I’ve spent years as an artist navigating the tension between authorial control and player agency. I texture characters, build shader networks, and encode symbolic meaning into every visual surface. But recently, I’ve been confronted with a more unsettling question: What happens when the world itself is no longer authored by human hands? Procedural content generation has evolved from a technical curiosity into an industry standard. Games like No Man’s Sky generate 18 quintillion planets algorithmically. Minecraft creates infinite worlds from seed values. Hades weaves handcrafted narrative through procedurally generated dungeons. And now, in 2025, AI tools like Unity Muse (Unity AI), Tripo AI, and Stability AI promise to automate asset creation entirely, generating characters, textures, and environments from text prompts in minutes. This isn’t just a workflow optimization. It’s a fundamental shift in how meaning is constructed in games.
When I design a character, I make thousands of micro-decisions. Eye shape and gaze direction convey vulnerability or menace. Texture breakup suggests wear patterns that tell a backstory. Asymmetric details communicate lived experience through scars, dirt, and weathering. Each choice encodes meaning. I’m not just creating a visual object; I’m building a symbolic container that players will decode through their interactions. But what happens when algorithms take over this process? When a designer types “medieval knight with weathered armor” into an AI prompt and receives a photorealistic 3D model in minutes, complete with PBR textures, rigging, and LODs, who controls the symbolic meaning embedded in that asset? The algorithm optimizes for statistical coherence, averaging millions of training images to produce outputs that look “correct” according to learned patterns. But it has no intentionality, no embodied understanding of what a scar means emotionally, no cultural knowledge of why certain color palettes evoke certain moods.
In 1967, Roland Barthes declared the “death of the author,” arguing that meaning is not fixed by the creator’s intention but emerges through the reader’s interpretation. The text becomes a “multi-dimensional space” where countless cultural references converge, and the reader, not the author, activates meaning. Procedural generation takes this further. In PCG, the algorithm becomes a co-author, or perhaps the primary author. The designer no longer crafts every tree, building, or character. Instead, they design systems: rule sets, constraints, and parameters that the algorithm executes. The designer fades into the background, becoming what I call a meta-author, someone who authors the conditions for authorship but does not directly author the final artifact. This raises profound questions. Who owns meaning when worlds are generated? If a player encounters a procedurally generated planet in No Man’s Sky, whose symbolic choices shaped that experience? The designer who wrote the algorithm, or the algorithm itself? Can algorithms encode intentionality? Intentionality implies consciousness, purpose, and cultural embeddedness. Algorithms operate through statistical patterns and mathematical functions. Can they “mean” anything in the semiotic sense?
Semiotics teaches us that signs derive meaning from their position within a system. A red door in a horror game signifies danger because of genre conventions, narrative context, and player expectations. The designer intends that meaning. But what happens when the door’s color, placement, and surrounding environment are procedurally generated? Consider Derek Yu’s Spelunky, a masterclass in constrained procedural generation. The game generates levels algorithmically, but every element (enemy placement, trap combinations, item distribution) follows strict design rules that encode intentional symbolic grammar. Spikes near ledges signal risk-for-reward, functioning as Peircean indexicality where proximity indicates danger. Shop placement creates moral dilemmas: steal and face consequences, or play fairly. Secret areas reward exploration through visual cues like cracks in walls or suspicious terrain. Yu doesn’t author every level, but he authors the symbolic vocabulary the algorithm uses. The procedural system becomes a semiotic engine, recombining meaningful units into novel but coherent experiences. Players can read these spaces because the underlying grammar remains stable across iterations.
Contrast this with No Man’s Sky, which generates planets using mathematical functions: Perlin noise for terrain, parametric equations for flora and fauna, color gradients for atmospheres. The result is visually impressive (18 quintillion unique planets) but semantically hollow. The algorithm optimizes for aesthetic variation, not symbolic meaning. A purple tree on Planet X and a blue tree on Planet Y are statistically different but semiotically identical. They don’t mean anything beyond “here is vegetation.” There’s no intentional symbolic architecture, no narrative embedded in the landscape, no emotional grammar encoded in the color palette. Players noticed this quickly. Early reviews praised the technical achievement but criticized the lack of meaningful discovery. Every planet felt like a reskin of the same underlying template. The algorithm generated difference but not significance. In Saussurean terms, the algorithm produces signifiers without stable signifieds. The sign system collapses into noise.
This distinction becomes even more troubling with AI asset generation. Tools like Unity Muse (Unity AI) allow designers to generate characters, textures, and environments from text prompts. This is revolutionary for production pipelines. What once took weeks now takes minutes. But can AI encode affect, or does it only mimic it? When I examine MidJourney-generated character portraits, I see stunning visual coherence but semantic emptiness. Eyes are symmetrical, static, emotionally neutral. Facial features follow idealized proportions, reflecting the algorithm’s bias toward “beauty.” Details are aesthetically coherent but narratively arbitrary. Why this scar? Why this expression? The images look like characters but don’t feel like characters because they lack the affective intentionality that human artists embed through lived experience, cultural knowledge, and emotional intuition. As I’ve argued in my work on ocular design, eyes function as primary vessels of affect. A character’s believability depends on micro-expressions, asymmetric gaze patterns, and contextually appropriate emotional cues. AI can simulate these visually but cannot encode them intentionally. The result is a kind of uncanny valley, not of realism, but of meaning.
Yet this doesn’t mean PCG and AI generation are inherently flawed. The solution lies in what I call hybrid authorship, using algorithmic systems as tools while retaining human intentionality at the design level. Henry Jenkins’ concept of “narrative architecture” is instructive here. Jenkins argues that games don’t need to tell stories the way films do. Instead, they can “stage narrative possibilities” through spatial design, allowing meaning to emerge through play. The designer becomes an architect of potential meanings rather than a dictator of fixed narratives. Applied to PCG, this means designing the semiotic vocabulary the system uses. In Spelunky, Derek Yu didn’t just code level generation; he coded a symbolic system where every element carries intentional meaning. Constraints aren’t limitations; they’re enablers of meaning. By restricting what the algorithm can generate, designers ensure that outputs remain within a coherent symbolic framework.
Hades exemplifies this approach brilliantly. Procedurally generated room layouts are constrained by handcrafted narrative beats, boss encounters, and character interactions. The algorithm provides variety, but the designers retain authorial control over symbolic meaning. When I play Hades, I experience genuine surprise at each run’s unique configuration, but I never lose the narrative thread. Zagreus’ relationships with other characters deepen through scripted dialogues that trigger regardless of procedural variation. The game feels authored because it is authored, just at a systemic level rather than a moment-to-moment level. The designers wrote the symbolic grammar; the algorithm executes it in novel permutations. This is hybrid authorship in action.
For AI asset generation, the workflow should preserve this principle. Use AI to generate base meshes, textures, or environments, then have human artists refine these outputs, encoding intentional symbolic details. The artist’s role shifts from creation to curation and encoding. This preserves efficiency gains while maintaining semantic depth. The algorithm handles repetitive labor; the artist handles meaning-making. When I think about my own practice, this resonates deeply. I already work with procedural systems in Unreal Engine (node-based shader graphs, parametric modeling tools, procedural texturing systems). These tools don’t replace my authorial intent; they amplify it. I design the system, set the parameters, and curate the outputs. The same logic should apply to AI-generated assets. The tool generates possibilities; I select and refine them according to the symbolic grammar I’m trying to construct.
Procedural generation doesn’t have to be a semiotic crisis. It can be a new semiotic regime, one where meaning emerges through the interplay of algorithmic systems, designer constraints, and player interpretation. But this requires us to rethink authorship itself. The designer is no longer the sole author of meaning. Instead, we become architects of possibility, encoders of symbolic grammar, designers of semiotic systems. The algorithm becomes a collaborator, executing our symbolic vocabulary in novel configurations. The player becomes the final interpreter, activating meaning through interaction. This triangulated authorship (designer, algorithm, player) mirrors Barthes’ vision of the text as a “multi-dimensional space.” But it also demands new design practices: semiotic auditing to regularly test procedural outputs for symbolic coherence, constraint design to define rule sets that encode intentional meaning, and hybrid workflows that combine algorithmic generation with human refinement.
I think about the regex systems I wrote about previously, where the period in Python shifts meaning depending on paradigmatic context. The same principle applies here. The algorithm is a symbolic operator, and its meaning depends on the system we design around it. When I write a procedural shader, the noise function doesn’t “mean” anything in isolation. It becomes meaningful only when I constrain its parameters, map its outputs to specific material properties, and embed it within a larger visual grammar. The same is true for procedural world generation. The algorithm is neutral; the meaning comes from the design decisions that frame its execution. This is why Spelunky succeeds where No Man’s Sky initially struggled. Yu understood that procedural generation requires more design intentionality, not less. He had to think systemically about how symbolic meaning would emerge across infinite permutations. That’s harder than designing a single, fixed level because it requires encoding meaning at an abstract, grammatical level.
The rise of AI tools intensifies this challenge but also offers new opportunities. If I can prompt an AI to generate a character, I can also prompt it to generate variations within a specific symbolic framework. “Generate a medieval knight, but make the armor asymmetrically damaged on the left side, with rust patterns suggesting coastal warfare, and eyes that convey exhaustion rather than heroism.” The more specific my prompt, the more I encode intentionality into the output. The AI becomes a collaborator in a semiotic conversation rather than a replacement for authorial intent. But this only works if I understand what I’m trying to encode in the first place. If I don’t have a clear symbolic grammar in mind, the AI will default to statistical averages, and I’ll get generic outputs. The tool amplifies my intentionality; it doesn’t create it.
This is why my background in semiotics, affect theory, and symbolic systems feels increasingly relevant. As procedural generation and AI tools become industry standards, designers who understand how meaning is constructed will have a critical advantage. We won’t just be operating tools; we’ll be designing the semiotic architectures those tools execute. We’ll be writing the algorithms that write the worlds. And in that recursive loop, authorship persists (transformed, but not erased). The algorithm writes the world. But we write the algorithm. That’s the paradox and the promise of procedural generation. It challenges traditional notions of authorship while opening new possibilities for systemic, emergent meaning-making. The key is to approach it with semiotic rigor, to treat it not as a shortcut but as a new medium with its own affordances and constraints.
As a game artist working at the intersection of technical pipelines and symbolic systems, I find this shift exhilarating and terrifying in equal measure. Exhilarating because it expands the expressive potential of games, allowing us to create vast, dynamic worlds that would be impossible to handcraft. Terrifying because it risks diluting authorial intent into statistical noise if we’re not careful. The future of game design depends on our ability to navigate this tension, to harness algorithmic power without surrendering symbolic depth. Procedural generation isn’t authorial abdication. It’s authorial evolution. And if we approach it with the same care and intentionality we bring to every texture, shader, and vertex position, it can become a powerful tool for encoding meaning at unprecedented scales.
References
Barthes, R. (1967). The Death of the Author. In Image-Music-Text. London: Fontana Press.
Farrokhi Maleki, M. & Zhao, R. (2024). Procedural Content Generation in Games: A Survey with Insights on Emerging LLM Integration. arXiv preprint arXiv:2410.15644.
Fukaya, K., Daylamani-Zad, D., & Agius, H. (2025). Intelligent Generation of Graphical Game Assets: A Conceptual Framework and Systematic Review of the State of the Art. ACM Computing Surveys, 57(5), 118, pp. 1-38. doi: 10.1145/3708499.
Jenkins, H. (2004). Game Design as Narrative Architecture. In K. Salen & E. Zimmerman (Eds.), The Game Design Reader: A Rules of Play Anthology (pp. 121-130). Cambridge, MA: MIT Press.
Peirce, C.S. (1998). The Essential Peirce: Selected Philosophical Writings, Volume 2 (1893–1913). Bloomington: Indiana University Press.
Saussure, F. de. (1916). Course in General Linguistics. Translated by Wade Baskin. New York: McGraw-Hill.
Togelius, J., Yannakakis, G.N., Stanley, K.O., & Browne, C. (2016). Procedural Content Generation in Games. Springer.
Tomkins, S.S. (1962). Affect Imagery Consciousness: Volume I: The Positive Affects. Springer Publishing Company.
