# Rewrite of Sections 2 and 3 of Environmental AI ## 2. [[The Natural Environmental]] Model 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) He does not regard the environment merely as a scenic collection of objects but as a dynamic 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 deeper textures, subtle variations, and evidence of geological history. 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. [NN-INTRO] Spinoza’s term *[[natura naturans]]* (nature in the act of producing) aligns with Carlson’s view of environments as active, evolving entities. Through partially autonomous processes, these natural systems exhibit ongoing developments that extend beyond mere human orchestration. This perspective reinforces Carlson’s emphasis on dynamic interactions, depicting nature as continuously ‘at work’ rather than as a fixed array of scenic elements. [/NN-INTRO] --- ## 3. [[Generative Environments]] We suggest that Carlson’s recommendations also hold for the [[aesthetic appreciation]] of generative AI: we should appreciate these systems 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. Olah’s depiction of neural networks stresses processes of growth rather than mechanical assembly. He points out that, although designers specify the overall architecture and training objective, the system itself grows in a way that resembles a living organism. Put differently, the network’s capabilities develop iteratively as it processes large amounts of data, so that the final patterns and functions emerge without being written out line by line. [MN-RESTRUCT] This resonates with Carlson’s ecological perspective: just as geological and ecological forces bring about certain formations without direct human orchestration, so too do these models evolve functionalities partly independent of any single engineer’s blueprint. The partial autonomy of this [[training process]] suggests that generative AI might usefully be regarded as “machina naturans,” echoing Spinoza’s term *natura naturans* (nature in the act of producing) rather than *natura naturata* (nature as already produced). Although the data are collected, cleaned, and curated by people, the resultant system exhibits configurations and behaviours that no designer can fully anticipate. In this sense, the ongoing, generative aspect of large AI models parallels self-producing processes in the natural world: they may be initially seeded and cultivated, but they continue to grow in ways that transcend straightforward commands or instructions. Conceiving of a large model as an evolving environment further illuminates its multifaceted character. Much like Carlson’s notion of learning about natural terrains, one cannot reduce such a model to a single function. Instead, it operates more like a dynamic landscape—users can move through various “microhabitats” of the model by applying different prompts, encountering diverse responses much as a naturalist explores distinct niches in a rainforest or coral reef. In this way, the system invites a form of open-ended exploration: repeated prompts reveal new or surprising areas of competence as well as intricate patterns of error. Such emergent features encourage an aesthetic mode of engagement that recognises ongoing processes rather than static design. Returning to Carlson’s insight, an informed aesthetic appreciation requires some familiarity with these algorithmic “forces.” Even if one is not a specialist in machine learning, noticing characteristic modes of response and testing prompts that yield consistent results cultivate an appreciation for the system’s internal organisation. Such experiential familiarity, like possessing basic geological or ecological knowledge, makes the system “intelligible” in Carlson’s sense: the more we understand the processes shaping its outputs, the richer our encounter with this generative environment becomes. 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 difference between a rainforest and a neural network, Carlson’s account acknowledges that natural environments can be subject to human intervention yet still exhibit spontaneous, ongoing developments. Generative AI meets that requirement to the extent that its training yields unexpected emergent properties. Like natural landscapes, these systems grow beyond any single blueprint, displaying partially autonomous qualities in their evolving functionalities. 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 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 to notice consistent stylistic patterns or recurrent biases in a model’s output. This practical familiarity provides a basis for more informed aesthetic appreciation, just as minimal geological awareness helps one perceive the shaping forces behind a coastal cliff. A further challenge is that generative AI changes rapidly, with new models or updates frequently appearing. Yet natural environments also evolve, sometimes dramatically, requiring observers to update their understanding. One can remain open to novel manifestations of generative AI in the same way an ecologist stays alert to environmental changes. Indeed, many users find aesthetic wonder in these ongoing transformations, as each system update might introduce unanticipated capabilities or reveal subtle shifts in the model’s “landscape.” Seeing the outputs themselves as belonging to an environment shifts the focus away from typical art-critical categories. Instead of judging an AI-generated poem solely by the criteria applied to human artworks, one can recognise how the text arises from both the user’s prompting and the system’s generative processes. The horticultural parallel again applies: a gardener can influence plant growth but never fully controls every branch or bloom. Similarly, a user can steer the model with prompts but may not precisely dictate the final outcome. Such a perspective acknowledges both human input and the partially autonomous unfolding of the system’s learned capacities. Carlson’s approach to natural environments also de-emphasises strict functional assessment in favour of understanding how ongoing processes create the scene. Likewise, generative AI’s aesthetic qualities can be illuminated by familiarity with how the system was trained, what data it encountered, and how prompts guide its responses. By attending to these factors, one can view AI outputs as emergent “fruits” of a dynamic environment rather than isolated, mechanistic products. While significant differences remain between actual natural ecosystems and any software-based creation, the key parallels—ongoing processes, partial autonomy, and iterative growth—suggest that Carlson’s framework can be fruitfully applied. If a primary lesson of environmental aesthetics is that deeper knowledge of underlying forces leads to a richer aesthetic experience, then the same lesson applies here. By seeing generative AI not as a mere functional tool or a fully authored agent but as a dynamic environment shaped by multiple forces, we gain a nuanced way of engaging aesthetically with these new artificial landscapes and their outputs. [/MN-RESTRUCT]