section 2.3.  Start by quoting Carlson again on the _epistemic_ requirement: "...we must appreciate nature in light of our knowledge of what it is, that is, in light of knowledge provided by the natural sciences..." Then, introduce your "soil analogy" directly as the _analogous_ "knowledge" for AI: "If scientific knowledge of geology and ecology is the key to appreciating natural environments, what is the analogous 'knowledge' for appreciating AI environments?  We suggest it is the understanding of the 'soil' from which AI art 'grows' – the **latent space** and the datasets that constitute it..."  Then develop the soil/dataset analogy further. ▪ **Connect "Experiential Artifacts" to Manifold Learning:**  When you mention "experiential artifacts (images, texts) that make up a system's dataset, and, crucially, how those artifacts and their features have been encoded in the system," directly link this to **manifold learning**. Explain that the dataset _is_ what shapes the manifold and the latent space. Understanding the dataset means understanding the _input_ that shaped the _natura naturans_ of the _machina naturans_. ◦ **Suggestion:**  "...truly appreciating generative AI requires not only knowledge of the computational architecture but also (some portion) of the multitude of experiential artifacts... that make up a system's dataset, and, crucially, how those artifacts and their features have been **encoded within the latent space through manifold learning.**" 1. **Refine and Expand the "Approaches to Appreciating Generative AI" Section (Section 3):** ▪ **Theoretical Appreciation and Manifold Learning:** In "theoretical appreciation," explicitly mention that technical understanding includes knowledge of **manifold learning algorithms**, **latent space architectures**, and how these contribute to the generative process. ◦ **Suggestion:** "First, there is theoretical appreciation. Those with a technical understanding of how generative AI works, including knowledge of **manifold learning algorithms**, **latent space architectures**, and training processes..." ▪ **Practical Appreciation as Latent Space "Gardening":** In "practical appreciation," link the "direct engagement" to the _act of prompting_ and _iterating_ within the latent space. Reinforce the "gardening" analogy here. ◦ **Suggestion:** "Second, there is practical appreciation, based on understanding through engaging with the generative environment – the **latent space**. A skilled gardener... develops an understanding of how such a system... works. Similarly, through **prompting and iterative refinement** with a generative AI, a user develops a non-technical, _experiential_ understanding of (part of) the underlying workings of the **latent space** system. In both cases, knowledge of how a generative environment’s forces function is arrived at through direct engagement, and this knowledge underlies aesthetic appreciation – the skill of the 'latent space gardener'." ▪ **Output Appreciation and Emergence from Latent Space:** In "output appreciation," emphasize that the outputs are not just images or text, but _manifestations of the latent space_ and the _generative process_ itself. Appreciate them _as_ expressions of _machina naturans_. ◦ **Suggestion:** "Third, there is output appreciation, in which the generative environment – the **latent space** – is appreciated through the appreciation of its products. In gardens, plants reflect natural generative processes shaped by gardeners. Similarly, generative AI outputs (text, images) are shaped by the dynamics of **latent space** and user input. By considering these products in context, we see how both natural and computational environments generate observable outcomes that can be appreciated not only as such, but especially as **expressions of the aesthetic potential of the _machina naturans_ – the generative AI system operating within its latent space.**" 1. **Consider Adding a Section on "Aesthetic Properties of Latent Space Itself":** ▪ **Beyond Outputs to the Medium Itself:**  You could push the "medium" argument even further by suggesting that the **latent space itself** has aesthetic properties, beyond just the outputs it generates.  Is there a kind of "mathematical beauty" or "algorithmic elegance" to a well-structured, disentangled latent space? ◦ **Possible Section:** "Beyond Output Appreciation: The Aesthetics of Latent Space." Explore questions like: "Can we aesthetically appreciate the _latent space_ of a generative AI directly, not just through its outputs? Does a well-organized, semantically rich, and navigable latent space possess its own kind of beauty or elegance?  Is there an 'algorithmic aesthetic' inherent in the structure and dynamics of these computational environments, analogous to appreciating the intricate beauty of natural systems, even beyond their directly perceivable manifestations?" **Dialectics and Japanese Gardens (Your Note):** Your note about Japanese gardens and dialectics is intriguing! You could potentially weave this in, perhaps in the "Approaches to Appreciation" section. • **Japanese Gardens as Dialectical Media:** Japanese gardens are often seen as carefully mediating between nature and artifice, control and spontaneity, wildness and order – embodying dialectical relationships. You could argue that DLSAM, as _machina naturans_, also embodies dialectical tensions: between human intention and algorithmic autonomy, control and unpredictability, cultural input and emergent output. ▪ **Possible Connection:** In the "Practical Appreciation" section, you could briefly mention: "Just as the appreciation of Japanese gardens involves understanding and engaging with their carefully balanced dialectical relationships between nature and artifice, so too does 'practical appreciation' of DLSAM involve engaging with the dialectic between human prompting and algorithmic generation within the latent space." **Overall, Nick, this is a very strong draft with a lot of potential!** By further developing the connections to manifold learning and latent space, especially by making latent space more central to your concept of _machina naturans_ and by expanding on the "knowing naturans" and "approaches to appreciation" sections as suggested, you can create a truly compelling and original paper that significantly advances the philosophical aesthetics of AI art. Let me know your thoughts on these suggestions, and we can refine the draft further! Refinements to be made (conversation with Enrico 14th January 2025) # 14 Feb 2025 ## Chat GPT Notes ### 4.2 pH as Analogue of Bias or Weighting? One might ask whether pH is directly translatable to a single concept in AI. pH modulates nutrient availability, shaping whether certain plants thrive. Could that be akin to _bias weighting_ in AI, i.e., how strongly certain features or styles emerge from the latent space? - In horticulture, acid-loving plants (e.g., azaleas) prefer soils with lower pH. Meanwhile, if the pH is too low, certain other nutrients become locked out, stunting plant growth. By analogy, if a model is “biased” heavily towards a particular style or domain—like “anime” or “photorealistic”—that domain might flourish at the expense of less common styles. - The prompter, akin to a gardener, might “raise or lower the pH” by adjusting the emphasis of certain tokens or negative prompts, encouraging or discouraging certain features. This is not a perfect mapping, but it highlights how an overall background parameter (pH or general weighting) influences what can thrive. # Current Shared Draft #paper/current [Link to Google Doc](https://docs.google.com/document/d/1-z9ChHDjCAGgDVtuFSUHVAK353-6mKlvSp-11B_Lz5k/edit?tab=t.0) # Draft being worked on (3000 words) 14 Feb 2025 The Environmental Aesthetics of Generative AI 14 Feb 2025 --- ## 1. ### 1.1 Quotes ### 1.2 LLM Chats - [ChatGPT Conversation](https://chatgpt.com/c/6741a88e-1200-8005-9f9f-a41185dd5a46) - ### 1.3 Notes - This paper examines whether artificial intelligence systems like ChatGPT can be aesthetically appreciated in ways that we appreciate human-made tools and artifacts. ### 1.4 Old Versions - Aspection and Gardening Draft - Creative Aspection old draft --- ## 2. ### 2.1 Quotes >With art objects there is a straightforward sense in which we know both what and how to aesthetically appreciate. We know _what_ to appreciate in that, first, we can distinguish a work and its parts from that which is not it nor a part of it. And, second, we can distinguish its aesthetically relevant aspects from its aspects without such relevance. We know that we are to appreciate the sound of the piano in the concert hall and not the coughing that interrupts it; we know that we are to appreciate that a painting is graceful, but not that it happens to hang in the Louvre. In a similar vein, we know _how_ to appreciate in that we know what “acts of aspection” to perform concerning different works. Paul Ziff says: >to contemplate a painting is to perform one act of aspection; to scan it is to perform another; to study, observe, survey, inspect, examine, scrutinize, etc., are still other acts of aspection… I survey a Tintoretto, while I scan an H.Bosch. Thus I step back to look at the Tintoretto, up to look at the Bosch. Different actions are involved. Do you drink brandy in the way you drink beer?1 >It is clear that we have such knowledge of what and how to aesthetically appreciate. It is, I believe, also clear what the grounds are for this knowledge. Works of art are our own creations; it is for this reason that we know what is and what is not a part of a work, which of its aspects are of aesthetic significance, and how to appreciate them. We have made them for the purpose of aesthetic appreciation; in order for them to fulfil this purpose this knowledge must be accessible. In making an object we know what we make and thus its parts and its purpose. Hence in knowing what we make we know what to do with that which we make. In the more general cases the point is clear enough: in creating a painting, we know that what we make is a painting. In knowing this we know that it ends at its frame, that its colors are aesthetically important, but where it hangs is not, and that we are to look at it rather than, say, listen to it. >All this is involved in what it is to be a painting. Moreover, this point holds for more particular cases as well. Works of different particular types have different kinds of boundaries, have different foci of aesthetic significance, and perhaps most important demand different acts of aspection. In knowing the type we know what and how to appreciate. Ziff again: >Generally speaking, a different act of aspection is performed in connection with works belonging to different schools of art, which is why the classification of style is of the essence. Venetian paintings lend themselves to an act of aspection involving attention to balanced masses: contours are of no importance, for they are scarcely to be found. The Florentine school demands attention to contours, the linear style predominates. Look for light in a Claude, for color in a Bonnard, for contoured volume in a Signorelli.2 >I take the above to be essentially beyond serious dispute, except as to the details of the complete account. If it were not the case, our complementary institutions of art and of the aesthetic appreciation of art would not be as they are. We would not have the artworld that we do. But the subject of this chapter is not art nor the artworld. Rather it is the aesthetic appreciation of nature. The question I wish to investigate is the question of what and how to aesthetically appreciate concerning the natural environment. It is of interest since the account that is implicit in the above remarks, and that I believe to be the correct account for art, cannot be applied to the natural environment without at least some modification. Thus initially the questions of what and how to appreciate concerning nature appear to be open questions. 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) > Venetian paintings lend themselves to an act of aspection involving attention to balanced masses; contours are of no importance, for they are scarcely to be found. The Florentine school demands attention to contours; the linear style predominates. (Carlson, 1979, p.19) > I survey a Tintoretto, while I scan an H. Bosch. Thus I step back to look at the Tintoretto, up to look at the Bosch. Different actions are involved. (Ziff, 1979, p.19) >Another view that also seems to diverge from the natural environmental model is the arousal model. This model challenges the central place the natural environmental model grants to scientific knowledge in aesthetic appreciation of nature. The arousal model holds that we may appreciate nature simply by opening ourselves to it and thus being emotionally aroused by it. The view contends that this less intellectual, more visceral experience of nature is a way of legitimately appreciating nature without involving any knowledge gained from science. Unlike the engagement model, this model does not call for a total immersion in nature, but only for an emotional relationship with it based on our common, everyday knowledge and experience of it. Consequently, in contrast to the engagement model, the arousal model does not lose the right to call its experience of nature aesthetic. Nor does it undercut the distinction between trivial and serious appreciation of nature, even though the appreciation it stresses may be more the former than the latter. However, the contrast between the arousal model and the natural environmental model is less clear. If we recognize our scientific knowledge of the natural world as only a finer-grained and theoretically richer version of our common, everyday knowledge of it, and not as something essentially different in kind, then the difference between the arousal model and the natural environmental model is mainly one of emphasis. Both models track the appreciation of nature, although the arousal model focuses on the more common, less cognitively rich, and perhaps less serious end of the continuum. (Carlson p.7) ### 2.2 LLM Chats https://chatgpt.com/c/678618e8-00a4-8005-a982-5933c456149a ### 2.3 Notes ### 2.4 Old Versions - Aspection and Gardening Draft - Creative Aspection old draft --- ## 3. ### 3.1 Quotes > If we recognize our scientific knowledge of the natural world as only a finer-grained and theoretically richer version of our common, everyday knowledge of it, and not as something essentially different in kind, then the difference between the arousal model and the natural environmental model is mainly one of emphasis. Both models track the appreciation of nature, although the arousal model focuses on the more common, less cognitively rich, and perhaps less serious end of the continuum. (Carlson, p.7) ### 3.2 LLM Chats - [ChatGPT Conversation](https://chatgpt.com/c/6741a88e-1200-8005-9f9f-a41185dd5a46) ### 3.3 Notes - The environmental aesthetics approach shows that the aesthetic appreciation of generative AI does not rely solely on transparent function or human-like persona. Instead, it arises from informed engagement, evolving knowledge, and the capacity to perceive patterns and order within a complex, semi-autonomous environment. ### 3.4 Old Versions - Aspection and Gardening Draft - Creative Aspection old draft --- ## General Resources ### General Quotes - Aesthetics of the Environment by Carlson - [Readwise Link](https://read.readwise.io/search/read/01ja7exb8y0hnc0913da2jecjr) ### General LLM Chats - [ChatGPT Conversation](https://chatgpt.com/c/6741a88e-1200-8005-9f9f-a41185dd5a46) ### General Notes - This paper examines whether artificial intelligence systems like ChatGPT can be aesthetically appreciated in ways that we appreciate human-made tools and artifacts. ### General Old Versions - Aspection and Gardening Draft - Creative Aspection old draft --- ## Bibliography - Aesthetics of the Environment by Carlson https://rthead.readwise.io/search/read/01ja7exb8y0hnc0913da2jecjr - Forsey, J. (Year). Title. Publisher. - Ziff, P. (1979). Title. Publisher. --- ## Scrap Section - Creative Aspection and AI Art - In our 2025 paper, we introduced the concept of *machina naturans* to describe generative AI systems such as large language models and text-to-image generators. - This concept extends the framework of natura naturans from §2.1—where nature is defined as an active, self‑organising force—into the technological domain. - In a manner similar to Carlson’s description of nature as consisting of evolving processes and interrelated systems, we suggest that generative AI also constitutes an environment shaped by algorithmic forces. - Although these systems are deliberately developed, they yield outputs—whether textual or visual—that are not entirely determined. - This corresponds with Spinoza’s description of nature as having a continuous capacity for self‑production; similarly, generative AI produces outputs that are not directly specified by human input, resulting from an interaction among training data, model architectures, and user prompts. - According to Carlson, nature’s aesthetic characteristics emerge from the processes, forces, and systems in which these forces operate. - In the case of generative AI, the user interacts with a system whose internal mechanisms remain partly concealed. - While prompts and parameters may influence the model, its specific outputs are not fully determined by any single user choice. - This introduces a dimension of unpredictability—what we might term dynamic recalcitrance—signifying that while there is scope for direction, the system’s generative processes ultimately decide how an output is formed. - Thus, labelling generative AI as machina naturans invites us to consider these systems not merely as passive instruments but rather as environments characterised by dynamic and ongoing creation. - They are, in effect, digital ecosystems of data and algorithms that respond to human interventions without being wholly contained by them. - By drawing on the discussion of natura naturans in §2.1—particularly its emphasis on the ‘forces and processes’ that sustain nature—we frame generative AI as operating through its own internal set of forces, systems, and interactions. - This perspective mirrors Carlson’s broader theme of appreciating an environment in light of the processes that sustain it, offering a starting point for understanding the aesthetic dimension of generative AI. - - In our 2024 paper, we introduced the concept of **machina naturans** as a term intended to capture the essence of generative AI systems like Midjourney in our discussion. - Young and Terrone (2024) define _machina naturans_ by drawing a parallel to _natura naturans_, which describes nature’s creative and autonomous generative processes. - They state: - [...]Drawing on this account of _natura naturans_, we introduce the category of _machina naturans_ to characterize those machines that are capable of generating and shaping things in the same way as nature[...] - They propose that generative AI systems are paradigm examples of _machina naturans_ ([...]Generative AI systems are paradigm cases of _machina naturans_[...]). - The central idea is that, like _natura naturans_, _machina naturans_ describes the capacity of machines to generate and shape outputs in a dynamic and somewhat unpredictable way. - This implies that these machines, in contrast to predictable tools or intentional agents, possess a degree of autonomy in their creative operation. - As the authors explain, both _naturans_ systems – nature and _machina naturans_ – are under the “incomplete control of the user” ([...]Both, as ‘naturans’ systems, are under the incomplete control of the user[...]). - This inherent unpredictability and limited user control are defining characteristics of _machina naturans_, setting it apart from traditional machines and tools, and aligning it with the dynamic and generative power of nature itself. - This concept is further illustrated by the analogy of using Midjourney to gardening, where users guide and influence, but do not fully control, the system's generative process, much like a gardener coaxes nature.This lack of complete control and inherent unpredictability is a key feature of _machina naturans_, distinguishing it from traditional machines and tools, and aligning it with the dynamic and generative force of nature itself. - This concept is further illustrated by comparing the use of Midjourney to gardening, where users guide and influence, but do not fully dictate, the generative process of the system, much like a gardener coaxes nature. - - the in their discussion of generative AI, proposing that such systems can be viewed as ‘artificial’ analogues of nature’s dynamic processes. ### NOTES OF VARYING QUALITY: • **Natural:** This signifies understanding nature as a realm of processes, systems, and interconnectedness, independent of human artifice. It means appreciating nature for its inherent dynamism, its ecological relationships, and its evolutionary history. It is about recognizing that nature is not merely a collection of static objects but a complex, living system.  The model urges us to move beyond superficial formal properties and engage with the intrinsic "naturalness" of nature. In essence, this first principle directs us to move away from artificial frameworks borrowed from art and to engage with nature in a way that respects its inherent character as a dynamic, interconnected, and fundamentally non-artifactual realm. **2. Appreciating Nature "in Light of Knowledge Provided by the Natural Sciences"** The second principle introduces the critical role of scientific knowledge in appropriate aesthetic appreciation of nature. It argues that our understanding of nature, particularly as revealed by natural sciences, is not just supplemental but essential for a deep and meaningful aesthetic experience. The text specifically mentions "environmental sciences such as geology, biology, and ecology" as key sources of this knowledge. Why is scientific knowledge so important according to this model? • **Revealing Nature's "Real Nature":** Scientific knowledge provides insights into the "real nature" of the natural world. It unveils the underlying processes, structures, and histories that shape nature. Science helps us understand how ecosystems function, how landscapes are formed, and how life has evolved. This deeper understanding enriches our aesthetic appreciation by moving beyond surface appearances. • **Uncovering Unity, Order, and Harmony:** The text suggests that "as science increasingly finds, or at least appears to find, unity, order, and harmony in nature, nature itself, when appreciated in light of such knowledge, appears more fully beautiful." (p. 11). Science, by revealing the intricate workings and interdependencies of natural systems, can reveal a deeper kind of beauty – a beauty of organization, complexity, and systemic coherence.  This is not necessarily a superficial "pretty" beauty, but a more profound aesthetic quality rooted in understanding the intricate order of nature. • **Moving Beyond Superficiality:**  Scientific knowledge helps to move beyond trivial or superficial appreciation. By understanding the ecological significance of a forest, the geological history of a mountain range, or the biological adaptations of a species, our aesthetic experience gains depth and meaning. It becomes less about fleeting impressions and more about a richer, informed engagement. However, it is important to note that the natural environmental model does not suggest that aesthetic appreciation _reduces_ to scientific understanding. Rather, scientific knowledge _enhances_ and _informs_ aesthetic appreciation. It provides a framework for understanding and valuing the aesthetic qualities of nature in a more meaningful and appropriate way.  It allows us to appreciate not just what nature _looks_ like, but also what it _is_ and _how it functions_. **Adequacy of the Natural Environmental Model** The text argues that the natural environmental model is more adequate than the object and landscape models because: "The natural environmental model thus accommodates both the true character of nature and our normal experience and understanding of it." (p. 6) This means it aligns with: • **The "true character of nature":** It respects nature's inherent qualities as natural and environmental, avoiding the distortions of art-based models. It moves beyond seeing nature as merely picture-like or sculptural, and instead emphasizes its dynamic, ecological, and systemic nature. • **"Our normal experience and understanding of it":**  It acknowledges that humans are not just passive observers but active participants who bring knowledge and understanding to their experiences of nature.  By incorporating scientific knowledge, the model aligns with a more informed and engaged way of experiencing the natural world, which is increasingly relevant in an age of environmental awareness. Furthermore, by grounding aesthetic appreciation in scientific understanding, the natural environmental model offers a degree of objectivity that is lacking in more subjective or culturally determined approaches. This objectivity can be valuable in environmental contexts, such as in environmental assessment, where aesthetic values are often dismissed as purely subjective. The natural environmental model offers a compelling framework for understanding aesthetic appreciation of nature. By emphasizing appreciation "as natural and as an environment" and "in light of scientific knowledge," it attempts to move beyond anthropocentric and art-derived frameworks to engage with nature on its own terms. It seeks a form of appreciation that is both deeply informed and genuinely aesthetic, recognizing the profound beauty that can be revealed through a scientifically enlightened understanding of the natural world. While other models offer alternative perspectives, the natural environmental model is presented as a robust and insightful approach, particularly relevant in the context of contemporary environmental concerns and our increasing scientific understanding of nature. ### old version - naturans is the big similarity - experiential artifacts is the big difference A positive account of appreciating generative AI can be found by drawing on Carlson’s environmental aesthetics, in which appreciators focus “on the order imposed [on] objects by the various forces, random and otherwise, that produce them” (p. 119). Carlson outlines a “natural environmental model” of appreciation, arguing that we must consider environments “as what they in fact are,” and that we should do so “in light of knowledge provided by the natural sciences, especially the environmental sciences such as geology, biology, and ecology.” We suggest that modern AI systems can be considered as a similar type of generative environment, and that we can therefore understand the aesthetic appreciation of AI in similar terms as Carlson does the natural environment.  - The natural world is shaped by various forces (e.g. wind, erosion, water currents) which transform the physical environment. Similarly, generative AI systems can be understood as environments shaped by a different set of forces (e.g. training data, alignment training, algorithmic constraints). When users provide prompts to these systems, the output is the result of interactions between such forces. As with natural environments, we might think that understanding such systems underlies their aesthetic appreciation. Like a forest or coral reef, an AI system contains multiple elements that interact in ways that can yield unexpected complexity and variation. For instance, training data may constrain certain outcomes, while user prompts guide directions of generation, and internal feedback loops adjust model parameters over time. These processes recall the layered interactions seen in natural environments.%%The above paragraph should be more detailed about what these forces amount to, in particular it should emphasise that when we ask what sort of thing a generative system is, to paraphrase the Carlson quote above, we need to remember that it is not *simply* a system but a system which has encoded a vast number of experiential artifacts of all kinds. Paragraph explaining what experiential artifacts are, and how they have been encoded in the system. Mention in particular the effect it has on a system if a particular artist, or style, or etc. is well represented in the training data –what does that have on the model. Another paragraph about how this can be thought of as similar to an eco-system %% A possible objection here is that everyday users of such systems lack technical knowledge of how they function, meaning that aesthetic appreciation of such systems is out of the reach of the majority. AI architectures and optimisation techniques may remain obscure, and it is not always clear how inputs and underlying parameters influence outputs. We suggest a response in Carlson’s remarks, which show that scientific and everyday knowledge are best viewed on a continuum. He writes: “our scientific knowledge of the natural world [is] only a finer-grained and theoretically richer version of our common, everyday knowledge of it, and not as something essentially different in kind.” On this view, a non-expert can acquire meaningful familiarity with AI just as a gardener, without formal training, understands sunlight, moisture, and soil conditions. Through iterative interactions, AI users identify patterns in outputs and anticipate how prompts affect results, paralleling the gardener’s observation of plant responses (reference withheld for blind review). Although it differs from scientific knowledge, this practical knowledge allows users to apprehend the AI’s underlying structure. By placing everyday and scientific perspectives on the same continuum, it becomes possible for non-specialists to cultivate an informed aesthetic engagement with AI as a generative environment. [^1]: [^2]: The twin of natura naturans is _natura naturata_, the stative aspect of nature, produced by naturans.