header image

Is an AI-controlled narrative really beneficial to players?

One would not expect an online trade rag for the IEEE to contain decipherable gems of content for the average Joe. Indeed, even the above-average Joe might struggle to convince himself it deserves an RSS subscription, especially with front page features like “Your quick-and-dirty guide to the fourth fundamental circuit element”.

But this article on artificial intelligence in games is a terrific read. The title “Bots Get Smart” is a little misleading – sure, it starts off talking about computer-controlled opponents in first-person shooters, but manages to segue into AI storytelling. The latter topic is what I will focus on.

I could spend a paragraph or two breaking it down, but the article does a good job of that already. Cue a mammoth cut-and-paste:

PaSSAGE uses the same game engine as Neverwinter Nights, a fantasy adventure set in medieval times, produced by BioWare of Edmonton. With PaSSAGE, scriptwriters determine only the most general arc to the story and provide a library of possible encounters the player’s character may have. The computer studies the player as he or she progresses and cues in the kinds of experiences that are most desired. For instance, if you like fighting, the game will provide ample opportunities for combat. If you prefer to amass riches, the game will conjure up ways for you to be rewarded for your actions. The software is able to make the sequence of events globally consistent by maintaining a history of the virtual world’s changing state and modifying the player’s future encounters appropriately. The game will therefore always appear to make sense, even though it unfolds quite differently for different people—or even for the same person as his moods and tastes change.

Neverwinter Nights appears a popular choice of tech among academics, be they teaching the fine points of journalism or building an MMO so realistic one cannot make armour without assembling the 50-odd components required for its construction. I’m guessing it’s because the Aurora toolset is so powerful and NWN can run smoothly on most PCs.

Fundamentally, PaSSAGE shifts the focus from the “one situation, many solutions” design philosophy of most role-playing games to a “many situations, one solution” one instead. This doesn’t mean it’s better, just different. Content creators go from crafting a limited set of specific scenarios to a metric crapload of generic ones. The smart thing to do, at least from a pipeline perspective, would be to make it so monsters, treasure and NPCs can be swapped out to recycle as much content as possible. However, a designer could manually create scaling scenarios if there’s a fear it’ll be too generic.

While I don’t think the actual scenario creation part would be hard (just time-consuming), threading these encounters together into a seamless narrative might be. Firstly, the system would have to learn what the player finds fun, and this would require throwing a series of pre-determined situations at them when they start. It’s not that much different from, say, the intro sections of Fallout 3 or Oblivion. But how many test scenarios must the player trawl through before we can take the leash off? Is there are smooth handover as fixed scenarios are mixed with generated ones, or is the first hour or two of gameplay the same for everybody?

So far, the player hasn’t experienced anything they wouldn’t normally get in a game using a standard narrative structure. The benefits might only reveal themselves on a second or third play-through. Even then, the player might have to radically change his playstyle, which might not be fun.

Another problem is sterilisation: How long could such a system run before the overall experience feels generic? I guess it can’t be any worse than grinding in an MMO, but if this is what it equates to, then why go to the effort in the first place?

Essentially, I see the main advantages appealing to developers rather than players. The burden shifts from building characters with massive dialogue trees and catering to sneaky, strong and diplomatic approaches, to programmers building intelligent routines to learn from the player and thread the scenarios together. Definitely an interesting option for a one or two-man team, but I think a full-blown studio would be hard-pressed to adopt such a model.

~ by Logan on December 11, 2008.

8 Responses to “Is an AI-controlled narrative really beneficial to players?”

  1. I wonder if you could use the Bartle Test as some kind of underlying framework for this sort of thing. I have some reservations about the concept though, mainly implementation-wise. Say that the game throws a repeated monster encounter at you, and those monsters drop a really rare, valuable item. A player starts grinding those monsters. How does the game know what they’re doing it for? Are they grinding to get the rare item? Are they grinding because they can sell that item for a ton of cash? Could they be grinding because they just enjoy killing the monsters? What if those monsters are just all over some mountain and the player is merely trying to reach the summit?

    Actually, a Fallout 3-style multiple-choice test at the start of the game that is essentially a simplified Bartle Test might do the job far better than trying to interpret the player’s behavior.

    I think that this is a potentially really useful concept regardless, but not as the be-all and end-all. I think it could be pretty useful for dynamically adjusting the game slightly – a sort of extension to the way Neverwinter Nights will spawn different numbers of mobs in an encounter if you’re under or over levelled – but within a standard game framework.

  2. “A player starts grinding those monsters. How does the game know what they’re doing it for? Are they grinding to get the rare item? Are they grinding because they can sell that item for a ton of cash? Could they be grinding because they just enjoy killing the monsters? What if those monsters are just all over some mountain and the player is merely trying to reach the summit?”

    That’s a really good point actually. I guess the easiest response is that, if the player is doing it repeatedly, then they must enjoy doing it, so why not provide more of the same. On a larger scale, it might be important for the AI narrator to not only understand that a player is doing a particular action, but why they’re doing it. It can make a rudimentary assumption based on scenario content (“He’s killing monsters, so he must like fighting”), but there’s definitely a benefit if it’s smart enough to read into that content (“He’s killing golden rats, so he must want/like their golden tails.”)

    Things get more complicated though if the game starts moderating content. Actually, it gets really complicated.

    In the end, I still think the work involved versuses the benefits gained aren’t strong enough to sway me from a more streamlined system, but then I’m coming at it from a production standpoint. I get the feeling there’s a type of game this system would fit perfectly with… it just might not be something as complex as an RPG. Imagine an FPS that adjusted its rooms (or swapped them completely) to provide gameplay catered to the player. A lot less variables and more predictable behaviour from a designer point view… a bit of potential there.

  3. “Imagine an FPS that adjusted its rooms (or swapped them completely) to provide gameplay catered to the player.”

    Played Left 4 Dead yet? :P

  4. Heh, yes I have. :)

    It’s a good example of the idea on a basic level, aimed at moderating tension, not playstyle. Regardless of L4D’s AI director, it’s still run and gun (which works awesomely and is a great deal of fun).

    If you do less shooting, the game doesn’t cater for that. In fact, it usually gives you more things to shoot! I was thinking single-player, and changing physical level design/puzzles/etc to match playstyle.

    EDIT: For example, take Bioshock. Imagine you use the Telekinesis power a lot. The game might detect this and provide rooms/sections of level and situations that let you go crazy with the ability. You’d still be able to shoot your way through, of course, but the player feels cooler because they get to use the plasmid they enjoy the most.

  5. Ah, I see.

    My worry would be that the logic that the game would be using may very well not reflect reality. If the player likes the telekinesis power then they will use it a lot. That follows. But you can’t turn the implication back on itself – the player using telekinesis a lot doesn’t necessarily mean they enjoy it. It’s been observed before that gamers tend to gravitate towards efficiency in gameplay mechanics. Since you mentioned Bioshock, take the electricity plasmid (I forget the exact name) that was able to stun people so you could crack their skulls with the wrench for example. I used that quite a bit, because it was often the most efficient way to dispose of a lot of enemies – in fact, for getting rid of security bots and cameras it was often the only option. Those things irritated me to no end and I didn’t want to waste valuable ammo on them. The last thing I’d want is for the game to throw more of them at me because it thought I enjoyed that. On the other hand, if the game was smart enough to see that I was using the machine gun a lot and spawned extra ammo for it more often, that would be fantastic.

    I think Bioshock basically did what you’re talking about as part of its design anyway. Different ways to approach stuff, different plasmids and so on. Lots of optional extras and stuff that you could get if you employed your skills the right way. That was one of the best things about the game, and something that isn’t achieved very often. Another example I can think of is Deus Ex. You can get through the game using a minimum of effort, but if you take some time and explore there’s a treasure trove of optional extra stuff that you can find if you have the right abilities.

    I think if you get to the point that you’re slotting in different rooms in your levels dynamically, then you might be getting the cart before the horse a bit. Why not just offer a few different rooms that have different methods for getting through? A designer would still need to make all the rooms regardless. Why go through the effort of trying to make the game guess when you can just provide options in the first place? As long as you put a decent amount of variation in, you’d probably find that the player would respond exactly the same.

    On the other hand, I do see a lot of potential to the idea if you have a game that dynamically creates its play environment completely. Take Diablo for example. In essence, Diablo was a Roguelike with graphics. What Blizzard did was build rooms with set pieces in them, randomly locate them, and then randomly generate a dungeon around those pieces. The result is a Roguelike with a plot. Building on that you could have a lot of rooms that would not all be used in a given area, so you’d get a random sampling of them. Putting intelligence into that randomness would be fantastic. I think you could probably implement it reasonably easily too, given enough rooms. Come up with a few stats – say, puzzle, event, magic combat, treasure, physical combat. Each room would be given a value in each of those stats. The routine which randomly selects the next room to place would take weightings in each of those stats, and based off those weights plus a random factor, try and choose the closest match. The player’s actions would modify the weightings, and you’d probably also need to balance the weights, i.e. if you’ve spawned a lot of treasure rooms, that weight would go down.

  6. I think the author of the origonal peice has sold the idea (and tech) short. What we have here, is a system that can:
    study and understand the players actions,
    track historic world events,
    discern the current world state,
    modularly and dynamically build a narrative.

    Thats rather impressive, but the authors ideal implementation of the tech is flawed.
    As nz mentioned, players look for the most efficient path (hmm.. must note to ensure the most efficient solution is the most fun), and as the ballance of the game is changed, so may be the most efficient solution.

    For example, Abe finds it best to sneak by enemies rather than confront them head on. The AI adjusts the game so Abe encounters less and weaker enemies. Abe, realising the enemies are now sword fodder and battles are further apart, begins to take a hack and slash approach. The AI increases enemy encounters and strength, Abe reverts to sneaking.

    So the question is, who is Abe really combating?
    The game may be chess.. but my opponent is the one sitting across from me.

  7. Some great points, especially about exploitation of the system (be it intentional or not). The most efficient solution or “dominant strategy” throws a spanner in the works. As has been mentioned previously, the AI would have to be able to discern the whys behind a player’s activity. But it would also need to know how to handle that activity.

    Shame we can’t always assume the player is playing for fun. :)

    It’s always easier to design and test a set piece over something dynamic, and that’s why I think hand-crafted narratives and scenarios will remain the popular choice from a production point of view.

    Give the AI system a few years to simmer, and keep an eye out for games that dabble in the idea (L4D). I’m sure we’ll reach a compromise. :)

    EDIT: In regards to the machine gun ammo thing – I know quite a few games that have intelligent powerups. Breakable crates are fantastic for that.

  8. […] Is an AI-controlled Narrative Really Beneficial to Players? […]