Jam & Tea Studios is bringing improvisation to digital games with Generative AI.
November 14, 2024
It is the year of our lord 2024, and you cannot take two steps in any direction without encountering discussions around generative AI.
This is definitely true within the games industry. Over the last year and a half, the company I co-founded, Jam & Tea Studios, has been building games with Generative AI at the heart of the technology we’re developing. We’ve made some very exciting breakthrough discoveries, but also have had to wade through a lot of hype and overpromises as well as fear and uncertainty surrounding the technology.
Right now, the ratio of noise to signal in conversations around this space is…high, to put it mildly. But I’m excited to share a few key takeaways, learnings, and foundational pillars we’ve come to through our work, and give a peek into the kind of future we—as artists, designers, and engineers alike—are excited to build and take part in.
There are many folks who are excited about Generative AI’s ability to speed up existing pipelines for making games. While some AI tools being developed are more robust and helpful than others, overall, we’ve recognized that developer tools and efficiency gains are ultimately incremental improvements, much like other technology and software innovations that we’ve seen in the industry in the past.
Jam & Tea’s excitement and focus of exploration is in Generative AI’s potential to unlock brand new genres of play.
New kinds of games are multiplicative opportunities, allowing us to make games for new and underserved audiences.
We’re not just making the same games faster. We’re making brand new kinds of play.
Specifically, we’ve found AI tools have allowed us to build game worlds that give players more agency, and allow our environments and characters to be more responsive to players doing unexpected things in the world. In our early prototypes, we were able to bring a level of spontaneity and improvisation previously only found in tabletop gameplay into digital spaces.
I come from a theater and themed entertainment background, and specifically have years of experience in improv and immersive theatrical design. When describing the new kind of genre that generative AI unlocks, I’ve found a simple analogy to another form of media helps express this most clearly:
In TV and film, there are scripted sketch shows like "I Think You Should Leave". There are also improv shows like "Whose Line Is It Anyway." Both are vibrant modes of entertainment, and neither is better than the other. Similarly, a generative AI-powered game (where players can steer the story) will be a different experience than a scripted, linear story (with carefully crafted and scripted lines).
Some folks are focused on putting AI powered mechanics into traditional games, thinking “AI NPCs or mechanics are just better and will replace all nonAI powered ones.” But that’s not the case. In a scripted game, an NPC who has infinite things to say is not inherently interesting or meaningful. In fact, you likely want an NPC to only say specific things on purpose, because they’re in the game for a particular purpose to guide the player onto the next stage of the story or quest. In those cases, application of AI actually dilutes your game experience!
But if you design an experience around your NPCs and your entire game world being able to react in the moment to whatever your player says or does, and you create mechanics that lets your players take advantage of that increased freedom and flexibility—you have a game that’s built to embrace what your players invent through play.
We’ve seen this kind of play be an absolutely magical experience, where NPCs know truths from player backstories that were generated at the start of a session, and so are able weave that into their interactions. We’ve built characters that react and change their objectives on the fly based on what players do or what’s happening in an environment. We’ve seen early playtesters invent their own ways to tackle the big conflicts they’re presented with in the game and narratives we’ve built.
One of my favorite examples of this occurred in an earlier tech demo. I watched players that were faced with a werewolf attack on a village… and decided to bake cookies to befriend the werewolves. It was incredible seeing the game react by creating a baker NPC based on our system that combined traditional procedural generation techniques with AI generation. When the players talked to that NPC, the game improvised and crafted a quest and list of ingredients for the players to pursue on the fly, all while staying in fantasy and in-world based on our designers’ specifications of the setting. It was so exciting to see the players succeed in creatively turning a werewolf hunt into an unscripted bake sale!
The more I work with the technology, the more it has become abundantly clear to me:
As generative AI gets better and better at producing high volumes of content, the need and hunger for human ingenuity, creativity, and expressivity will only increase, not decrease.
One obvious use of generative AI is in churning out more stuff. Already today, you see images and text flooding the internet and diluting search results from showing non-generated results and outputs. But we know that infinite content, without meaning and purpose, is just infinite noise.
It turns out, writing naturalistic webpage copy or creating a believable stock photo is very different from composing an experience that resonates yet surprises, that feels fresh and new, yet inviting and familiar.
As a studio, we invest in our creative team because we know our collective creative voice will always be our key differentiator and competitive edge. Human ingenuity and storytelling imbues our games with meaning. Generative AI serves as but one component among the many tools we have to create games that offer people a shared experience that connects them.
After all, at the end of the day, if no human bothered to make the thing, why should we assume anyone would bother to consume it, much less love it?
While we’ve reached these learnings through working on [Project Emily], our codenamed flagship co-op multiplayer RPG game, we decided to take a subset of the underlying tech and design lessons learned to build a smaller, indie scale game. As a studio, we believe in getting things into the hands of players as soon as possible, and that’s why we built an entire game in just 6 months that shows off some of the possibilities in this space.
Retail Mage has you and up to 3 friends playing as customer support wizards working in a magical furniture store. Customers come in with all sorts of eccentric wants, and your job is to earn as many 5 star customer reviews as possible.
In the game, players can say (through speech to text) or write in what they want to say or do, and the game generates dialog options and determines if they’re successful or not based on their backstory, character traits, environment, and more. While the stated goal of the game is to earn 5 star reviews, I’ve seen players get playful and creative in how they tried to achieve this goal, or decide they want to try to burn down as much of the store as possible, or turn the game into a dating sim by flirting with every customer, to everything else in between.
In building Retail Mage, we’ve refined, stress tested, and improved so many systems and underlying components of our unique Emily Engine that powers the AI components of the game, and it’s been a blast seeing folks realize that they can do almost anything, and that so many of the rules of “how video games work” just doesn’t apply here.
Retail Mage is available on Steam now if you’d like to check it out. We’re also always excited to start a dialog with more developers interested or curious about this space. You can also follow along with our work at https://jamandtea.studio, find us on LinkedIn, or find us as @jamandteatime on your social platform of choice.
Read more about:
BlogsYou May Also Like
