Authors:
Samantha Stahlke
Four years ago, I completed my graduate research on AI in games. Today, I count myself among the creatives for whom AI has escalated from a friendly acronym to a four-letter word. Here, though, I do not wish to retread the long and storied road of AI research and its inextricable entwinement with digital games. Instead, let's talk about Clyde.
Insights
→ Generative AI is just one part of a rich history of AI and content generation in games.
→ We must not wield generative AI without critically considering the problems it presents and implications for the treatment of workers.
→ Reflecting on technological advancements can remind us of the magic in human creativity.
Clyde tells us something about technology and about ourselves. Clyde speaks volumes in his simplicity. Clyde is the orange ghost in Pac-Man. His behavior is randomized, unlike that of his compatriots Inky, Blinky, and Pinky. He exists at the intersection of the mindless and the mindful. In isolation, randomness is meaningless, but in contrast to his teammates and in combination with the human zealotry for pattern recognition, this becomes interesting. The absence of personality becomes personality. To players, Clyde is a he, perhaps a they or a she, but decidedly not an it. He is, quite literally, a ghost in the machine.
At the time of Pac-Man's release, Clyde joined a small but rapidly growing league of AI characters; one that quickly welcomed the aliens of Galaga, Pookas of Dig Dug, and shellcreepers of Mario Bros. Early game AI obeyed simple rules, but its dynamism elevated and diversified the challenges available to players. As the years passed, it became more complex. More-sophisticated techniques joined the fray, bringing us the eerily lifelike non-player characters of F.E.A.R. and the omniscient AI Director of difficulty and drama in Left 4 Dead.
Even today, we derive great experiences from basic game AI. The pattern-based bosses of Hollow Knight or Cuphead lack neither personality nor difficulty, and the idle roaming of a farmyard cat in Stardew Valley is charming in its simplicity. While these constructs are operated by code, they are brought to life by something very human: deliberate decisions in how they look, what they do, and how they interact. All of these are sparks of creativity that players appreciate and grow to love. Just a few lines of clever dialogue can bring an astounding level of depth to a character with little more to do than wander around.
Delusions of Grandeur
Today, the phrase game AI is more complicated. Games have a long history with machine learning and generative content, but the public discourse around AI in games exploded in 2021 and 2022 with the release of DALL-E and ChatGPT.
I went through an uncomfortable upheaval in my own relationship with the technology around that time. I previously adored generative AI and had noodled around with a few projects myself. But in 2022 I was an artist in the industry. News on generative AI quickly turned from awesome to sickening as I heard of seemingly endless ethical breaches and read the musings of evangelists all but screaming for the death of human creativity.
But here's the thing about human creativity: It's wonderful.
In The Art of Game Design, Jesse Schell speaks of "resonant themes" that permeate games to deliver "transcendent and transforming" experiences [1]; to "think [another] person's thoughts and feel their feelings" is "an integral part of gameplay." Inspiration, research, and iteration are essential in translating our experiences into games. When it is applied recklessly, generative AI jeopardizes the integrity of these processes.
Still, there are obvious benefits to be had (e.g., AI-assisted searches that make finding an old Jira ticket a bit less painful, automatically replying to an already useless email). Assuming ethically trained models, who are we to judge if some find them inspiring and helpful? Our primary focus is on the player, though, and most player-facing applications are dubious at best.
Consider voice acting, where AI-powered text-to-speech promises to replace human actors. Commercial interests threaten to eliminate a swath of the workforce or coerce individuals to sell their likeness (or simply steal it)—never mind the fever dream of posthumous performance. Ethically, it is a nightmare. But it is also one of many knives in the back of player experience.
Voice cloning, impressive as it is, still rests in a stilted and over-rehearsed uncanny valley. Even with perfect delivery, the result would forever remain incomplete. Appreciating good voice acting necessitates understanding that there's an actor behind the microphone who became part of a story they cared about. It's not just hearing words read a certain way, it's knowing someone felt something when they said them.
Those words themselves are also under threat, as the worst kinds of product managers pass on to their subordinates that most excruciating command: "Just use ChatGPT." Need a few quick ideas for a pitch? ChatGPT. Rough dialogue? ChatGPT. Catchy name for a feature? ChatGPT. Beg the gods of semantic prediction for 10 ideas, and ye shall be blessed with eight bullet points, impressive in their sheer mediocrity, spat out in less time than it takes to tell the junior writer they've been laid off.
The outcome of such endeavors is often hopelessly shallow, if not outright nonsense containing a few pearls of plagiarism. However, even the most impressive pieces—those otherwise indistinguishable from human wit—can provide only a pale imitation of meaning. Great writing exists in the context of its creation. Its power comes not only from what lies on the page but from knowing that there exists a voice, a lived experience, that brought those words to the fore. Within the fundamentally emotional domain of art, the perception of soul is of the utmost importance.
This brings us to visual art, ostensibly the most widely discussed (and controversial) application of generative AI. Visual media has served as both a showcase of the technology's most immediately impressive results and a lightning rod for its flaws.
Generative AI systems rob us of the joy in our work or the ability to work at all, before going on to rob players of the intent and meaning behind the games.
Beyond the obvious perils of derivativeness and repetition, generated content often loses intentional details, while others are added seemingly at random, with neither having deep reasoning in their presence or connection. In the context of games, maintaining stylistic consistency among hundreds or thousands of assets is nigh impossible (Figures 1 and 2). And output is often not quite right in a disturbingly not-remotely-human way: A human artist makes mistakes, but they won't be swirling a seventh finger into an unholy fractal before blending it into the side of a lamppost.
 | Figure 1. Left column: Sprites drawn for Cook Serve Forever (CSF). Right grid: Assets generated with Adobe Firefly struggle in style and perspective. The concept of sliced bread also proves challenging. |
 | Figure 2. Top: CSF's individually painted ingredients combine into burgers, temaki, and doughnuts. Bottom: Firefly-generated foods marry initial appeal with off-putting errors. |
It is possible that we find ourselves on an interminable technological plateau of diminishing returns. But even if these challenges vaporize, AI-generated work builds on a largely meaningless foundation. When dealing with human creations, players can ask a question like "Why does this character look this way?" and know there is an answer somewhere, whether based in lore, usability, or an artistic whim. When that answer is discoverable, it is a delight. But even the mere fact of its existence makes a game more interesting. With generative AI, the only viable answer is something akin to "because this is what the black box of linear algebra pulled from the statistical cloud associated with this prompt." And here, less is not more.
The Greatest Generation
It is bemusing that much of the discussion around generative AI in games seemingly ignores the fact that in some practical sense it already exists. It's called procedural generation. The preexistence of infinite game worlds is the 40-year-old elephant in the room.
Well-executed procedural content generation (PCG) expands creative possibilities while enhancing player experience. Take the cavalcade of sandbox survival games epitomized by Minecraft, with a new world to explore and transform in every playthrough (Figure 3). Or the genre ignited by 1980s classic Rogue that includes modern gems like The Binding of Isaac, Hades, and Balatro, where players learn more with every attempt at the same core journey. Or experimental designs like Façade, whose developers thoughtfully crafted a procedural story system to test the boundaries of narrative engagement.
 | Figure 3. Minecraft is possibly the most renowned example of PCG in games. Over time, a cavalcade of biomes, structures, and critters have sprung forth to populate its endless worlds. |
Procedural content can boost replayability and make individual experiences more unique. It also intertwines beautifully with the concept of emergence—the spawning of complex behaviors from the cumulative interaction of simple rules—just ask any Dwarf Fortress player.
Naturally, PCG is vulnerable to some of the same pitfalls as generative AI. It can be buggy. Its output can feel boring. With public frustration repeating the sentiment that games with increasingly massive scope ultimately emerge feeling rushed or unfinished, caution should be exercised with any approach that bears the siren's call of unending content. But another advantage of PCG systems over generative AI tools here is that they are more readily adaptable to feedback, instead of relying on muddling with prompts or tacking on an extra GPU and hoping for the best.
If you want a quick set of character portraits, it is faster and easier to snag them from a Stable Diffusion wrapper than to program a bespoke system working with assets created by a human artist. But the PCG system will not be a black box. It can be more easily integrated with other game systems. It will encapsulate meaningful decisions from a talented team of humans with creative intent. It will not rely on a model that uses the energy of a small town to train on data harvested without permission. And characters will always have the desired number of fingers.
The Heart of the Matter
To love something, you must engage with it, and here a cruel deception comes to pass with generative AI. You may be captivated by a snippet of prose, only to find yourself befuddled by little factual inclusions that aren't so factual. You may be drawn into the eyes of a character, only to grimace at a Gigeresque mutation of melted teeth. Or maybe you just realize that whatever you're consuming is devoid of the heart that permeates the things you love. To engage with this content is to reach for cherries and find yourself chewing on a mouthful of red gravel. Maybe the first few times, you're impressed by the illusion, but that feeling quickly changes into disappointment and resentment.
Here, we return to the critical understanding of art as a medium of emotion. Tolstoy said of art and empathy that "it is on this capacity of people to be affected by the feelings of others that the action of art is founded" [2]. As with Schell's "integral part of gameplay," the emotions of both artist and audience are essential; to evoke a meaningful experience, there must be meaning in its creation.
Much of the rhetoric surrounding these systems is so perversely focused on efficiency that it disregards the value in the creative process, in the human expression of feeling. Who, really, was clamoring for the automation of concept art? These systems rob us of the joy in our work or else the ability to work at all, before going on to rob players of the intent and meaning behind the games they play.
The sad reality is that the current discussion is dominated by a chorus of self-described prompt engineers insisting that typing "fantasy character anime girl trending on artstation 4k" into Midjourney makes them some sort of 22nd-century da Vinci. It is polluted by the exact same crowd that flooded the Internet with their cigar-smoking chimpanzees and hexagonal Twitter profile frames three years ago. Regrettably, much of the commercial interest has followed suit. The buzzword cycle must be fed, and generative AI is just the latest thing to fill the void infiltrated by NFT games and the metaverse.
In spite of the real and justifiable fear and anger surrounding this technology, this is not meant to be a condemnation but a cautionary tale. A reminder that not every hammer we create must in turn make every aspect of our culture a nail to be driven down. That we should treat these things as tools to be used thoughtfully and responsibly.
Even among prominent researchers, there is plenty of discussion about AI's flaws, besides the sensationalist bluster around paper clips and gray goo. Turing Award recipient Yann LeCun, one of few bearing the AI godfather moniker, has noted that generative AI lacks the "essence" of emotional communication necessary for art forms like improvised music [3]. Disclosure in its use is also critical. LeCun notes that "truthfulness is an essential part of the artistic experience."
Certainly there are worthwhile experiences to be created here. The weirdness of latent-space interpolation can create mesmerizing works that do not imitate existing art but embrace something new. Generative AI can be used intentionally to comment on technology, society, misinformation, labor, and the nature of art. There is value in using AI tools to make media more equitable. Generative systems, trained and applied ethically, can help empower independent developers. AI screen readers, described video, and captioning can contribute to more universal media accessibility.
It is easy to write cynical quips condemning the state of the AI industrial complex. But maybe what we really need is a love letter to the people who help fill our lives with meaning. To the countless creatives who pour their hearts out so that we can share in their dreams, ideas, and hopes, and derive joy and meaning from the experiences they create. The humanity in art is what makes it beautiful. And people deserve beauty in the games they play.
References
1. Schell, J. The Art of Game Design: A Book of Lenses, 2nd ed. CRC Press, 2015.
2. Tolstoy, L. What Is Art? Henry Altemus Company, 1898.
3. Levy, S. How not to be stupid about AI, with Yann LeCun. Wired. Dec. 22, 2023; https://www.wired.com/story/artificial-intelligence-meta-yann-lecun-interview/
Author
Samantha Stahlke is a researcher and instructor in games and HCI at Ontario Tech University. She is also an artist and developer and has worked as a UX consultant. She holds a graduate degree in computer science and coauthored The Game Designer's Playbook: An Introduction to Game Interaction Design. [email protected]
Copyright 2025 held by owner/author
The Digital Library is published by the Association for Computing Machinery. Copyright © 2025 ACM, Inc.
Post Comment
No Comments Found