Gabe Newell Writes for Edge

Left4Dead released this week, and I’m extremely pleased with the reception so far.

We set out to create a first-person four-player game in which the action changes each time it’s played. We wanted to use procedural narrative to simulate stories because, in a multiplayer environment like this, we felt a more linear, scripted approach would not deliver in the same way.

I am going to discuss here how procedural narrative works as a story-telling, gameplay device; some of the issues we faced in developing Left4Dead and a few of the lessons we’ve learned for the future.

Behavior Models

To come up with a theory for how to do narrative in a multiplayer game we started by looking at the players’ different behaviors – exactly how they’re playing, what their actions are, where they move.

We tried to note as many interesting contexts of their actions as possible. Are they moving together as a group or are they splitting up? Is their mouse jerking around a lot, or are they interacting smoothly? Are they agitated or are they relaxed? How much damage are they taking? How accurate is their shooting?

Then, using all that data we tried to come up with a model to tell us how the player is faring in the game.

The data can tell if a player is comfortable or if they are starting to be overwhelmed. Once you have that concept, then you can start to create pacing and events in response to the player.

Pacing and Events

The pacing part of it is the first piece of the puzzle. Once you are monitoring all four players as they’re playing you can take the group right to the edge and then back off. You can give them any sort of pacing you like and you can throw appropriate events at them.

The events are trying to give them a sense of narrative. We look at sequences of events and try to take what their actions are to generate new sequences.

If they’ve been particularly challenged by one kind of creature then we can use that information to make decisions about how we use that creature in subsequent encounters.

This is what makes procedural narrative more of a story-telling device than, say, a simple difficulty mechanism.

To achieve a sense of story you need there to be some notion of intentionality on your opponents’ part. This is entirely different from a difficulty level which just ramps up to a constant level.

In terms of the signal that you’re giving the player, a  difficulty level is like a flat line response as opposed to a wave. We tend to think of it almost in terms of signal processing. A difficulty level just says ‘go up to this level and remain constant’ in terms of the experience that it’s giving to people. That isn’t really the most entertaining experience that you can give people. They want peaks and valleys and really big reactions to the choices that they make.

They want the consequences to be clearly connected to their actions – yet also to seem believable to them – so that when they do something particularly well or particularly poorly they want a large amplitude response to that, as opposed to just have more monsters thrown at them.

Much to Learn

We’re very much at the beginning, not only of figuring out the shared procedural narrative, but of storytelling in games generally.

It’s been one of the topics that’s been super interesting to us as a group of developers since the original Half Life. At that time, we looked at the shooter genre, which really had degenerated into a shooting gallery, and we believed there was a lot more room for storytelling. The first person perspective really opens up opportunities for storytelling and so we’ve always been interested in the ways in which games can become a storytelling medium.

What we’re trying to do now is to create a shared story that you and your friends can all be part of rather than just the experience that you go on by yourself.

Going forward, we’re definitely going to use some of the things that we’ve learned – what worked and what didn’t work – with Left4Dead not only in multi-player but also in our single player games in the future.

It’s just given us a better way of reacting to a set of player choices that span time. We’ve always wanted games to have more memory of your actions and some of the things that we’re doing in Left4Dead are a response to paying more attention to what groups of players are doing which you could just as easily apply to a single player game.

One of the theories of fun we use is that the more ways in which the game is recognizing and responding to player choices, the more fun it seems. There are simple, narrow ways in which that’s true. For example, shooting the wall when nothing happens is not as fun as shooting a wall when something happens and dust flies up. That’s a really simplistic example of why you want the game to be paying attention. It’s a narcissistic experience. You want the game to pay attention to what the player is doing.

Development Challenges

I’ve been asked a few times about the technical challenges of this project, and they’ve definitely been there. But one of the most interesting challenges has been organizational.

Traditionally, a lot of game design occurrs in the level design department, so people are actually creating geometry and placing entities and controlling the logic of those encounters. So our level designers are used to having a bunch of tools that they can use to create experiences like, ‘oh the player will do this and then I’ll drop this thing on their head’.

But with moving to a more procedural approach that doesn’t take place so much. They don’t place weapons. They don’t place monsters. They don’t control the number of monsters that show up. Those very simple knobs that they’re used to turning just aren’t there.

They’ve gone from being the director of the experience to handing that off to a big chunk of AI code. That was an interesting thing for them to wrestle with – to have to learn to have faith that the in-game director would do something interesting with their spaces. They’re used to having a lot more control than that.

There have also been lots of advantages, in terms of customizability, about having a procedural approach. We can turn off the graphics and have the game play itself and then use much more statistical methods for analyzing outcomes. We can play tens of thousands of games every night and then spend some time looking at graphs rather than watching individual players play.

It definitely moves us in a much more abstract and quantitative way of thinking about big chunks of gameplay while moving a lot of the responsibility out of the hands of level designers and into the hands of programmers.

Test, Test, Test

But I don’t want to give anyone the impression that we’ve abandoned human play testing. The automatic testing gives us a security blanket but we still want to watch people and make sure that the things that we create quantitative metrics for actually have real world applicability.

We’re theorizing about a lot of this stuff  –  we want to actually watch players and find out how they’re reacting. We have to test each play theory and make sure it’s actually true. Exposing the theoretical model with actual players was a super important part of testing the game.

We want to take that even further in the future. One of the areas that we’re really interested in is testing biometrics on player state.

Right now we have to filter our observations of players by simply by watching them and then we have to guess as to how certain things are affecting the player. But there are new technologies where we can wire players up with EEGs and actually have direct exposure to their physical reactions to the games.

We can know for sure of something is actually frightening the player – their heart rate is going up, their respiration stats are peaking, appropriate parts of their brains are being activated. Direct measurement of players’ arousal states and responses to the things we’re doing is super exciting. It just will allow us to be much more analytical about the decisions that we’re making and the roller coaster ride we’re trying to create for the player.

Ultimately, statistics are guiding stories and generating narrative, but they are all based on actual, real-world, human experiences of playing the game.