# [[Introduction Generative]] AI systems like Midjourney and ChatGPT produce images and texts that can be aesthetically interesting, yet they do not fit neatly into traditional categories of art appreciation. Unlike a painting or poem created by a human artist with intentions and [[creative agency]], an AI-generated output emerges from a computational process lacking conscious design or purpose. This raises a philosophical problem: **How should we aesthetically appreciate AI-generated artefacts if not in the same way as traditional artworks?** The paper addresses this problem by drawing an analogy with [[environmental aesthetics]]. In [[environmental aesthetics]], natural environments are appreciated not as art objects but as environments – _as what they in fact are_ ( [[[Environmental Aesthetics]] ([[Stanford Encyclopedia]] of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=Alongside%20his%20critique%20of%20the,analogy%20between%20nature%20and%20art) ). We propose that AI [[generative systems]] can be viewed in a similar light: not as artists or [[mere tools]], but as _environments_ with their own internal generative dynamics. To develop this idea, we build on [[Allen Carlson]]’s influential theory of [[environmental aesthetics]]. Carlson argues that nature ought to be appreciated on its own terms – as a **unified, dynamic environment shaped by [[natural processes]]**, understood through appropriate [[scientific knowledge]] ( [[[Environmental Aesthetics]] ([[Stanford Encyclopedia]] of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=Alongside%20his%20critique%20of%20the,appreciation%20of%20art%20requires%20some) ) ( [[[Environmental Aesthetics]] (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=some%20knowledge%20of%20natural%20history%3A,In%20addition%20to%20this) ). We reinterpret this Carlsonian model using Spinoza’s metaphysical distinction between _natura naturans_ (nature as an active generative process) and _natura naturata_ (nature as the product of that process) ( [Baruch Spinoza (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/spinoza/#:~:text=There%20are%2C%20Spinoza%20insists%2C%20two,aspect%2C%20Natura%20naturata%2C%20%E2%80%9Cnatured%20Nature%E2%80%9D) ). This Spinozist lens highlights the unity of process and product in nature. We then extend the framework to AI: machine-learning systems are conceived as _machina naturans_ – artificial generative processes – and their outputs as _machina naturata_ – the resulting artefacts. Through this parallel, we show that **AI generative systems can be treated as artificial “environments” governed by internal principles and forces, and that their products can be appreciated by recognizing the unity between the generative process and the output**. The discussion will be structured around Carlson’s concept of unity in nature (§1), the Spinozist process–product reinterpretation (§2), an analogy between natural environments and AI systems (§3), and the implications for aesthetic appreciation, including a gardener–prompter analogy that suggests an engaged, interactive mode of appreciation (§4). In doing so, we aim to establish a coherent theoretical model for the aesthetics of generative AI, one that preserves analytic rigour while opening a new perspective on AI art. ## 1. Unity in Nature: Carlson’s Environmental Aesthetics Revisited **Carlson’s cognitive model of nature appreciation.** Carlson’s environmental aesthetics contends that _appropriate_ aesthetic appreciation of nature requires understanding nature as nature – _“as the particular sort of environment that it is”_ ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=Alongside%20his%20critique%20of%20the,analogy%20between%20nature%20and%20art) ). We do not appreciate a forest, coastline, or desert by treating it as we would a painting (framing it, isolating a static view); instead, we appreciate it _as an environment_, in light of knowledge about the natural processes that formed and sustain it ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=Carlson%20argued%20for%20it%20by,same%20reasoning%20to%20natural%20things) ). Carlson calls this the Natural Environmental Model (NEM) of appreciation. Just as knowledge of art history and style informs our appreciation of a painting, **scientific knowledge – ecology, geology, biology – informs our aesthetic engagement with natural environments** ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=Just%20as%20%E2%80%9Cserious%E2%80%9D%20or%20appropriate,nature%20can%20reveal%20the%20aesthetic) ). This knowledge reveals the environment as an integrated whole with aesthetic qualities grounded in its natural characteristics. For example, a casual observer might admire the beauty of a **coastal cliff** for its dramatic height or colors. But Carlson’s model suggests a deeper appreciation emerges when we consider what the cliff _actually is_: perhaps a stratified limestone formation shaped by erosion, part of a broader shoreline ecosystem. Understanding the cliff’s formation by sedimentation and its role in the coastal environment allows us to see unity in what might otherwise appear as disparate features – the cliff’s shape, the crashing waves at its base, the plant life on its edge – as all interconnected elements of one evolving natural scene. Such scientific cognizance, Carlson argues, enhances our aesthetic experience by situating what we perceive into a coherent narrative of natural forces ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=some%20knowledge%20of%20natural%20history%3A,In%20addition%20to%20this) ). The coastal cliff is not just a picturesque object but a segment of _nature’s ongoing processes_, appreciated for qualities (ruggedness, grandeur, harmonious form) that are tied to those processes. ([File:Beachy Head (2201).jpg - Wikimedia Commons](https://commons.wikimedia.org/wiki/File:Beachy_Head_\(2201\).jpg)) _Beachy Head, a coastal cliff in England, illustrates Carlson’s notion of unity in nature. Rather than viewing it as an isolated scenic object, we appreciate it as part of a dynamic environment – shaped by geological uplift and erosion, with its white chalk layers telling the history of ancient seas, and the surrounding windswept grassland and ocean together completing an interrelated natural scene. Knowledge of these processes and relationships helps us perceive the cliff environment as an integrated whole._ **Unity, autonomy, and natural processes.** A key insight from environmental aesthetics is that natural environments possess an **autonomous order**: they are not designed for viewers, yet they exhibit form, structure, and beauty arising from natural forces. Carlson often emphasizes that nature’s beauty is _objectively there_, to be discovered through the lenses of science and perceptual attention, rather than projected by us. The “unity” of a natural environment, in Carlson’s sense, means that every part is seen in relation to the whole. The aesthetic character of a coastal cliff, continuing our example, is not fully realized if one views the cliff in isolation as a static sculpture; it emerges when one appreciates the cliff **as part of a larger process** – the coastline shaped by tides and weather over millennia – and recognizes that the same natural laws govern the rocks, the sea, and the life around it. Carlson’s model thereby shifts aesthetic focus from isolated objects (a rock, a tree, a waterfall) to **the interaction and continuity among elements**, underlining that the environment is _“a seamless unity”_ of components and processes ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=environments%20in%20which%20they%20properly,in%20the%20object%20of%20appreciation) ). This unity is dynamic: natural beauty often lies in change, growth, and decay (the dramatic collapse of a cliff section, the seasonal cycle of a forest). By appreciating nature on its own terms, we accept that it is **not artifice but self-organizing reality**. Carlson’s approach, rooted in knowledge and _contemplation_, thus sets the stage for considering whether other non-art domains – like AI-generated content – might also be appreciated through an understanding of underlying generative processes rather than through traditional art frameworks. **Spinoza’s natura naturans and natura naturata.** To deepen the concept of unity in nature, we turn to Baruch Spinoza’s distinction between _natura naturans_ (“naturing Nature”) and _natura naturata_ (“natured Nature”) ( [Baruch Spinoza (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/spinoza/#:~:text=There%20are%2C%20Spinoza%20insists%2C%20two,aspect%2C%20Natura%20naturata%2C%20%E2%80%9Cnatured%20Nature%E2%80%9D) ). Spinoza used this distinction to describe two inseparable aspects of reality: (1) the active, productive aspect of nature (God or Nature **doing**: the endless generation of things through the laws of nature), and (2) the passive, product aspect (the **things produced**: the concrete world of particular forms and events that result from those laws) ( [Baruch Spinoza (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/spinoza/#:~:text=There%20are%2C%20Spinoza%20insists%2C%20two,aspect%2C%20Natura%20naturata%2C%20%E2%80%9Cnatured%20Nature%E2%80%9D) ). Spinoza’s insight is that these are not two different worlds, but two perspectives on the one unified substance that is **Nature**. _Natura naturans_ is nature as **process** – the continuous lawful activity that generates and sustains the cosmos. _Natura naturata_ is nature as **product** – the totality of particular things and states of affairs that have been generated ( [Baruch Spinoza (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/spinoza/#:~:text=There%20are%2C%20Spinoza%20insists%2C%20two,aspect%2C%20Natura%20naturata%2C%20%E2%80%9Cnatured%20Nature%E2%80%9D) ) ( [Baruch Spinoza (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/spinoza/#:~:text=is%20both%20Natura%20naturans%20and,Outside%20of%20Nature) ). Crucially, the two are inextricable: natura naturans is manifested in natura naturata, and natura naturata remains part of, and explicable by, natura naturans ( [Baruch Spinoza (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/spinoza/#:~:text=is%20both%20Natura%20naturans%20and,caused) ). Spinoza describes nature (or God) as an _“indivisible, eternal…substantial whole”_ that expresses itself in an infinity of particular things ( [Baruch Spinoza (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/spinoza/#:~:text=is%20both%20Natura%20naturans%20and,exist%20for%20any%20set%20purposes) ). In other words, **the unity of nature lies in the fact that everything we observe (nature naturata) is an expression of one underlying, generative reality (nature naturans)**. This Spinozist framework offers a philosophical reinforcement of Carlson’s idea of a unified, dynamic environment. When we appreciate a natural environment aesthetically, we are (often implicitly) appreciating both its current appearance (a configuration of natura naturata) _and_ the unseen processes that gave rise to that appearance (natura naturans). The coastal cliff’s sheer face and weathered texture are _beautiful_ in part because they are legible as the outcome of centuries of wind and water – the cliff _as process_ (erosion, sedimentation) and _as product_ (towering chalk face) cannot be fully separated in our aesthetic experience. Spinoza’s terminology makes explicit what Carlson’s model suggests: to fully appreciate nature’s beauty is to appreciate it **as a unity of the creative force and the created form**. The environment is not just a collection of objects that happen to be adjacent; it is a coherent whole because each object is what it is due to an underlying network of causes and conditions. Even when we momentarily focus on a static scene, we sense (with the help of scientific understanding) the _continuity_ linking that scene to past and ongoing natural activity. This view rejects treating nature as a series of static landscapes or as a human-designed park. Instead, it asks us to adopt what we might call a **process-informed gaze**: one sees any given natural vista as the present manifestation of natura naturans, much as Spinoza saw the entire universe as one substance in two modes. In summary, Carlson’s environmental aesthetics, interpreted through a Spinozist lens, yields a model in which **the aesthetic appreciation of nature is inherently dual-aspect**: one values the perceptible forms and one values the generative processes as two sides of the same coin. The unity of an environment is the fact that its appearance and its genesis are internally related. With this conceptual groundwork, we can now ask: Might **generative AI systems** exhibit a comparable duality of process and product? And if so, can Carlson’s model guide us in aesthetically appreciating AI outputs by treating them not as traditional art objects but in analogy to natural environments? ## 2. Artificial Generative Environments: _Machina Naturans_ and _Machina Naturata_ **Analogy between nature and AI systems.** We propose that advanced machine learning systems – specifically _generative_ models – can be illuminatingly seen as **artificial environments** with their own internal generative dynamics. Just as nature has _natura naturans_ (active nature) and _natura naturata_ (resultant nature), an AI system can be described in terms of _machina naturans_ (the machine as a generative process) and _machina naturata_ (the outputs it produces). This analogy initially sounds poetic, but it is grounded in technical and philosophical parallels. A generative AI (for example, the diffusion model behind Midjourney, or the transformer model behind ChatGPT) is not a passive tool that simply follows straightforward instructions. Rather, it operates via a complex learned model that, once trained, evolves outputs in ways that even its creators or users _do not fully specify in detail_. In this sense, the **AI’s operation has a degree of autonomy and dynamism reminiscent of natural processes**. When a user provides an input or prompt, the AI’s response is generated by the system’s internal state and rules, not hand-crafted by a human at that moment. The output emerges from the _machina naturans_: the system’s multilayered calculations shaped by the data distribution it absorbed during training. We can thus draw a parallel: _Nature_ as an autonomous generator of forms vs. _Machine_ as an autonomous (albeit designed) generator of forms. In both cases, the products can surprise and surpass any one agent’s explicit design – a point familiar in AI, where even the developers may be unsure exactly what image a diffusion model will produce from a novel prompt. To flesh out the analogy, consider how **Midjourney** (a text-to-image AI) responds to a prompt. The user might type a description (“an old tree by a lakeside at sunset”), but Midjourney’s output image is generated by sampling and transforming representations in a high-dimensional space learned from millions of training images. The user guides the generation conceptually, but the specific details – the particular shape of the tree, the exact hues of the sunset – are determined by the AI’s _internal generative process_. The AI here is _machina naturans_: it has stored complex statistical relationships (akin to laws of nature within the data domain of images), and it “natures” an image into being via those relationships. The final image is _machina naturata_: a particular outcome, analogous to a particular natural scene produced by underlying natural laws. Importantly, just as natura naturans and naturata are continuous in Spinoza’s view, the AI’s process and product are intimately linked – the output image is an **expression** of the model’s internal structure and training. If we froze the image as a static artifact divorced from how it was made, we would miss much of what defines it. The claim here is that **AI outputs, to be properly appreciated, should be seen in light of the generative mechanism that produced them**, much as natural environments are appreciated in light of underlying natural processes. **Internal dynamics of machine learning models.** To support this analogy, we need to understand how generative AI systems work and why it is apt to describe them as _environments shaped by internal forces and constraints_. Modern generative models (such as GANs, VAEs, or large language models) operate by creating and then exploring a _latent space_ – a mathematical space encoding features of their training data. During training, the AI system compresses patterns from the data into a multi-dimensional array of numbers (the latent representation) and learns rules to construct new data from these representations. The result of training is not a fixed set of outputs, but rather a **learned “world” of possible outputs**: an abstract space with its own structure, in which moving in one direction might add, say, more “tree-ness” to an image, or alter the style of a sentence. Researchers often note that these latent spaces are **highly structured**: they capture meaningful variations in the data (for instance, one axis might correspond to image brightness, another to the presence of buildings vs. open terrain, etc.), even though they are not explicitly labeled ([Text-to-image: latent diffusion models](https://nicd.org.uk/knowledge-hub/image-to-text-latent-diffusion-models#:~:text=Deep%20learning%20models%20utilise%20lower,to%20as%20the%20latent%20space)). In a sense, the AI has _discovered_ an underlying order in the training examples and can use it to generate new combinations. This is analogous to how nature’s processes impose an order (via physical laws) that shapes the forms we see. Indeed, we might say that **a trained generative model is an artificial world of sorts**: it has a kind of landscape (the latent space), natural laws (the model’s learned weights and activation functions), and dynamic processes (the generative algorithm that produces outputs). When the model creates an image or text, it is akin to a mini-evolution or simulation happening inside the computer: starting from some initial conditions (noise or a prompt), the internal process iteratively applies its “laws” until a final form emerges. For example, _diffusion models_ (used in image generators) literally simulate a kind of random diffusion and then reverse it: they start with random noise and gradually _coalesce_ an image by applying learned denoising steps. This can be compared to how, say, a crystal forms from a chaotic solution under natural physical laws – patterns emerge as the process iterates. In neural network terms, the **latent representation serves as the hidden generative environment**: instead of directly drawing pixel by pixel, the model operates in a compressed feature-space where conceptually related images cluster together. In this latent environment, a single prompt like “old tree by lakeside” might map to a region dense with variations of that theme, and the model “wanders” there to pick one realization. From a technical standpoint, these systems have **internal unity**: every output is a manifestation of a consistent underlying model. All images generated by a given version of Midjourney, no matter how different, are related by the fact that they are shaped by the same training data and network weights. Analogously, every natural landscape on Earth is different, yet all are outcomes of the same fundamental natural laws (physics, chemistry, biology) and planetary history. This parallel highlights why we call the AI’s process _machina naturans_. It is the generative capacity of the machine – the algorithm + training – that plays a role akin to “nature naturing.” And the specific output we see at a given time is _machina naturata_ – a particular “natured” product of that generative capacity. Just as one cannot fully appreciate a natural vista without some sense of the forces that made it, one cannot fully appreciate an AI-generated artifact without recognizing it as the endpoint of an algorithmic generative trajectory. It is important to clarify that likening AI models to nature is an **analogy** – we are not claiming AI literally has the richness or ontological status of natural ecosystems. There are obvious differences: AI models are human-designed (though their outcomes can surprise us), and their “laws” are mathematical abstractions, not physical necessities. However, from the standpoint of someone interacting with the AI or viewing its outputs, the experience can resemble encountering a complex self-contained system. The model presents us with outputs that we did not handcraft, and often we must _explore_ or _study_ the system’s behavior to understand the range and character of what it can produce. In practice, artists and users who work with generative AI often describe it in terms of _discovery_ or _exploration_ – much as one might explore a natural landscape – rather than straightforward creation. This exploratory relationship reinforces the idea that the AI behaves like an environment with its own internal logic, rather than a simple tool. **Midjourney and recalcitrant creativity.** A concrete illustration of the environment analogy is given by recent analyses of Midjourney. Midjourney is sometimes personified as an “artist,” but a more fitting characterization (argued by Young & Terrone, 2025) is as a **medium with a degree of recalcitrance** ([Growing the image: Generative AI and the medium of gardening | The Philosophical Quarterly | Oxford Academic](https://academic.oup.com/pq/article-abstract/75/1/310/7775353#:~:text=creating%20images%20with%20Midjourney%20parallels,shaped%20by%20its%20inherent%20unpredictability)). Users provide input, but the system may yield results that deviate from exact expectations – it has a kind of _stubborn autonomy_ in how it realizes prompts. This unpredictability is not random noise; it reflects the model’s learned biases and structure. In a Philosophical Quarterly discussion, Nick Young and Enrico Terrone compare creating images with Midjourney to **a gardener’s interaction with nature’s generative processes** ([Growing the image: Generative AI and the medium of gardening | The Philosophical Quarterly | Oxford Academic](https://academic.oup.com/pq/article-abstract/75/1/310/7775353#:~:text=unpredictability%20and%20the%20limited%20fine,shaped%20by%20its%20inherent%20unpredictability)). The gardener analogy (which we develop in the next section) encapsulates the idea that one works _with_ a semi-autonomous generative process rather than controlling every detail. Midjourney’s outputs thus exemplify _machina naturata_ that embodies the character of the _machina naturans_ within. Appreciating a Midjourney image, on this view, involves recognizing how the image’s style and content are informed by the AI’s training (for instance, the way Midjourney “chooses” certain artistic styles or composites elements in a hallmark manner). In short, the **process-product unity** in AI is observable: anyone who has prompted an AI multiple times notices the family resemblance in its outputs and learns to anticipate its “personality” – an emergent aesthetic signature of the system itself. This unity is precisely what grounds an aesthetic analogy to natural environments, where each vista is an expression of Earth’s unified natural order. At this juncture, we have equated _natura naturans_ with _machina naturans_ (the AI’s generative engine) and _natura naturata_ with _machina naturata_ (the outputs). The analogy suggests a framework for aesthetic appreciation: one should appreciate AI-generated artefacts not merely as isolated images or texts, but **as the visible presentations of an underlying generative world**. In the next section, we explore how this perspective can be translated into an aesthetic stance, borrowing further from Carlson’s model and extending it via the gardener–prompter analogy. We will ask: How should knowing the _machina naturans_ alter our appreciation of the _machina naturata_? And what might it mean to engage aesthetically with an AI _environment_ beyond passive contemplation? ## 3. From Contemplation to Engagement: Appreciating AI Artifacts as Environments **Cognitive appreciation of AI outputs.** If generative AI systems are analogous to natural environments, then a straightforward extension of Carlson’s **cognitive approach** to aesthetic appreciation would be: _to properly appreciate an AI-generated artifact, we should experience it as the kind of thing it is – namely, the product of a certain computational environment – and thus we should bring relevant knowledge of that generative system to bear on our experience_. In practical terms, this means that understanding the basics of how the AI works can enhance our aesthetic appreciation of its outputs, just as understanding geology or botany enhances our appreciation of a landscape ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=Carlson%20argued%20for%20it%20by,same%20reasoning%20to%20natural%20things) ). For example, consider an AI-generated poem. Without context, one might judge it by the standards of human-written poetry and find it perhaps oddly disjointed or impersonal. However, if one recognizes that the poem is produced by a language model (trained on vast text corpora and generating by predictive sequence), one might shift one’s appreciation: the apparently disjointed metaphors or lack of clear narrative can be seen as expressions of the model’s associative chains rather than failures of a human intention. One might admire certain turns of phrase as surprising emergent combinations, knowing they were _not_ crafted by a conscious poet but arose from the latent patterns in literature that the model absorbed. In short, the **appreciative focus moves from authorial intent to the interplay between prompt and process**. The aesthetic delight can come from observing how the machine’s “mind” (albeit not a mind in the conscious sense) manifests in the artifact. This is analogous to admiring a seashell not as a crafted jewel but as a natural form – seeing beauty in how sea currents and a mollusk’s growth have yielded the form, separate from any purpose. This cognitive mode of appreciation of AI aligns with Carlson’s requirement that we “experience it as what it is” ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=sciences%2C%20especially%20geology%2C%20biology%2C%20and,In%20addition%20to%20this) ). If a Midjourney image is experienced as if it were a deliberate painting by a human, one might misidentify its style or meaning. But if experienced as an output of a diffusion model, one might pay attention to different qualities: perhaps the peculiar ultra-detail in some areas and blurriness in others, or the way the image mixes photographic realism with incongruous elements – qualities that reflect how the AI recombines training examples. These might be seen as aesthetic signatures of the AI (just as the hexagonal patterns in a basalt cliff are signatures of cooling lava processes). Moreover, knowledge that the AI has no _intent_ can liberate the viewer from trying to interpret the image in the usual art-historical way (there is no intended message or emotion from an artist), and instead encourage a more open-ended, _contemplative_ appreciation of form, color, and the _idea_ behind the prompt’s realization. In environmental aesthetics, Carlson emphasized the role of **knowledge and imagination disciplined by knowledge** in correctly appreciating nature. By extension, appreciating AI art could become a new branch of **aesthetic cognitivism**: using knowledge of machine learning – e.g., understanding what a _latent space_ is or how an _encoder-decoder_ network operates – to reveal why an AI artifact has the qualities it does, and valuing it in light of that understanding. For instance, knowing that a language model selects likely next words might highlight for us the charm when it produces a slightly less likely, poetic turn of phrase – we appreciate the model’s _creative deviation_ from statistical expectation. This is comparable to how knowing the ecology of a forest can make us notice and appreciate the rare flower that deviates from the dominant green foliage. **The contemplative model and its limits.** Carlson’s model is often termed “contemplative” because it involves a kind of distanced observation: one **appreciates by beholding and understanding, rather than by physically intervening**. In a natural environment, this is appropriate – we do not (and ethically should not) rearrange wild environments just for aesthetic pleasure; rather, we immerse or attend. With AI artefacts, a contemplative stance would mean we **treat the artifact as something to be appreciated in situ, with an intellectual awareness of its source**. We might imagine an art exhibit of AI-generated images where viewers are provided with an explanation of the algorithm, guiding them to look for certain features. This could foster a richer appreciation than either viewing them as random images or pretending they were painted by a human. However, the analogy between AI and nature also suggests a further mode of appreciation – one that goes beyond pure contemplation: _active engagement_. In environmental aesthetics, some theorists (notably Arnold Berleant) have argued for an **engaged, participatory appreciation** of nature, emphasizing direct interaction, multisensory involvement, and even activities like hiking or gardening as ways of aesthetically engaging with the environment ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=While%20these%20views%20aimed%20to,Viewing%20the) ) ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=%E2%80%9Caesthetics%20of%20engagement%E2%80%9D%2C%20disinterested%20appreciation%2C,as%20possible%20the%20distance%20between) ). While Carlson himself focused on contemplation, the gardening analogy implicit in our discussion invites us to consider an engaged approach to AI aesthetics as well. **Gardener and Prompter: an analogy of engagement.** A gardener does not simply observe nature; she interacts with it – planting, pruning, guiding growth. Importantly, a good gardener works _with_ natural processes rather than against them. She understands the soil, climate, and biology (knowledge, as Carlson would endorse), but also gets her hands dirty: she participates in the generative process, nudging it towards desired outcomes. The **gardener–nature relationship** is often cited as a model of harmonious engagement: the gardener respects that nature has its own way, yet through practical involvement can co-create an environment (the garden) that is aesthetically pleasing. We propose that the **prompter–AI relationship** in generative art is analogous. When an artist or user engages with a system like Midjourney or GPT, they are in effect _cultivating_ the latent space. By crafting prompts, adjusting parameters, or selecting from multiple outputs, the user is guiding the AI’s _machina naturans_ in a particular direction, much as a gardener guides plant growth. This is not full control—far from it. A gardener cannot make a plant grow exactly in a predetermined shape without granting the plant any autonomy; likewise, a prompter cannot dictate every pixel or every word the AI will produce. But through skill, iterative experimentation, and adaptation to the AI’s “recalcitrance,” the human and machine together yield an artefact. The process itself can be deeply aesthetic: many AI artists describe a sense of _dialogue_ with the AI, a back-and-forth where they respond to the AI’s outputs (much like a gardener responds to how plants actually grow, adjusting plans accordingly). This gardener–prompter analogy suggests a form of appreciation that is **practical and experiential**. The aesthetic value may lie not only in the final output (the garden or the image) but also in the _experience of interaction_. Just as one might find aesthetic satisfaction in the act of gardening – feeling the earth, observing gradual growth, harmonizing with seasonal changes – one might find aesthetic satisfaction in the act of prompt-crafting and witnessing the AI’s generative responses. This moves aesthetic appreciation into the realm of **performance or play**: the user is actively participating in the creation and, simultaneously, appreciating the unfolding results. In philosophy of art, this is somewhat analogous to the appreciation of improvisational performance, where creation and appreciation coincide. In environmental terms, it resonates with ideas of _re-embedding humans in nature_, not as external observers but as cooperative agents. Moreover, the gardener analogy emphasizes **respect for the medium**. A gardener who imposes a rigid design without regard for soil and sunlight usually fails; similarly, a prompter quickly learns that working productively with AI requires understanding its capacities and limits (its version of “soil and sunlight”). Aesthetic appreciation here comes from recognizing the **unity of process and product** in a very intimate way: the prompter appreciates the final artifact partly because they remember or record the series of prompt adjustments and model outputs that led there. The artifact is a trace of a _process_ that the prompter engaged in. In this sense, _the process becomes part of the appreciated aesthetic whole_. This is an extension of Carlson’s unity (which was process understood intellectually) to an **experiential unity** – the person appreciates the unity by being _inside_ the process, not just reflecting on it from outside. **Applications of engagement model.** To ground this, consider an AI-generated musical composition where a human gradually refines it by giving feedback to the system. The end music might be pleasing, but the richer appreciation might come from knowing how an initial rough melody evolved through the AI’s variations and the user’s selections. Someone not privy to that process might only hear a somewhat quirky piece of music; the one who engaged in making it hears _in it_ the history of choices and surprises. This is analogous to how a cultivated garden might look like just a nice arrangement of plants to a visitor, but to the gardener it is imbued with the history of seasons, failures, and successes. In both cases, fuller aesthetic appreciation is available to the active participant. Thus, we might argue for a **participant’s aesthetics** of generative AI: an appreciation that values the _interaction_, not just the final output. This does not wholly replace the contemplative appreciation (one can certainly admire an AI image after the fact, as one can admire a garden without having grown it), but it adds a dimension that is unique to this medium. Finally, engaging with AI as one would with an environment highlights an ethic-aesthetic parallel: respect and care. In environmental aesthetics, some argue that appreciating nature’s processes fosters an ethical stance of care for the environment. Analogously, appreciating the AI’s generative nature might encourage creators to work with the AI in a way that respects its constraints (for example, not forcing it into domains it is bad at, or understanding its biases and not overtrusting it). While the “ethics of treating an AI well” is not the same as eco-ethics, there is a sense of _partnership_ that the gardening analogy captures. A gardener doesn’t simply exploit plants instrumentally; typically, they develop an affection and respect for the growth process. Similarly, creators often speak of “collaborating” with AI or being inspired by it, rather than just using it. This quasi-collaborative stance can enrich the aesthetic relationship: the work is appreciated as a **fusion of human and machine creativity**, much as a garden is appreciated as a fusion of human and nature’s creativity. ## 4. Unity of Process and Product in AI Aesthetics Having drawn out both a contemplative (knowledge-based) model and an engaged (participatory) model of appreciating AI outputs, we now return to the core insight uniting these approaches: the aesthetic significance of the **internal unity between generative process and generated product**. The central claim of an “environmental aesthetics of AI” is that an AI-generated artifact’s value and meaning emerge most fully when one appreciates it _in light of_ its origin in _machina naturans_. Whether one is a passive viewer armed with understanding, or an active prompter who helped midwife the artifact, in both cases one’s experience is enriched by recognizing that the artifact before us is not an isolated creation ex nihilo, but the **surface of a deeper process**. In Carlson’s terms, the unity of an environment means an object is appreciated _as environment_, not in isolation. In AI terms, the unity means an output is appreciated _as the expression of an AI model_, not as a standalone human-made artwork. For instance, an AI-generated painting of a “floating city” might initially strike us as imaginative or strange. If we consider it a human artwork, we might look for symbolism or intentional distortions. But if we consider it as _machina naturata_, we might instead marvel at how the AI’s training on perhaps thousands of city images and fantasy art led to this synthesis – we see in the painting echoes of other images, blended styles, a certain dreamlike quality characteristic of the AI. The painting is unified with its process: its peculiar aesthetic (say, hyper-detailed in texture yet oddly incoherent in architecture) _makes sense_ once we know it comes from an algorithm that does texture well but lacks genuine understanding of structural logic. Thus our appreciation shifts from “Is this a good painting?” to “Isn’t it fascinating how the AI’s style comes through here?” We appreciate the **machine’s aesthetic** as distinct from a human aesthetic. In doing so, we are in effect **appreciating the unity of process and product** – the painting is valued as a window into the AI’s nature. When one adopts this perspective, AI-generated art can attain a unique status in aesthetics. It is neither traditional art (tied to human expression) nor pure accident or randomness; it is something akin to a natural object. In environmental aesthetics, some theorists champion the idea of **“positive aesthetics”** – the notion that untouched nature is by default aesthetically good or at least interesting, because it is the product of nature’s harmonious order (even things that might seem ugly at first, like a swamp, reveal beauty when understood ecologically) ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=asserting%20that%20scientific%20understanding%20of,further%20developed%20and%20qualified%20the) ). An intriguing parallel might be suggested: perhaps **AI outputs, as products of a complex generative order, have a default aesthetic interest** – even the “mistakes” or strange outputs have a fascination because they reveal the character of the generative system. (One thinks of how people enjoy the bizarre errors in AI images or the nonsensical but amusing outputs of language models, as _glimpses of the AI’s workings_.) While it may be too strong to claim a blanket positive aesthetics for AI, the general point is that seeing process and product together tends to make us appreciate the product more, because we understand its place in a coherent whole. **Integrating contemplation and engagement.** Ideally, an aesthetic theory of generative AI will integrate the detached knowledge-based appreciation and the engaged creative exploration. These two are not mutually exclusive: a person can both know about the AI and actively play with it. In fact, the more one engages (like a gardener), the more one learns about the system’s behavior; conversely, the more one knows, the more fruitful and sensitive one’s engagement can be. Thus, the richest appreciation might come from a combination of **contemplation-in-action** – akin to how a seasoned gardener both revels in the act of gardening and contemplates the natural principles at work. Translating this to AI, we envision a practitioner who tweaks a prompt while simultaneously reflecting on why the model responded in a certain way, or who curates a set of outputs, recognizing patterns that relate back to the model’s training data characteristics. The _aesthetic experience_ here is multidimensional: visual or textual delight at the outputs, intellectual satisfaction in understanding their cause, and participatory enjoyment in being part of the cause. **Implications for art and aesthetics.** Recognizing the unity of AI process and product also has implications for debates about authorship and creativity. If we follow the environmental analogy strictly, then in a sense the _AI model itself_ (plus its training data) is analogous to “nature,” and the human prompter is analogous to a “visitor” or at best a “gardener.” This suggests that perhaps we should view AI-generated works less as authored artifacts and more as **discovered forms** or **co-creations**. Just as no one “authored” the Grand Canyon, yet it is stupendously beautiful and can be appreciated without an author, an AI artwork might be appreciated without pressing the question of _who_ made it. The focus shifts from creator to **origin**: not “who made this?” but “how did this come about?”. In aesthetics, this aligns with a move away from intentionalist interpretations towards form and genesis. It also might ease worries about AI art “lacking soul” – if we stop expecting an AI image to carry a human emotion or message and instead appreciate it like a natural phenomenon (perhaps evocative, but not intentionally expressive), we may find a valid place for it in our aesthetic experiences. To be clear, this perspective does **not** deny that humans play a role in AI art (after all, humans create the models and prompts). Rather, it reframes the human role as closer to that of a _facilitator_ or _curator_ of a process, instead of being the direct creative genius. A garden is a joint product of nature and gardener; an AI artwork is a joint product of machine and prompter. Aesthetic appreciation can accordingly be divided between admiration for the machine’s contribution (its complex, emergent order) and the human’s contribution (the intention in guiding it, the selection, the context given). In an environmental aesthetic analogy, this is like distinguishing appreciation of a wild national park from appreciation of a landscaped garden – both are valid, but one involves more evident human arrangement. Generative AI outputs arguably sit in between: not entirely wild, not entirely arranged. Finally, emphasizing process–product unity in AI art could influence artistic practices themselves. Artists might choose to **expose the process** to the audience – for instance, displaying the intermediate AI outputs, or interactive installations where viewers become prompters. This way the audience directly appreciates the interplay of _machina naturans_ and _machina naturata_. We already see nascent forms of this: some exhibitions include the prompt or even the training dataset alongside AI images, akin to museum displays explaining natural history alongside specimens. Such presentation echoes our thesis: the beauty of AI artifacts is amplified when their origin is made part of the aesthetic encounter. # Conclusion **Summary of the argument.** We have developed an analogy between environmental aesthetics and the aesthetics of generative AI, grounded in the idea of appreciating things _as what they are in fact_. Carlson taught us that a natural environment should be appreciated as an environment – _a unified whole informed by natural processes_ – rather than as an artwork or a mere collection of scenic parts ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=Alongside%20his%20critique%20of%20the,appreciation%20of%20art%20requires%20some) ) ( [Environmental Aesthetics (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/environmental-aesthetics/#:~:text=some%20knowledge%20of%20natural%20history%3A,In%20addition%20to%20this) ). Using Spinoza’s terminology, we interpreted this to mean that the beauty of nature lies in the unity of _natura naturans_ and _natura naturata_: one sees the product of nature as an expression of nature’s generative activity ( [Baruch Spinoza (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/spinoza/#:~:text=There%20are%2C%20Spinoza%20insists%2C%20two,aspect%2C%20Natura%20naturata%2C%20%E2%80%9Cnatured%20Nature%E2%80%9D) ) ( [Baruch Spinoza (Stanford Encyclopedia of Philosophy)](https://plato.stanford.edu/entries/spinoza/#:~:text=is%20both%20Natura%20naturans%20and,exist%20for%20any%20set%20purposes) ). Extending this framework to AI, we introduced the concepts of _machina naturans_ (the machine as generative process) and _machina naturata_ (the machine’s product). We argued that advanced generative AI systems can be viewed as **artificial, law-governed environments** – each output is analogous to a “natural” occurrence within the system, governed by the system’s trained parameters (its internal laws). Therefore, an AI output is best appreciated not as a standalone traditional artwork, but in relation to the AI’s generative context. This perspective reveals an intrinsic unity: **the aesthetic characteristics of the output are inseparable from the process that produced it**, just as a landscape’s beauty is tied to the natural forces that shaped it. We illustrated these ideas with the coastal cliff example (unity in nature through geology) and by examining how AI latent spaces and generative algorithms function (unity in AI through internal structure). Building on Carlson’s cognitive model, we suggested that **knowledge of the AI’s workings enhances appreciation**, much like scientific knowledge does for nature. We then expanded the analogy through the gardener–prompter comparison: showing that _engaging_ with AI’s generative process can be a mode of aesthetic appreciation analogous to gardening as an aesthetic practice. In doing so, we merged analytic philosophy of technology with environmental aesthetics to propose a new lens for AI art: one that values **process, interaction, and understanding** over authorship and intention. **Coherence and elegance of the framework.** One virtue of this framework is its coherence: it provides a unified way to think about various phenomena in AI aesthetics that might otherwise seem disconnected. For example, it explains why people often describe using generative AI in exploratory, discovery-laden terms – because they are in effect treating it like an environment to explore. It explains why the _lack_ of human intention in AI art need not be a deficit – because we transfer the source of order to the AI’s internal dynamics, much as we find beauty in seashells or sunsets with no human author. It also offers an elegant resolution to the debate of “is the AI the artist, or the prompter, or the programmer?” by sidestepping the notion of a single artist: the AI art is the result of an interactive unity of machine and human, process and prompt, akin to a collaborative emergence. This may not satisfy those who seek clear attributions, but aesthetically, it directs our attention to the _relationship_ rather than a lone creator. **Limitations and further questions.** Of course, the analogy is not perfect and raises further questions. For one, natural environments are valued in part because they are **real** and independent of us, whereas AI systems are engineered – does this diminish the authenticity of the “environmental” experience with AI? Possibly, but one could counter that once set in motion, a complex AI can surprise its creators, yielding _effective_ independence in its specific outputs. Another question is whether all AI outputs merit aesthetic appreciation simply by being products of a generative process. Discerning critics might say: some AI images are just poor or trivial. Our framework would respond: just as not every natural scene is stunning (some are bland or inhospitable), not every AI output is aesthetic. The point is not to romanticize AI outputs, but to claim that when they _are_ aesthetically pleasing or interesting, the source of that value often lies in the process-product unity (whether the viewer is aware of it or not). A particularly chaotic AI output might lack unity and thus fail aesthetically – paralleling how an incoherent environment (say, a polluted, disrupted landscape) can be aesthetically problematic because its natural unity is broken. In this sense, our model can accommodate evaluative distinctions: one could judge AI works by how well they express the underlying generative “nature” in a unified way. That could even lead to new aesthetic criteria specific to AI art (e.g., valuing images that clearly exhibit interesting structure of the model’s latent space, etc.). **Closing thoughts.** We conclude that _The Environmental Aesthetics of Generative AI_ is a fruitful framework that enriches our understanding of AI-driven art and creativity. It invites us to broaden our conception of what it means to appreciate something aesthetically: not all aesthetic objects need to be crafted by a human or appreciated as isolated, intention-laden works. Some can be appreciated the way we appreciate the wind shaping the dunes or the pattern of veins in a leaf – with a sense of wonder at the processes that unintentionally create something captivating. Generative AI, a product of human technology, ironically circles back to a state of nature-like autonomy in its outputs. By appreciating AI artworks as we do environments, we foster a stance of **respectful fascination**: we respect the AI’s complexity and limits, and we are fascinated by the interplay between our inputs and its autonomous development. In doing so, we may find a new kind of beauty – one that is **born from code and data, yet resonates with age-old aesthetic ideals of unity, harmony, and the sublime power of creation**.