#llmtext #draft for the new AI and the [[Aesthetic Appreciation]] of [[Latent Space]] # Generative AI and The [[Aesthetic Appreciation]] of [[Latent Space]] ## Abstract We propose that generative AI systems themselves can be aesthetically appreciated by users in a manner analogous to [[the appreciation]] of natural environments. By extending [[Allen Carlson]]'s [[environmental model]] of [[aesthetic appreciation]] to the domain of generative AI art, we argue that users interact with AI systems like Midjourney as artificial environments—conceptualised here as *[[machina naturans]]*. Drawing parallels between environmental appreciation and the exploration of AI-generated outputs, we emphasise that knowledge—practical, technical, and art historical—plays a significant role in both contexts. This perspective aims to enhance [[the understanding]] of AI art appreciation and creation, highlighting [[the role]] of knowledge in engaging with these artificial environments. ## Introduction The advent of generative AI art challenges traditional notions of artistic creation and appreciation by blurring the distinctions between artist, tool, and artwork. Systems like Midjourney, which transform textual prompts into detailed images, necessitate a reconsideration of existing aesthetic frameworks. We propose that generative AI systems themselves can be aesthetically appreciated by users in a manner comparable to [[the appreciation]] of natural environments. By conceptualising AI systems as artificial environments—referred to here as *[[machina naturans]]*—we aim to extend Carlson's [[environmental model]] of [[aesthetic appreciation]] to the domain of generative AI art. Traditional models of art appreciation rely on clear boundaries and intentionality established by human creators. In contrast, AI-generated images result from processes that lack direct human authorship and operate within complex, abstract representations of data. This raises questions about how such art can be meaningfully engaged with and appreciated. We suggest that Carlson's environmental model offers a framework for understanding and engaging with both the AI systems and their outputs. By appreciating the AI system through its outputs, users engage with the generative process itself, akin to how one appreciates a natural environment through informed exploration. In a recent paper, Young and Terrone (2024) argue that creating images with generative AI systems like Midjourney is best understood through a comparison with gardening. They introduce the concept of *machina naturans* to characterise machines that autonomously generate and shape outputs in ways similar to nature's processes. We build upon this idea to further explore how users can aesthetically appreciate generative AI systems as artificial environments. ## Carlson's Environmental Aesthetics and Generative AI Allen Carlson's work on environmental aesthetics addresses the challenge of how natural environments can be aesthetically appreciated without the predefined boundaries and focal points found in traditional artworks. He observes that while art objects are human creations with clear demarcations and intentional aesthetic features, natural environments lack such explicit guidance for appreciation. Carlson proposes that knowledge—common-sense and scientific—about natural environments informs both what should be appreciated and how it should be appreciated (Carlson, 2000). In traditional art appreciation, familiarity with artistic traditions and styles allows for meaningful interpretation and engagement with artworks. Acts of aspection, such as surveying a Tintoretto or scanning a Bosch, are informed by knowledge of artistic conventions. In natural environments, Carlson suggests that knowledge about the environment guides the observer in identifying appropriate focal points and acts of aspection. For instance, appreciating a prairie landscape involves different sensory engagements compared to appreciating a dense forest. We propose that generative AI systems can be conceptualised as artificial environments, functioning as *machina naturans*. Young and Terrone (2024) introduce the term *machina naturans* to describe machines capable of generating and shaping outputs autonomously, akin to nature's *natura naturans*. They argue that creating images with Midjourney parallels a gardener's interaction with nature's autonomous generative processes: "Midjourney acts as machina naturans, where users engage with its dynamic recalcitrance to influence, but not fully control, the creative process" (Young & Terrone, 2024, p. 5). By appreciating the AI system through its outputs, users engage with the generative process itself, much like engaging with a natural environment through active exploration and sensory engagement. Interacting with AI involves unique acts of aspection analogous to those in natural environments. Users navigate the AI's latent space—the high-dimensional abstract representation where data features are encoded—by refining prompts and interpreting the AI's outputs. This engagement is dynamic and exploratory, requiring adaptability and responsiveness. Different regions of latent space may necessitate different prompting strategies, just as varied environments demand different approaches. An understanding of art history and technical aspects of art enables users to navigate the AI's latent space more effectively. Generative AI models like Midjourney are trained on vast datasets comprising millions of images from diverse sources, including artworks across different periods, cultures, styles, and mediums. During training, the AI identifies and encodes features such as brushstrokes, colour palettes, compositional techniques, and thematic content. These features are abstracted and stored within the latent space, forming the conceptual landscape through which the AI generates outputs. Knowledge about art allows users to craft prompts that guide the AI to explore specific regions of the latent space, influencing the outputs. By referencing artists or movements in their prompts, users can invoke specific styles or techniques, effectively engaging with the AI system's generative capabilities. Understanding the nature of latent space enhances the user's ability to interact with the AI meaningfully. The concept of latent space is central to understanding how users engage with generative AI systems. Latent space refers to an abstract, multidimensional space where data is encoded such that similar data points are positioned closer together. This space captures the underlying patterns and features learned from the training data, forming the landscape within which the AI operates. By interacting with the latent space, users effectively explore this hidden structure, positioning themselves within it through informed prompting. Knowledge informs both what to appreciate—the features and styles encoded within the AI—and how to appreciate them through appropriate acts of aspection. This mirrors Carlson's notion that appreciating nature involves engaging with it through informed acts guided by knowledge. Users who understand the AI's training data and generative processes can better appreciate the complexity and depth of the outputs. ## The Prompter as Gardener: Grappling with Dynamic Recalcitrance We suggest that users appreciate the AI system by engaging with its outputs, refining the analogy between prompters and gardeners. Young and Terrone (2024) propose that creating images with Midjourney is akin to gardening, where the gardener interacts with *natura naturans*—nature's autonomous generative processes. Similarly, the prompter engages with *machina naturans*, navigating its dynamic recalcitrance to influence, but not fully control, the creative process. Gardeners have a unique appreciation of nature that goes beyond the passive enjoyment of a casual observer. While anyone can appreciate nature by walking through a forest or admiring a landscape, gardeners engage with nature on a deeper level by actively working with its dynamic processes. They grapple with the inherent unpredictability and autonomy of *natura naturans*, shaping and cultivating their gardens while acknowledging nature's independent agency. This engagement involves grappling with what Young and Terrone (2024) describe as dynamic recalcitrance—the challenges and possibilities presented by working with an autonomous generative system. Gardening is not only about the end product—the garden or the plants within it—but also about the process of interacting with nature's generative forces. This process constitutes a form of aesthetic appreciation, where the gardener values both the creation and the act of creation. Similarly, prompters engage with generative AI systems by crafting prompts and iterating on outputs, navigating the AI's latent space. They grapple with the AI's dynamic recalcitrance, influencing but not fully controlling the creative process. The AI system's outputs are manifestations of its generative processes, and by engaging with these outputs, users appreciate the system itself. This interaction is not merely technical but constitutes an aesthetic engagement with the AI environment. Users adapt their strategies based on the AI's responses, learning from the outputs and refining their approach. As Young and Terrone (2024) note, "The prompter can iterate on Midjourney's products to get (or get closer to) the result for which they are aiming... Doing so, the gardener can exert skilful yet incomplete control over a piece of natura naturans which has a significant degree of autonomy in its own development" (p. 6). The parallels between gardeners and prompters highlight a form of appreciation that arises from engaging with an autonomous generative system. Both the gardener and the prompter collaborate with a system that exhibits dynamic recalcitrance, appreciating both the process and the products of their interaction. While the gardener cultivates plants and shapes a garden, the prompter explores the latent space and generates images, both appreciating the emergent results of their engagement. An analogue to appreciating nature by walking in it, without the active engagement of gardening, could be the passive consumption of AI-generated art without interacting with the AI system. In this case, users appreciate the outputs but do not engage with the generative process. Our focus, however, is on the deeper appreciation that comes from active engagement and collaboration with the AI system. ## A Deeper Understanding of AI Art Appreciation Applying Carlson's environmental model to generative AI art enhances the understanding of AI art appreciation by providing a framework that accounts for the unique characteristics of AI systems. This perspective acknowledges the collaborative dynamic between human users and AI systems, recognising the fusion of human intentionality with the AI's autonomous generative capacities. Understanding enables users to harness the AI's capabilities more effectively, leading to more intentional and nuanced creative outcomes. Informed engagement allows users to communicate their artistic intentions to the AI more effectively, navigating the latent space to achieve desired results. This informed interaction enriches the aesthetic experience, as users appreciate not only the outputs but also the underlying generative processes of the AI system. Moreover, this approach highlights the importance of knowledge—practical skills, technical understanding, and art historical awareness—in engaging with AI systems. By drawing parallels with environmental appreciation, we emphasise that engaging with AI art is not a passive experience but an active exploration informed by understanding and intentionality. ## Conclusion We have proposed that extending Carlson's environmental model of aesthetic appreciation to generative AI art provides a framework for understanding and engaging with this emerging form of creativity. By conceptualising AI systems as artificial environments—*machina naturans*—we suggest that generative AI systems themselves can be aesthetically appreciated by users in a manner comparable to appreciating natural environments. The appreciation arises not only from the outputs but also from the process of engaging with the AI system's dynamic recalcitrance. Drawing parallels between gardeners and prompters, we have highlighted how grappling with an autonomous generative system constitutes a form of aesthetic appreciation. This perspective underscores the role of knowledge in engaging with these artificial environments. Understanding enables users to navigate the latent space effectively, enhancing both the creation and appreciation of AI-generated art. Integrating Carlson's model enriches the understanding of AI art appreciation, facilitating a deeper, more informed engagement with generative AI systems and their outputs. --- **References** - Carlson, A. (2000). *Aesthetics and the Environment: The Appreciation of Nature, Art and Architecture*. Routledge. - Cooper, D. E. (2006). *A Philosophy of Gardens*. Oxford University Press. - Young, N., & Terrone, E. (2024). Growing the Image: Generative AI and the Medium of Gardening. *The Philosophical Quarterly*, 0(0), 1–10. [https://doi.org/10.1093/pq/pqae120](https://doi.org/10.1093/pq/pqae120)