In ...Mantic Pixel Dream Girls, Frank Lantz points at the failure of approaching "game design with AI" from the perspective of modern game design:
I don’t think you can add a drop of machine learning, with its deep, psychedelic weirdness, to the clockwork cathedral of a modern AAA videogame, which is a carefully-arranged puzzle-box of precisely interlocking gears. I think you have to start from scratch, start from the drop of deep, psychedelic weirdness and grow something completely new.
I suspect we will have to start from scratch to a large degree, but there's a flavor of play that exists today that I suspect might have a similar flavor—I call it Big Gameplay.
Big Data is the idea that you’re receiving data faster than you can process it. Big Gameplay is when you’re creating and changing gameplay systems faster than anyone can understand them.
In Big Gameplay, the core of the game is figuring out enough of how the game works, before the systems inevitably shift again and the discovery process begins anew.
Player community interaction/feedback/discovery has always been a big part of competitive games. But competitive games are different in their striving. Their ideal outcome more similar to inventing a new version of basketball or football. The end goal is something more-or-less static, rather than a system where engaging with systemic change is the core interaction.
This distinction can be seen in Counter Strike 2, the 2023 sequel to the original from 2000, and the re-release of the original 2004 version of World of Warcraft as World of Warcraft: Classic in 2020.
Counter Strike, and now Counter Strike 2, are competitive games. To the layman the sequel is functionally identical to the original game released in 2000. Yes, there are numerous small tweaks, but when one of the headline changes is about backend server architecture, you know it’s gonna be pretty much exactly the same.
World of Warcraft: Classic, in contrast, showed just how divergent the path of modern World of Warcraft (WoW) has become from the original game. The path of modern WoW being much more exemplary of Big Gameplay.
In the original 2004 version of WoW, optimal play could be deterministically calculated on a piece of paper. In 2004, however, no one really did so (and when they tried, they did it poorly). The game was new and novel enough that just existing in the game's world was engaging and rewarding.
In contrast, modern WoW has become much more centered around optimal play. But now rather than having napkin-calculable gameplay systems, they're complex enough that optimal play is incalculable. Instead, players have resorted to creating gameplay simulations like SimulationCraft (and the cloud-hosted Raidbots) to simulate gameplay.
Increasing class synergy and the prevalence of proc-based combat modifiers have eroded the accuracy of traditional calculators that rely upon closed-form approximations to model very complex mechanics. The goal of this simulator is to close the accuracy gap while maintaining a performance level high enough to calculate relative stat weights to aid gear selection.
When you get a new piece of in-game gear that is somewhat similar to what you already have, there is no way to determine if it is an upgrade without machine-assisted help.
In 2004 you could say something like “critical strike is the most important stat for mages.” You can no longer make simple statements like this in the modern game. The only way you can optimally fight the machine is by simulating it with your own machine.
When they first made movie cameras you might have thought - hey you can point this thing at a play and it’s just like being in the audience. But eventually we figured out the important thing about movie cameras was the way film could be chopped up and re-arranged to form a new visual language, the grammar of cinema, out of which emerged a whole new dimension of synthetic dreams.
Big Gameplay is a non-traditional exploration of an ongoing system that neither the developer, nor the player, can fully understand on their own. A large part of “the game” is a collective poking and prodding at the gameplay systems to even understand what is going on.
The designers of WoW do not abdicate their responsibility for a creative direction for the game, but there is a sense that the approach they take is trending towards pushing a big ball of mud rather than polishing an intricate clockwork.
This engagement with a system that is larger than we can understand sounds a lot how we interact with AI. It’s people trying to jailbreak ChatGPT through convoluted conversational situations. It’s Ben Thompson trying to find Sydney within Bing’s chat. It’s Go players trying to understand why AlphaGo makes the moves that it does.
As Frank describes, we’re still in the “point a video camera at a theatre play” stages here, but I think these flavors of interaction point towards a potential path to walk down.
Games have a job to do, beyond entertain us and make money. Games have always been about figuring out the beauty of logic, the meaning of systems, the purpose of computers. In game design, we search through the possibility space of possibility spaces, looking for those ghostly shapes that connect things to each other, things to us, us to the world, and us to each other. Game design is hard.
Rather than the possibility space of a game being a defined and static thing we explore, what happens when we have machines that can constantly define engaging and shifting possibility spaces? What does it look like when play is a rough exploration of the shapes of those spaces, and just as we begin to feel secure in our understanding they shift again?
A rudimentary version of this is essentially what modern WoW has become, and it's very engaging! There are Discords full of people dissecting the massive variable set at their disposal. There's 13 minute explainer videos on minor ability-rotation changes that give you 4% damage increase. And baked into these discoveries is the understanding that it will likely be invalidated by a game patch very shortly!
But that is the joy of this sort of play. We all love a carefully-arranged puzzle box of precisely interlocking gears, but what does it look like when the game is figuring out what exactly is inside the puzzle box in the first place?