1.30.2010

On pointlessness

If you've played through Mass Effect 2, you've met Thane. He's an assassin with morals: a stoic figure who only kills those that he believes are causing suffering to others. He has a deep belief in the old gods of his people's native religion; after completing each assignment, he retreats to his quarters for solemn meditation.

His characterization works well in the game; I found him to be the most interesting supporting character, anyway. It made me think, though: how much more interesting would an instance of "the assassin who prays after each assignment" be if it weren't a pre-baked, special case occurence? This is outside the expectations for Mass Effect obviously, but it got me thinking.

Consider a game in which a couple of elements exist: some method for being assigned assassination targets, and a church or altar at which characters are able to use the verb "pray." If both of these elements can be used by either the player or AI characters, the potential for self-expression and discovery are enormous: the player is able to roleplay the above "assassin who prays after each assignment" in a completely self-driven way, imbuing his avatar with a unique and specific character in the gameworld; alternately, the observant player could follow an AI to the mission assignment-dispensing element, observe them tracking down and killing their target, and then follow them to the church to see them pray. The discovery of this systemic characterization might be that much more memorable than encountering a pre-scripted story character.

For this sort general paradigm to be successful, a few things would have to be true about the gameworld:

  • A plethora of unique interactive objects such as the above altar/church would need to be present.
  • The majority of these objects would need to have absolutely no input into the game's central success mechanics.
  • The objects would need to be interactable by both the player and AI.
The first two points are required to make any potential chain of interactions unpredictable and personally meaningful. If there's a game with only assassination missions and a church, well, the potential combinations are not especially surprising. But as the number of mundane interactable objects rises into the dozens or hundreds, the potential for drawing meaning from performing one interaction after another increases as well.

Divorcing these expressive interactions from success-based systems is important, otherwise the player has a purely optimal reason to interact with them aside from expressivity. So if the player receives bonuses from a "Serenity" stat, and killing someone lowers that stat, but praying at the church raises it, then the designer is telling the player in a fairly straightforward way to pray after killing someone. This makes the chain of interaction less an autonomous player choice, and instead simply the most optimal reading of the game's numerical systems.

The third point is less essential, but preferable: if only the player can interact with the expressive elements in the world, their use feels less authentic, more special-case, more predestined in function. If only I can pray at the church, then this church has been put here for me to pray at, and I as the player am separated from the gameworld. But if an AI is able to perform the same actions I can, it confers not only the advantage of the above player integration into the gameworld, but hooks into discoverability: I can see someone walk into the church and kneel down to pray, which clues me in organically to the fact that this interaction is possible, without simply scrubbing the world for interactable objects.

Allowing NPCs and the player equal interactive access to these objects gives the designer the ability to script characters with specific cycles of expressive behavior: one could create an NPC named Thane (for instance,) then set him up so that he tended to take assassination missions, only accepted assignments for targets with certain traits ("criminal," "corrupt,") then always went straight to the church and prayed as soon as his assignment was completed.

Fostering this sort of "systemic characterization" would clearly require a lot of work in a game's development be put towards completely "pointless" interactions. This is already done with some frequency in certain aspects of mainstream games: for instance, visual avatar customization is completely pointless, but it's been acknowledged that many players see value in imprinting a specific appearance on their in-game cipher, and so the work is expended. Extending this kind of personalization into the interactivity of the gameworld, into not just how your avatar looks but who they are, seems that much more valuable. Broadly, it might help foster the feeling of a gameworld where "anything is possible," and the specific occurences played out or observed are authentic and unique expressions of that potential.

This entry was inspired in part by Alex Hutchinson's talk at GDC09.

14 comments:

Michel said...
This comment has been removed by the author.
Michel said...

This kind of thing was standard amongst roleplayers in Ultima Online. There were so many pointless interaction possibilities that it allowed people to create real identities. For example, getting drunk, marriage, praying, sailing (instant teleportation was ubiquitous), etc, etc.

It's something lacking from current MMORPGs. I personally have no desire to roleplay in a singleplayer game. They are better suited to allowing me to inhabit or sustain a role, not create one from scatch. I suppose if there was some recognition from the NPCs as to what I was doing it might work. Praying for a couple of polygons has no purpose unless you can talk to someone about it at some point, like Thane did.

Rob Zacny said...

Would it be preferable that other rewards be made available through the player's decision to pray at the altar? Specifically, I'm thinking of religious services, or characters that you can only meet and talk to if your character begins regular worship. Nothing that provides new quests or bonuses, but just opportunities to explore character and the world more deeply.

I ask because I know, just from the way I play (and having observed countless D&D sessions) that most players treat bits of "pointless" self-characterization as a matter of complete indifference. The cleric in the D&D campaigns says, "I go pray in the woods," when the party makes camp for the night, but religious observance rarely impacts his relationship with the character or his decisions.

If, in an RPG, I can go to an altar and "pray", I would want some sense that prayer is a real thing in the gameworld (even if it doesn't confer some stat boost). If hitting the "pray" button just has me watch my character go through a brief prayer animation, then it becomes a pointless and meaningless novelty, like having toilets that actually flush when you hit the "use" button.

Matthew Burns said...

In school I played around on MUDs for a while; one had a monk class along with a fully developed fictional religion to go with it. It had a mechanic where monks attending a daily service at the main temple got some kind of stat boost. The service was a fully scripted event that happened over the course of a few minutes in the temple's main hall.

People ended up doing one of two things: writing a script that would take them from wherever they were into the temple, then zip back to where they were the moment the ceremony was over, or sticking around after the ceremony and using the hall as an impromptu chat room.

I never once encountered anyone who was role-playing actually being pious and taking the monk religion seriously. This was probably due to the way the MUD was structured (it was fairly PvP oriented), which meant most of what people were there to do was get stronger & "beat" others (definitely not role-play).

As you say, then, perhaps the ceremony would have been better served as an interesting narrative or character development vehicle if it had no effect at all.

Borut said...

I love this sort of stuff. :)

On point 2, "The majority of these objects would need to have absolutely no input into the game's central success mechanics." - I think it's really more that they must not be singularly tied to success. They cannot simply have a net positive effect.

But they could be used by the player to succeed - they could have an effect on the world mechanically, but the effect must be ambiguous. So some players might use the mechanic to succeed at the game by incorporating it to a specific strategy, but there must be many other strategies to follow for success as well.

As an example, maybe if you pray a lot there are unique, powerful items that you can use. There would also have to be other unique, probably orthogonally powerful, items you can use if you explore other mechanics instead.

So the player is encouraged to use those mechanics for expressivity, OR to explore them for impact on the world. Then they are still there by choice, which should still have the same impact on perceiving the characterization.

I think that's how you encourage more roleplay as well - if we present the player with 5 paths, let's say, and they're confident all of them will lead to success, they're encouraging to pick the one they find interesting for whatever purpose (hopefully because they like the role). If there's no mechanic effect, some people will not choose at all.

I'd argue there's a fourth, murkier requirement as well - all the communication it takes to build the context to that systemic moment, which is a lot harder than it sounds. You might see the man praying, but you need to know he's an assassin, so you either need multiple interactions with a single character driven through these systems, standard storytelling elements mixed with it (Thane appears as a character in a scripted mission), or you need visual language that hints that he's an assassin. There have to be opportunities to convey the types of targets he's picking, for instance, if you want to go that deeply into the characterization.

Greg Sanders said...

It's a fun idea, but combining Michel and Borut's discussion, roleplaying needs an audience of some sort to really reward pointless activities. This isn't to say I've never done single player pointless, in Deus Ex 1 I dragged all of the Chinese sailors I knocked out to the deck of the ship I was going to blow up, so they would have a chance to make it out alive. But that's really the exception.

Mechanics are one way to have the game notice what you do, but I agree that roleplaying requires some not rewarded activity. I'd say Borut has it right here, perhaps a serenity stat that could be restored by a wide range of activities, ones that wouldn't stack so there wouldn't be a reason to do everything. Also you could reward habits/quirks; going to church or eating apples by given them increasing returns over time, again mutually exclusive ones so there's not a drive to collect them all.

Alternately, your NPCs can notice in dialog. Bonding via that method risks making it automatic, but again multiple mutuallly exclusive options for any given character would work well,

Really the key thing is that roleplaying is a group activity, your group could be AI, but that doesn't mean they aren't an audience.

Sam C said...

I think there would have to be some reward mechanic to make roleplaying worthwhile, even if the reward is a single comment on your behavior by an NPC. There's no interesting decision if it has no effect on the game world. What do you get for investing your time? I understand wanting to leave the choice up to the player, but there can still be choice even if there is a reward. If you're roleplaying, you're probably not worried about mix-maxing the system in the first place.
An example I can think of where there's a reward for making a certain moral choice is critically wounded enemies in STALKER. They still show up on your radar as a red dot, moaning in pain, suffering a lot. There's no in-game reward for letting them live, and killing them lets you loot their body, but there's still an interesting moral choice, just because something seems wrong about shooting someone while they're lying down. Sure, you can justify it as putting them out of their misery, but it seems wrong to shoot someone who's not a threat after you've already defeated them. Shooting the wounded enemy is the optimal game choice, but you can still have the choice to not shoot them.
In short, I think it's important for there to still be rewards and punishments for decisions to make them meaningful.

Amanda Cosmos said...

Your post reminded me of how in Dragon Age any Chantry leader you would speak to would yield a dialogue option along the lines of, "Can you bless me?" Choosing it didn't really confer any benefits that I could notice, just a cutscene, but I felt compelled to ask any later Chantry leader I could find to do it for me, especially if I just did a task for them. Maybe it's something I thought the character I was playing would do (she was a Dalish Elf—going native?). It was a nice little piece of detail and I appreciated it.

Jamey Stevenson said...

I am really intrigued by the notion of "systemic characterization", but I am also confused by your specific proposal of what this might entail.

You wrote, "how much more interesting would an instance of 'the assassin who prays after each assignment' be if it weren't a pre-baked, special case occurence?". But then later on in your own example you mention that "one could create an NPC named Thane (for instance,) then set him up so that he [...] always went straight to the church and prayed as soon as his assignment was completed."

I may be missing something, but I fail to see how this kind of highly deterministic scripting is any less "pre-baked" than the static narrative of Mass Effect 2. If anything, it seems like a more circuitous route to achieve essentially the same end, and one rife with problems as it is reliant upon a player choosing to follow this NPC around for long enough to witness all of this behavior, rather than having the same characterization communicated concisely via a compact piece of static content. It's a simple matter of fidelity.

Now, I don't want to seem like I'm completely down on this idea. I agree that having a variety of non-player-centric interactions happening simultaneously can absolutely help create the impression of a more vibrant world. But I'm not sure that the most crucial distinction to be made here is whether or not these little snippets of characterization are "systemic" or not, because the most effective examples of convincing NPC behavior that I can conjure, from GTA to The Last Express, are still ultimately reliant on explicitly authored components.

I say all of this as an AI programmer currently working on a sandbox game, and I think perhaps my skepticism arises from being acutely aware of the limitations of current AI with regard to building agents that are highly dynamic but also expressive and (most importantly) coherent. So I just want to make it abundantly clear that I wholeheartedly believe that this idea of systemic characterization is a laudable goal. I'm just having a hard time overlooking the multitude of pragmatic issues that need to be sorted out before we can even begin to scale that mountain. Until then, I expect we'll continue to mitigate these issues by leaning heavily on the crutch of statically authored content.

Steve gaynor said...

Jamey: Yeah, you raise a fair concern, though I think that the specific definition of "scripted" in use might come into play here. You'll note my use of the word "tends" in describing that the NPC is scripted to do. The system I'd picture (to dive deep into bullshit fantasy design doc land) would generate an AI goal stack based on a number of scripted "desires" which define the NPC's reaction to their current state. So, like all NPCs, "Thane" might need food every ~8 hours. Food costs money; if he hits a low threshold on the money in his possession, he might be scripted to put acquiring an assassination mission at the top of his goal stack (assuming no other more immediate situation took precedence such as, let's say, fending off a mugger in the street or pathing around a car that's stopped in a crosswalk.) Until he has completed his mission, whatever next step he needs to take to complete it would be his current goal; if he were to fail it he would want to acquire another mission, etc. Once his goal was completed (and therefore money replenished) he might react to the "assassination assignment completed" event by "wanting" to find an altar to pray at. Once the prayer goal was completed, he might go buy that food he was making money for and get on with his life.

Let's assume that in this design the player's core loop relies on tracking NPCs in their movements around the world, meaning a system would be in place for the player to tag and track any given NPC in their daily routines, perhaps via some sort of PDA-like device. This way the player could spy on an NPC without following them around physically; perhaps a log of their actions could be kept automatically as well so that you could check it later instead of staring at the NPC's map blip obsessively.

I personally find this general concept more compelling than a purely prebaked approach because the characterization of the NPC is native to the workings of the gameworld itself and, assuming that all the NPC's verbs are 1:1 with things the player can also do, the observation/interruption of/participation in the NPC's behaviors are that much more integrated with and authentic to the player's own experience.

I'm trying to consider what parts of narrative are actually accomplishable (at least theoretically) via simulation as opposed to static, authored content; we still can't simulate speech and conversation systemically in a convincing way, but AIs generating goals and performing behaviors is something we do all the time. Building a framework for sequences of these systemic behaviors to convey meaning is at least an interesting tangent to consider.

Steve gaynor said...

Sorry to go on, but: another advantage of this approach would be extensibility in exchange for fidelity: hand-crafting each character from scratch requires an enormous amount of single-use content creation, whereas--Sims-like-- unique characters that emerge from the game's low-level possibility space are recombinations of existing content, and therefore only limited by the total number of variables provided by that content. So even if the designer were to manually script an NPC's potential behavior within that framework, the investment in this setup would be a fraction of that put into hand-crafting a more traditional one-off, special-case character. The end result is of course not equal, but the tradeoffs are fairly clear.

Steve gaynor said...

I guess in a lot of ways I'm just describing Dwarf Fortress as well.

Jamey Stevenson said...

Steve, thanks for your response(s) and clarification. I think we agree on more points than we disagree here.

You're right that there is some leeway when we're talking about "scripted" behavior, since there are certainly varying degrees of complexity that can be involved that could go beyond the simple if/then rules of the Thane example. As you mention, goal hierarchies and other data structures that strive to achieve complex simulations of intent and desire are not only plausible, but already exist in many games, even if the desires don't often go beyond "attack" and "take cover". In hindsight, I probably overstated by implying that I don't see any differences between this kind of behavior and statically authored content.

As far as which approach is more compelling, I share your general bias toward systemic approaches, but I still think there are certain circumstances where static content is simply more effective at conveying certain things, and as much as it pains me to say it I think that something as nuanced as characterization might fall into this category. A simulated hierarchy of needs just isn't enough to constitute a compelling personality, for me at least. It might be fine for an ambient NPC, but it doesn't really hold up well to scrutiny. Our natural inclination to project motivation onto opaque systems helps out a little, but even that can only take us so far.

Of course, my criticism regarding the communicative fidelity of simulation is a bit of a paper tiger, since it's not like static content and simulation are somehow mutually exclusive. For instance, there is no good reason why the sort of combinatorial approach to NPC characterization that you described can't exist alongside statically authored content, with both contributing to an individual NPC's characterization. If nothing else, I think we can both agree that there are many potentially valid juxtapositions of these methods, and it's a fascinating space to explore.

I'm ashamed to say I haven't played Dwarf Fortress yet. I will have to check it out.

Huggernaut said...

I think many of the NPC problems could be solved with a "habit" system, wherein at the origin of an NPC personality, it has a wide range of activities it will want to participate in. As it does more, it tends to want to "return" to previous activities more than it wants to experience more things, but the desire is still there. They won't get so stuck in a routine that they seem lifeless, as they will always "want" to try new things, but at the same time they will hopefully have a list of things they are most likely going to be doing.

Different activities could take care of different "needs," in quotes because feeding onesself is not as necessary for survival as, say, bungie jumping, but the latter does fulfill a "need" for thrills.

It may end up with a ton of characters with nonsensical lives, but reining it in a bit would allow interesting characters (praying assassin, fireman with a gambling problem, etc.) to develop naturally, rather than needing to be forced in.