Quantum Storytelling with Generative AI
On a recent hike through Joshua Tree National Park I was scrambling off a smallish rock formation in this pristine place, in the middle of nowhere. I looked down to secure my footing and noted that the ground beneath my feet was staring back at me. Literally! It was a single little plastic toy googly eye in the sand of the desert. Of course I couldn’t help but think of the movie Everything Everywhere All At Once. For those not familiar, the movie portrayed multiple universes with multiple potential outcomes for its multiple characters. By some accounts, the googly eyes represent a joyful search for meaning across time and space. Naturally, I wanted to find meaning in this googly eye’s very existence at that time, in that space. And since I’m professionally preoccupied with multiverses and generative artificial intelligence — admittedly the buzziest of buzzwords over the past couple of years — I let my imagination loose. Now, I may be forcing a connection because I’m mired in these topics, but hear me out.
Starting almost eight years ago I began carving out a niche expertise around interactive entertainment. Specifically, branching narrative, choose-your-own-adventure style storytelling. The kind of experiences we would produce certainly felt generative to the audience. We were using AI-powered smart speakers to allow one’s voice to be what controlled the action, which allowed us to plunge each player into immersive worlds where they were one of the characters, alongside SpongeBob Squarepants, Arnold from Westworld, or the zombies of The Walking Dead. We had incredibly talented writers, sound designers, composers, and voice actors who brought these experiences to life. They would craft the overall narrative, build the crucial elements around the characters and their choices, and carve what we called the “happy path” — the most direct line from the entry point of the story through to the end of the arc. Choices made by the players that strayed from the happy path would be embraced through alternate scripted scenarios that ended with them back on the path after either a satisfying side story, or via what we called “fallback” responses. These responses were really just cleverly scripted error messages delivered in character with mostly appropriate context to keep players immersed in the world.
It was these fallback responses that first got me thinking about the promise of generative AI for interactive storytelling. I knew that replacing the human creators was a non-starter. What I really wanted to do was to supplement parts of the script that were creative but not necessarily critical to the narrative arc. The fallback seemed like a great place to start. We would sometimes write multiple fallback responses at every node that required player input. In our larger experiences, there could be dozens of decision points each with potentially dozens of fallback options. The more context we built into the fallback, the more time it took to write. For something like The Walking Dead Pathways with over five hours of final recorded audio, this translated to nearly a month’s worth of writing time. The body of written fallback responses could potentially be larger than the original script. User testing showed that, while fallbacks didn’t necessarily add to the story itself, they significantly contributed to a positive player experience. I thought this was a perfect use case to test my hypothesis. Unfortunately, back in 2019, the availability and affordability of large language models (LLMs) such as ChatGPT felt like years away. Turns out they were. But they’re here now.
Today, you can feed LLMs a prompt that includes details of a plot and characters and a few specific context cues and ask it to write 100 short response variations in the voice of one of your characters that are all versions of “are you sure you want to do that?” or “I didn’t catch what you just said” or “okay, let’s see where this path leads,” etc. Our experience shows that even when these fallbacks don’t make perfect sense, they mostly keep a conversation moving forward without taking players out of the scene. Beyond fallbacks, the rapid evolution of LLMs and their consumer interfaces like ChatGPT combined with the proliferation of generative AI tools like Midjourney, Craiyon, Runway, Stable Diffusion, etc., represent an opportunity to reimagine storytelling.
World-building and franchise development is difficult and resource intensive. Whether it’s Tolkien’s fictional middle-earth, or Tolstoy’s very real 19th-century Russia, the robust armature of these universes is critical to evolving narratives and characters across generations. Coming back to the googly eye for a moment. That eye, in the sand of the desert, was a construct that interrupted my understanding of my world in that instant. The eye represented a journey that led it to being in the desert at that very moment. It also represents the prospect of a new adventure for me beginning with the choices I make going forward. It’s a simple device with nearly infinite narrative possibilities. My imagination is really the only limiting factor to exploring these possibilities. And my imagination has limitations. I can conjure scenarios based on my admittedly narrow set of experiences and world-view. But there are many, many more possibilities that I cannot imagine. This is where I believe generative AI has the potential to expand how we think about creating and consuming stories. I think of it as quantum storytelling.
Back when I was in school I did the bare minimum when it came to math and science. I barely passed my core requirements and actively avoided those subjects in college. But since then, I’ve been pretty consumed with the intersection of science and humanities. I learned several programming languages, worked as a data analyst in public policy, and spent a fair bit of time trying to wrap my mind around math poetry (yep, that’s a thing). My latest brain hobby is quantum physics, which I’ve been reading about for years with varying degrees of success in unpacking even simple theories. I don’t let my layman status deter me from thinking about how certain aspects of quantum mechanics are relevant to understanding how the new era of AI can be applied to world-building, and the future of entertainment.
Forgive me in advance for skimming over some complex topics. I’m merely making an argument for how quantum mechanics can be useful for understanding a new paradigm in storytelling. In most simplistic terms, quantum mechanics is concerned with the behavior of particles and energy at a fundamental level. Whereas storytelling is concerned with the behavior of characters and events at a narrative level, typically within one or more narrative arcs. In quantum mechanics, particles can exist in multiple states simultaneously, while in traditional storytelling, characters can have multiple motivations and desires that can lead to multiple outcomes, but the ultimate outcome is fixed in the final narrative. The opportunity, as I see it, is to employ AI to remove the notion of fixed outcomes. Everything everywhere all at once, if you will.
I’m not suggesting that AI be given creative control. Just a bit of artistic license. Creators should absolutely define or originate a narrative universe, beget characters, construct plots, and determine the choices their characters make within scenarios to result in the arc of a given story. Quantum storytelling acknowledges these structures and follows parameters established by the storyteller for instructing AI to guide the behavior of narrative components leading to side plots, new arcs, and alternative endings. I have lots of ideas for how this relationship between creator and AI will play out (more on this later). But I thought I’d ask ChatGPT directly for some suggestions. Here are a few options it put forward:
Idea generation: Generative AI can be used to generate ideas for plotlines, characters, and settings that can be used as a starting point for the human author to build upon. The AI can draw on a vast database of existing stories to identify patterns and themes that might be relevant to the author’s vision.
Expansion and variation: Once the author has established the basic elements of the story, generative AI can be used to generate variations and expansions on those elements. For example, the AI could suggest different directions the plot could take, or different ways that a character could respond to a particular situation.
Editing and refinement: Generative AI can be used to analyze the author’s writing and provide suggestions for improving the story’s structure, pacing, and other elements. The AI could identify areas where the story is dragging, or where certain elements could be better integrated into the overall narrative.
Real-time feedback: As the author is writing, generative AI can provide real-time feedback on the story’s progress, identifying areas where the author might be deviating from the established structure or suggesting new directions to explore.¹
Not too bad. Right? Actually, it’s a pretty close, but also fairly narrow assessment of what’s possible. It lacks the creativity to fully imagine the relationship from the perspective of the creator. It also lacks currency. It doesn’t realize how rapidly things are moving in the world of AI. It can’t see itself in its own future. Good thing it has us humans to help it evolve. Which brings me to my main point. AI can fundamentally transform how stories are told and consumed as a subordinate partner to humans.
To riff on ChatGPT’s answer, the tools I’m envisioning bring AI and generative technologies to the frontline of creativity for both linear and interactive content, while delivering the following benefits:
- Greatly improve the process of creative ideation
- Speed up creative iteration within teams
- Rapidly generate suggestions and placeholders for styles, references, storyboards, and shot lists
- Reactively generate satisfying narratives for interactive experiences that respect and extend user actions
- Generate new characters / new behaviors / new scenarios within parameters established by the creator
I work with a group of talented creatives who consider language to be a powerful expression of human thought, but certainly not the only one. Our technology approach introduces AI at a foundational level working with traditional production pipelines and multiple other disciplines. Our long-term plan considers this language-based foundation as the jump-off point of next-generation processes that seek to empower creatives of different disciplines, such as actors, visual artists, and composers, to co-create content in digital environments, with creative outputs being translated to and from language by our tools.
Our ultimate goal is to create a tool suite that fully generates compelling narrative content — from real-time epic story arcs down to character dialog, in an interactive environment that is responsive to evolving contexts and user actions. The company we’re launching to achieve that goal is currently in stealth mode but watch this space.
And so it is with the proverbial googly eye firmly affixed to my forehead that we embark on achieving this vision of generating everything everywhere all at once. I hope you’ll follow us on this journey. And if you’re interested in contributing in any way, don’t hesitate to reach out. I’d love to hear from you.
¹ ChatGPT Feb 13 Version. Free Research Preview.
