2.27.2010

State change as the key to emergent play

Here's something that I'd only recently considered concretely (or that I'd probably heard in one of Clint's talks years ago and forgotten), which is elementary yet worth restating:

The key to fostering emergent play is the introduction of meaningful state change into a game's sytems.

Consider a game with little emergent play in its combat encounters: your verbs are bullets and grenades, and so are the enemies'; battle lines are clear and enemies are aggressive toward the player; you attack the enemies until they're dead and move on.

Alternately, consider a game with highly emergent outcomes to combat encounters-- unsurprisingly, I'll use BioShock as my example. Your verbs are bullets and explosions, as well as abilities that can freeze, burn, or turn enemies against one another, or be deployed as traps; the spaces are open-ended and enemies roam around freely; environments contain hazards that the player can affect the states of such as pools of water and flammable oil spills.

Say the player encounters a neutral Houdini Splicer (a teleporter that throws fireballs) and a Leadhead Splicer (standard firearm enemy.) They might just shoot the Splicers on sight. But they might instead Enrage the Houdini, who starts throwing fireballs at the Leadhead, igniting a nearby oil puddle which spreads fire to an explosive barrel, which then explodes and kills them both at once. Well that was unexpected.

This outcome emerges from the range of possible state changes applicable to the pawns in the scene: the Houdini can be neutral or aggressive, and made aggressive toward other enemies by the player's Enrage ability; the Houdini's fireballs have imperfect accuracy and carry a fire stimulus, which can change the state of the oil puddle to "burning"; the fire from the oil puddle can spread to the explosive barrel, causing damage to all pawns in the area.

Emergent outcomes are arrived at when the player brings those outcomes about indirectly; the method that allows the player to cause indirect outcomes is state change, and furthermore state change that can propagate through the world. By introducing a single meaningful state change into the world, the player kicks off an unpredictable chain of causality from which a final outcome emerges.

A matrix of the different pawns in the world (bots, turrets, Splicers, Big Daddies, oil spills, water, etc.) their potential states (neutral, friendly, aggressive, burning, frozen, shocked, ragdolled,) and how they can be changed (hacking, Enrage, Incinerate!, Winter Blast, Cyclone Trap, electric tripwires, rocket spears) defines the map of potential emergent outcomes in the game's systems. Simply having a high number of verbs that do damage to enemies does not change the end result; fostering meaningful state change of pawns both by the player and propagated through the world enables the indirect inputs that result in emergent play.

The payoff is for the player to be surprised (Enrage -> ??? -> "Ha! Both of those guys blew up!") Surprise is valuable in all entertainment: plot twists, novel settings, shocking spectacle, dramatic turns of phrase-- all are meant to present us with something unexpected, something different from our normal experience that we couldn't have predicted if we'd tried. The key to humor is surprise-- if you expected the punchline of a joke, it wouldn't be funny; it's the key to drama-- if you saw the ending coming, you wouldn't be satisfied. Fostering emergent play encourages the player to be surprised when your mechanics crash into each other, and better, gives them the tools to surprise themselves.

11 comments:

Pavel said...

Have you played Stalker - Call of Pripyat ? : ). That's one hell of an emergent game.

Greg Kasavin said...

I agree with everything here and wish more games provided open-ended approaches to combat, let alone non-combat. One thing I'd add, though, is that I think it's incorrect to assume that players naturally gravitate toward state-changing systems when predictive ones are available to them. Surprises are exciting but I think people tend to avoid them if they can.

This is a not-entirely-true generalization, but I think it's reasonable to say that many players naturally take the path of least resistance, which often involves the path where the outcome is the most certain. Following your BioShock example, while the most exciting and entertaining combat experiences in those games involves using those systems that cause emergent behaviors, I bet you the vast majority of BioShock and Bio2 players stuck with bullets plus Electro-Bolt / Incinerate pretty much the whole way through. These are the tools placed on the critical path for the player by the designers and are the most dependable and efficient.

I agree that state change is the key to emergent play, but a game needs to provide significant incentive for the player to induce those types of state changes in order for the emergent play to occur frequently (assuming you want most players to experience it frequently). Otherwise you're depending on players' innate desire for exploration (which is probably low in most people playing a story-driven, linear game), or on extrinsic rewards like achievements for unlocking all Plasmids.

If BioShock "forced me" (meaning, seduced me) into using Enrage or Security Command on the critical path, instead of Electro-Bolt and Incinerate, or made those types of plasmids feel more powerful than the direct-damage ones, I think the outcomes you describe in this post would be experienced by most players.

But I don't mean to keep speaking for others' experiences. I've finished BioShock 1 three times and BioShock 2 twice, and barely ever used most of the plasmids and tonics, even when playing on the highest difficulty with vitachambers off.

I can't prove that most BioShock players don't play the game the way you describe here so it's just a hunch, but it'd be fascinating to see data on which types of plasmids get the most use through the campaign.

In short, I think unless a game introduces state-changing mechanics as par of the player's core tool set, as opposed to something off to the side that's there only for the curious, the sort of emergent gameplay you describe won't take center stage.

Michel said...

Isn't good AI a series of state changes? "I am unaware of the player and in idle patrol mode--oh shi there he is--I'm being attacked so I'll take cover, roll, flank, etc." Halo has no environmental state changes (except maybe huge boxes or turrets being flung around), yet I would say the Legendary difficulty combat is highly emergent.

I agree that surprise is key to a fun battle, but I think it can still be achieved with boring verbs like shooting and flanking and taking cover.

Lucas said...

What about the granddaddy of emergent sims, Conway's Life? In that, there a very small number of potential states (a cell is on or it is off), but many interesting, hard-to-predict consequences.

In Life, surprising emergent behaviour comes from the big number of interactions among elements, and keep coming because these are generative. The consequences don't have to be numerous (fire, frozen, enraged, etc.), but they should be able to kick off a chain of further consequences.

For example, groups of cells turned on or off in The Game of Life themselves generate new patterns of on/off cells, which generate new patterns, etc. The enraged Houdini Splicer generates fireballs, the fire spreads via the oil, the barrel generates explosive damage in an area.

The result of one change leads to another, and it's this sequence of consequences, each reached by an interaction among elements, that is cool and unpredicted.

Having many states makes a richer space of consequences and allows for a greater number of interactions among elements (lighting one another on fire, running to water, etc.), but in itself doesn't bring about emergent behaviour. For example, we could imagine a huge number of states that simply change a Splicer's mask type, which would not interact with other elements and cause them to behave differently. The player could cause a bunch of changes between mask states, but because the masks don't interact with the world in a meaningful way (they don't light anything on fire), they don't set off any cool behaviour.

Lucas said...

On re-reading the post, I realize you've made the same point I have in my last comment ("their potential states… and how they can be changed").

I guess Conway's Life came to mind as soon as I read about states and emergence. Yeah…

Have a better one,

Steve gaynor said...

Michel: I'm with you as far as AI leading to unexpected moments. AI definitely changes states based on awareness, state and position of the player. I would say however that the distinction is the player's ability to make intentional inputs into the AI's states, aside from being in the AI's line of sight or not.

Good examples of games that have fewer state-changing verbs but give the player intentional hooks into the AI's behavior might be Thief with its Noise Arrows and Water Arrows, or Splinter Cell with its Sticky Cams that can make noise and draw the AI to a different location. Though even these don't necessarily encourage highly emergent outcomes-- you can alter the AI's behaviors intentionally, but mostly to predictable results. Placing a gas mine in Thief then drawing a guard over to it using a Noise Arrows is intentionally and allows the player to execute a multi-step plan, but the end result is making the player feel clever, not surprised.

But, to Greg's point, I think that it's true that to make indirect player abilities the core experience, direct-damage abilities need to be de-emphasized-- this is true of Thief and to a lesser degree Splinter Cell, in that a direct attack is the least effective tactic. BioShock could theoretically be rebalanced in a number of ways-- weapons do less damage, player can carry less ammo, player can only slot a limited number of weapons at once-- to de-emphasize frontal assault and force the player to rely on indirect abilities, but one of the goals of BioShock is accessibility, and the simplest approach is also the most accessible to the most players, so weapon-only playthroughs are pretty well supported.

I would say BioShock is still successful on the combat emergence front for most players simply due to the extreme density of overlapping states and pawn types-- the nice thing about systems of this sort is that players can experience unexpected outcomes that were initially triggered by some other actor in the world, even if they themselves are not actively engaging with the richest mechanics.

Harvey Smith said...

Good writeup. Some of your "range of state inputs/outcomes" is what Randy I were trying to capture with some of our wacky diagrams here:
http://www.witchboy.net/articles/practical-techniques-%E2%80%A8for-implementing-%E2%80%A8emergent-gameplay/

Clint said...

Steve

Good stuff. I think Randy and Harvey nailed this topic in their GDC talk about emergence from 4 years ago or so. I elaborated on it in one of my later talks - intentionality or improvisation or both.

I think the key is not in the meaningful state changes, but in the ability for them to *propogate* through the world. You can have oil puddles that burn, pipes with freezing gas that break, barrels that explode, and all that, but if burning oil puddles don't IGNITE explosive barrels and exploding barrels don't BREAK pipes with freezing gas in them, and freezing gas does not EXTINGUISH fire, then you have a whole bunch of objects that DISCRETELY change state and nothing unpredictable can emerge. State changes don't trigger other state changes, so every action is entirely player triggered...

Its in the *interconnection* of the systems that things get interesting - and IMO the challenge is to ensure highly interconnected systems remain readable instead of collapsing into unparsable chaos. There are times in FC2 - for example - when a single bullet can lead to chain reactions of explosions and fire that end up killing 10+ enemies and burning entire areas to the ground - that often doesn't feel right because so much happens that the player has no idea what is going on in that chaos (fortunately, in our case, it was almost exclusively constrained to fire and explosions and AI behavior, so at least mass insanity can be parsed as 'holy - that place was excessively flammable...'.

In a game like Bioshock, it's easy to imagine many interconnected systems that feedback negatively as well as positively, and when those are hidden (either not designed to be readable, or literally obscured from view in 3d space), I think emergent things can be happening between systems 'listening to each other', and then can effectively be turning one another on and off - making it hard for the player to see the cascade. I think you guys mostly succeeded at solving this in BS2, and typically find it quite clear what is happening when behaviors emerge.

I also agree with others that designs that emphasize player directly damaging a primary enemy counter (ie: health) tend to be degenerate and work against any goals to create highly emergent play.

elias said...

I don't think "emergent" and "unpredictable" are synonymous, and I agree with Greg that players will avoid actions with unpredictable outcomes. Surprise may be a key element of humor and drama, but I think games are a different beast. Surprise through unpredictable outcomes of player actions can reduce the player's sense of agency and thus enjoyment of the game.

Emergent gameplay through state changes as described here does have value, but I don't think it needs to be unpredictable. In the scenario with splicers it seems the player could have predicted that to be a likely outcome (not 100% guaranteed--either splicer could have moved out of the blast radius of the barrel) if given enough time to assess the situation, and knowledge of how the state-based interactions of the system work. So I wouldn't really say it was unpredictable once the player learns what to expect, which is a good thing.

I see "emergent gameplay" as meaning strategies which were not predicted by the game's creators. This would include things like the Javelin glitch in Modern Warfare 2, and underrepresented classes charging gold in WoW to join instance groups and then drop out once in the instance, so that the group will be able to get first pick of the people of that class who are in the queue.

xan said...

Some very interesting points. I see exactly where you are all coming from, and I agree that the key to emergent experiences is having multiple states which can be triggered in unpredictable ways.

That being said, it doesn't need a HUGE number of states and outcomes to give the feeling of emergent behaviours. As an example, take the original Doom. There you could quite easily engineer monsters to fight one another. The impression was of a much more complicated environment that it actually was.

Greg Sanders said...

Responding a bit to Greg Kasavin's point, I think players often try to avoid state-changing systems because we've been trained not to use them.

In traditional RPGs there are often a variety of states called 'status effects.' Things like poison, slow, petrify, confuse, etc. The degree to which these states can propagate is often limited by having separate battle screens or no environment to speak of, but they're still there.

I personally enjoy emergent behavior and love doing roundabout stuff like causing enemies to target one another or the like; however, I err against using status effects in most RPGs. Why? Because they typically cost resources and don't work on bosses. I think the emergent nature of states can throw off the balance which results in designers nerfing them until they're not any good.

Happily, I think this is changing in recent games. Action RPGs have gotten better with status effects by making them time variable, they still work on tough enemies but just not very long. In tabletop RPGs but also beyond I think there's been a lot of pushback against use of immunities to balance powers, if half of the enemies are immune to status effects but few are immune to bullets, why explore? Finally I think reducing the numbers of states and clearly advertising them reduces the frustrating parts of uncertainty. In this I'm quite agreeing with Xan, status effects/states work much better when they are limited in number but can be activated in multiple ways.