Last week we talked about this:
That guard helped me kill the dragon in the background five seconds earlier.
So what’s going on here?
The Sandbox Narrative Conflict
Narrative in games is still a trigger and state-based affair.
Let’s say I have two NPCs — Bob and Mary. If I talk to Bob before Mary, it advances a game state and Mary gives me new dialog. If I talk to Mary before Bob, it advances a different game state and Bob gives me a different dialog than he would if I talked to him first. If I walk within 10 meters of Bob (specifically a bounded area that encompasses Bob), he will walk over to me. While if I’m wearing a certain hat, Mary will attack.
These things are typically implemented by designers in a high-level scripting language, after coders have added the functionality to the game editor. Note that quests are just game states which have been flagged to show up in the journal, with all the surrounding icons and text.
At a rough count, Skyrim has about 1200 named NPCS. Someone has to dress them, give them an inventory, dialog, whatever. NPCs like “Falkreath Guard” would be cloned to save a lot of time. They’re given stock phrases when you click on them — things that make sense in any situation. If the player has a particular game state set (he’s the Archmage, for example) there will be additional stock phrases added to the pool.
This is how things worked in the original Fallout. And indeed, pretty much every RPG from the 1990s. That’s right, our fundamental paradigm for quest and world-building hasn’t drastically changed for twenty years.
The problem is that as world sizes grow, and the number of actions the character can perform grows, the number of possible connections between every NPC and every other NPC, and the player, also gets very large. And designers are still mostly doing this by hand.
There is a fundamental conflict between traditional RPG narrative and a sandbox open world:
As world sizes, NPC count, and player freedom increase, the percentage of the player’s actions which register as meaningful to the game necessarily decrease.
In short, the semiotics become increasingly dissonant with the gameplay because the semiotics of go-anywhere-do-anything sandbox RPGs is about choice and permanence.
This is just a function of how games are made. Skyrim was a huge world, and it took about 90 developers and (supposedly) $85 Million. MMOs have the added constraint of multiplayer completely breaking traditional quest logic. Which is why phasing, and similar work-arounds were invented. Basically, MMOs have devolved into single-player games to protect quest logic, and single player RPGs are decreasing the meaningful choice as the world gets bigger and the player can do more and more of everything.
This is the part of the article where the author usually advocates his solution and then tries to sell you something. I would say something vague like “well, what we really need is more AI generated non-linear storytelling.” And then just casually slip in there that the project I’m currently working on is doing just that.
But that’s just a string of buzz words.
When you really get down to how the AI would look, it’s not that easy. For example, most game dialog is (in theory) crafted by serious writers who work to give the NPC a personality and a unique voice. Seems obvious that if you want a story to flow, you need a human to at least check it over. So there are challenges.
The proposition is to write AI that:
- picks dialog snippets from different styles
- knows how to construct sentences from phrases based on parameters
- can dynamically create new game states and connect them to other game states
- is smart enough to construct an over-arching story
It makes a quest for Bob, a quest for Mary, and determines how the quests interact. If it wants Mary to attack the player on sight, it draws a bounding box around her that doesn’t interact badly with the environment. And the writing has to be passable, and it has to feel right (also note VO is out, unless you use integrated text-to-speech). I want to stress that this is really, really, really hard. And it probably sounds absurd to anyone reading this with a background in computer science.
But out of the 100 people who directly develop a modern game, a large portion of the coders are working on the technology around the game. Instead of what players think of as “the game.” This is everything from writing shaders, to implementing features on graphics cards, to helping the artists get a pipeline from 3dMax to the game.
Game features just tend to be engine capabilities that allow designers to make something in the editor or native scripting language. So for example, the Skyrim bookshelves are functionality that some coder probably added to a container, and I’m guessing it relies on very specific parameters for the bookshelf 3d model for it to even work.
All I’m saying is that maybe games could live with less of the latest-and-greatest-in-graphics-technology and neat little widgets like bookshelves and instead spend time reinventing how narrative gets implemented. Graphics have gone through literally a dozen revolutions in the last 20 years. Because the industry cared about it.
Bandwidth on graphics cards have increased by a factor of 16 since 1999. In 1996 the Diamond Monster 3D was a revolution — a 3d accelerator on a card! (in those days, you had separate 2d and 3d cards). Then the GeForce 1 changed it all again with on-board GPU. We’ve been through several different form-factors. SLI was popular, and then it was unpopular, and now it’s popular again.
We are doing things now that seemed laughably impossible in the 90s. Go Google search any video game from the mid-90s and realize that pixel art wasn’t there to be cute, it was there because VGA graphics were all you had. It was made to look cute, because art style can compensate for low quality graphics. You had to do it like this.
And VGA itself was a revolution from the world of 16 colors — where there wasn’t enough graphical fidelity to even have an art style. All of these advances in graphics, but we’re still making state-based quests and scripted dialog basically the same way we’ve always had.
Obviously, nobody is going to start a revolution with an $85 million game — that’s too much money to be risky. And to Skyrim’s credit, it’s taken baby-steps in generated stories with their Radiant system for dynamic content. And they made $450 million their first week of sales so I’m not saying their formula doesn’t appeal to the market.
And I played Skyrim, it definitely passed the time. But Daggerfall was released in 1996. It had spell-creation, enchantment, and a political system. It had a great story. Does the narrative and open-world gameplay of today’s sandbox Action/RPGs really represent a 15 year evolution?
I just think we can do better as an industry. Especially in sandbox games that are supposed to be about freedom and choice, where everything in the game but the narrative is giving you freedom and choice. And without consistency in permanence, you get a game filled with semiotic dissonant moments. Because can something really be permanent if it doesn’t register with the NPCs who you have come to believe exist in this living world?
And yeah it may take us 20 years to get there. But then shouldn’t we get started?