There's been a lot of talk on the forums recently about dynamic games, which I think is great. I see this as one of the two ways that games can become about narrative progression as opposed to level/gear progression. I've mentioned parts of my ideas elsewhere, but I wanted to create a thread to get out my cohesive thoughts about how a fully dynamic MMORPG could work, and get some discussions about the pitfalls and limitations of such a system.

Dynamic Story Telling Engine (DSTE)

None of the stuff I'm going to talk about is going to be possible without a robust DSTE. What is a DSTE? The easiest way to describe it is as an artificial intelligence dungeon master. It is a system that dynamically comes up with content and provides it to the player with context that make sense to that player. This is NOT just a set of text associated with a quest type and a monster (though obviously that has to be part of it). Creating bland KTX quests does not add anything to the genre.

What does a DSTE have to do to "work" 

• Understand a story grammar. This is an important concept: stories can be broken down into components, similar to a computer language grammar (or formal grammar). Following specific rules within the grammar will allow the computer to (randomly or procedurally) create legal "stories" that fit within the grammar. An example line of this grammar might be "quest -> (enemy) (villainous act) with the intent of (intent)." So the quest "Sauron creates an army of Orcs with the intent of conquering Middle Earth" becomes a valid quest in the grammer. With each of the words in parenthesis having their own grammar rules as well. There could and would be more quest grammar alternatives as well.
• Understand a story context. Unfortunately, the above grammar would also validate the following quest "Your mother's secret lover destroys all beets with the intention of winning a tournament." There needs to be context that tells the computer what makes sense, and what doesn't. This is a bit trickier, but can be achieved with moderate success through creating relationships between the individual components. For instance, the component "Your mother's secret lover" would be restricted to a smaller subset of villainous acts that make sense to it, such as "runs your family out of business." This will likely be improved over time as people use the system and data is mined from it. Once we have both grammar and context, with have a communicable language. The computer can understand what a player has done as a story, and can create stories for that player to further engage in.
• Understand plot. A valid language is only able to create a single, straight forward bit of information. If we want a dynamic world, that feels valid, a computer will need to be able to create a plot tree. This takes a similar role to context, by tying one sentence of the language to another in a plot tree. So we'll have two different ideas: a state, and an effect. A state is the current status of a quest, such as the "your mother's secret lover is trying to..." An effect is something that can happen based on your actions or inactions. And effect will always lead to another state, or an end of a quest. So, the system may generate (and find a way to communicate) the options of exposing this man, foiling his plans, sabotaging him in some unrelated way, or not acting in time (important concept discussed later), each of which will lead to a different state, until an end state is reached (ensuring that it reaches an end state is an issue as well.
• Find proper components. When a quest is generated it will ideally be personal to all the people involved in it. To do this, every quest needs to pick compatible components (characters, players, objects, villains, etc.) that already exist in the world. This is more connections, where components are eliminated based on their history and status.

• Unique items and quests. Everyone can have items that matter to them, and stories to go along with it. When you do a quest, it's very likely that no one else has done it before, and that gives a sense of novelty to every quest.
• Narrative investment in quests including possibility of failure. This is a big one. PLAYERS CAN FAIL to succeed. In doing so, there will be natural consequences that they player wants to avoid. Suddenly more is at stake than just their xp/minute. Quests can also be timed, meaning that players can know that what they are doing actually needs to be done. Players play to advance their story, not their character level.
• Natural player interaction. Quests can easily be associated with multiple players. A werewolf is attacking a town? Everyone who lives in the town knows about it. Vampire turning members of a family? That player (everyone can have a family, if they choose) will be involved in that quest, as will friends of the family members turned. They will naturally work together to solve the problem.
• Roleplaying. Everything you do is playing in the role of the world. The gameworld can literally flip from RPing being just talking IC to the only time you're NOT roleplaying is when you are talking OOC.
• Changing your world. The game world is affected by your choices, and your skill in implementing those choices. Other player's stories can literally be created based off what you choose to do. And that is true for every single player in the game.

Basically, Madlibs inside an elaborate decision tree that is seeded by a player's unique ID?  That's an oversimplification, of course, but that's what your left with after boiling off the fancy jargon.  ( ' :

I think most of it is doable, with the exception of the dynamic world -- that's a lot of moving parts, especially if it turns out another player kills off a unique quest contact or plunders a clue/item you need.

Originally posted by therain93

I think most of it is doable, with the exception of the dynamic world -- that's a lot of moving parts, especially if it turns out another player kills off a unique quest contact or plunders a clue/item you need.

Definitely a challenge to actually tie the system into the world and the moving parts. I've thought of the item/quest NPC issue, and the best solution I can come up with is to have all NPCs be unattackable unless allowed to do because of a quest in the DSTE. Similarly items would only be visible or obtainable by those with a relevant story.

Computer generated stories aren't going to work. I see you also bought into axehilt's claim that a themepark can be dynamic. it can't. Not unless you extend the definition of themepark to meaningless.

Originally posted by Cuathon

Computer generated stories aren't going to work. I see you also bought into axehilt's claim that a themepark can be dynamic. it can't. Not unless you extend the definition of themepark to meaningless.

They might eventually, AI is a interesting research project and many scientists are working on it.

But it ain't working right now at least, the Swedish IRS have an AI that are tryibng to answer your questions and it is totally off 90% of the time. If you can't even put it to answer simple questions I don't see how it would work as a GM.

What would work is to have a similar function to Biowares Neverwinter nights and let some players GM the game like in all pen and paper RPGs. The problem is just how you would select the good GMs and avoid the bad ones. Because paid GMs aren't an option here.

Maybe you could mix GW2s system with some real dedicated player GMs to get a truly dynamic game. Some DEs will be premade, others gM created and GMs can anytime posess NPCs or mobs.

Originally posted by Cuathon

Computer generated stories aren't going to work. I see you also bought into axehilt's claim that a themepark can be dynamic. it can't. Not unless you extend the definition of themepark to meaningless.

This has nothing to do with anyone else's claims. This is something I've been interested in long before I began posting on mmorpg.com.

Anyway, saying that something can't exist  without any explanation of why doesn't provide a compelling argument or useful feedback. I mean, if I were to say a crafting based world with no levels that had enemies spawn couldn't work, that wouldn't be very helpful to you, would it?

Part of the requirements of this working, however, is that the DSTE understands context, meaning that it is inherantly not meaningless. Assuming I can get it to work.

I would rather just harvest player greed to the maxium extent.   Probably the best resource an indy developer has, and why games that are essentially contentless by traditional standards(Haven and Hearth, WurmOnline) exsist so happily.

Turn players into monsters, and give them bonuses for how much damage they can do.

Tie a web empire building game to a normal MMO(EvE market+PvP).

Make some resources super limited(50 released a day), and watch as players fight for it.

Make other resources super common to use, but most easily gathered at low levels.

Make some resources harvestability just plain dissapear every other month.

Add forums, and actually allow a little bit of non-personal in-character fighting.

make unfinishable goals.

Originally posted by Loke666

But it ain't working right now at least, the Swedish IRS have an AI that are tryibng to answer your questions and it is totally off 90% of the time. If you can't even put it to answer simple questions I don't see how it would work as a GM.

There is a HUGE difference between trying to make an AI interpret the human world, and creating an AI world that humans can interpret as a story.

Even a very basic quest generator that provided a small set of quests with random variables could be used at the base of a dynamic story telling engine.

Just plugging something like Quest (Find, kill, capture, rescue, deliver) a (item, npc(random race, class, sex)) of number (1 to random) etc...

You could even take one story then randomize elements in it. This sint' something new pen and paper has been doing it since forever. http://anarchyonline.wikia.com/wiki/Missions Anarcy Online has had a random mission generator forever. I get what your saying tho and having the computer generate a story and quests etc dynamic and on the fly taking into acount things that are actually happening in the world would be way awesome.

A mission generator with a robust story built around it. Sounds great. Well it is better than KTX and goes a step beyond the Rift/ dynamic events for endless (but newish) content. Would it really more than a randomly generated mission with a few lines of text, though? A good story needs more than [villan] causing [conflict] at [location] and [NPC] telling you [solution]. I'm afraid the creativity for mission's story simply can't be thrown together by a clever mix of contitions and varibles.

Thinking back on what Todd Howard was talking about with the early implementation of Oblivion's radient AI...NPCs were constantly causing issues where a shore keeper would have all of their wares stolen or a town was empty on the player's return due to a rogue NPC that decided to murder everyone as it traveled by...

You will likely be stuck between a limited set of conditions with a rather bland mission context and a wildly irratic and uncontrollable AI mission system that will constantly be causing more problems than creating fun missions.

Creating a fully dynamic MMORPG would be an enormous feat by itself and a DSTE would be the icing on the cake. Possible, maybe in the distant future.

Originally posted by mmoguy43

A mission generator with a robust story built around it. Sounds great. Well it is better than KTX and goes a step beyond the Rift/ dynamic events for endless (but newish) content. Would it really more than a randomly generated mission with a few lines of text, though? A good story needs more than [villan] causing [conflict] at [location] and [NPC] telling you [solution]. I'm afraid the creativity for mission's story simply can't be thrown together by a clever mix of contitions and varibles.

 I too doubt that there would be a good outcome from purely automated system (at this time at least).  It hurts to say that.

Thinking back on what Todd Howard was talking about with the early implementation of Oblivion's radient AI...NPCs were constantly causing issues where a shore keeper would have all of their wares stolen or a town was empty on the player's return due to a rogue NPC that decided to murder everyone as it traveled by...

You will likely be stuck between a limited set of conditions with a rather bland mission context and a wildly irratic and uncontrollable AI mission system that will constantly be causing more problems than creating fun missions.

 There are lots of forces at work in that problem.  Combat exclusive (because "centric" doesn't go far enough in describing it) design, has major issues in a dynamic world.  From that we end up with absurd power curves (both player and NPC alike) and only one way to deal with anything (kill it).  Instead we need the "real evil" that only comes from well intentioned people trying to change the world.  More Lex Luthors and fewer "Solomon Grundys". This, along with options aside from killing would go a long way in widening the sweet spot between boring and uncontrolable swings. 

Creating a fully dynamic MMORPG would be an enormous feat by itself and a DSTE would be the icing on the cake. Possible, maybe in the distant future.

The problem with "distant future" thinking is how rarely it ever comes to be (still waiting for my flying car).  UO had an ecosystem for spawns and at least a few other games have had migration patterns.  As simplistic as that sounds, that is a pretty good start.  Meaningful weather can be built, but it is ignored and the same can be said for seasons.  Wurm shows that there is no reason that you cannot update maps (add/ remove structures) and even terraform.  Placing those elements together, you'd have a pretty darned dynamic stage.

Building NPCs for that world is a bit trickier, but adding those extra conditions even to an otherwise fairly bland mission generator would still be a leap forward.  Figuring out economic, social and political layers to add into their decision making (where applicable) would probably be rather similar to the previous steps.

Years ago, when I was working on random dungeon/maze generation I created a system where the random story/plot was generated first then the dungeon/maze was built from the plot points with extras.  The end result still looked like a randomly generated maze but with story added.  Nothing has come close to the quality of hand crafted content.

I think the evolution of voice acting is gona hamper progress in this area. People are taking to VA very well, and it would seem like a step backwards to start doing "text again".

Yeah, it may be possible with text. yes it could maybe be done with a synthesized voice although were not quite there yet. but currently we need actors for VA.
Which poses a problem..

Originally posted by mmoguy43

A mission generator with a robust story built around it. Sounds great. Well it is better than KTX and goes a step beyond the Rift/ dynamic events for endless (but newish) content. Would it really more than a randomly generated mission with a few lines of text, though? A good story needs more than [villan] causing [conflict] at [location] and [NPC] telling you [solution]. I'm afraid the creativity for mission's story simply can't be thrown together by a clever mix of contitions and varibles.

The point of the DSTE is to be able to break down any story into a core plot that both the player and the engine can perceive in the same way. Obviously the grammar is going to be far larger than just a few lines. Also, I want to try to find ways to eliminate the NPC telling you solution part, because it's probably my biggest pet peeve in mmorpgs. It leads to me feeling like the NPCs just want me to do their chores for them.

Thinking back on what Todd Howard was talking about with the early implementation of Oblivion's radient AI...NPCs were constantly causing issues where a shore keeper would have all of their wares stolen or a town was empty on the player's return due to a rogue NPC that decided to murder everyone as it traveled by...

You will likely be stuck between a limited set of conditions with a rather bland mission context and a wildly irratic and uncontrollable AI mission system that will constantly be causing more problems than creating fun missions.

These are fair points. I'm still defending the abstract here, but keep in mind this is not an AI system in the typical sense. I'm not giving each NPC a list of needs and desires and sending them out into the world to cause chaos. Remember I'm trying to emulate a Dungeon Master, not a world. A DM only needs to affect things when they matter to the story.

By default, NPCs will do nothing other than provide the information about the world that they know. They will do other stuff only when the DSTE determines that there needs to be more "stuff" happening in a particular area, so the chaos is a bit more confined than the Radiant AI's initial implementation. This is all happening at a higher level than individual NPCs too, so the system is aware of what it's already doing, and can check to make sure a quest is not going to cause issues with other quests or histories (though that may not always be a bad thing).

Yes, the components will have a limited numebr of connections to other components (which will be further limited due to the current world context, whatever that is at the time), both to keep things responsive and to keep things sane. I don't think that will lead to bland missions context, but I'm not sure yet.

Oh, and despite my arguments here, I'm sure that there will be plenty of problems with irractic quests.

Originally posted by Mellkor

I think the evolution of voice acting is gona hamper progress in this area. People are taking to VA very well, and it would seem like a step backwards to start doing "text again".

Yeah, it may be possible with text. yes it could maybe be done with a synthesized voice although were not quite there yet. but currently we need actors for VA.
Which poses a problem..

VO is obviously not an option. Even detailed quest text like we see in today's themeparks are not an option.

You'll have to pull meaning from relatively sparse discussions with NPCs, and the actual objectives that you're looking to do.

First of all: YES, I've been thinking a lot about things like this, and how dynamic content could get some REAL use instead of just mad libs.

But they are just mad libs. The goal is to make them not mad, that is, that just as the verb of the sentence is limited by the subject, the subject is limited by the object of the previous sentence, so that quests form chains, with a clear traceable cause and effect, and a history attached, perhaps in lengthy 'quest text' about how things go to be this way, and what will be affected depending on how you finish the quest. That requires a lot of prep work that the computer can't intuit on its own. Every thing you can fill in a blank will need to be coded and told what it can and can't relate to. You can make your life easier with some class inheirtance type work so you don't need to go one by one, but overall, the programmer will have to have said what a angry person is and what a step mother is, so that the game will know what sets off an angry step mother and what she can do about it, and who she notifies, and etc.

The next problem comes in because even if you have a chain of cause and effect, you don't have a story. You don't have a beginning middle and end. That kind of framework, that the mad libs fit into, might be really interesting to code, and the most challenge, because that is the part that players will see through. In its best implementation, it'll just be like a procedural episode. "Oh, House's team guesses wrong now, House insults them and leaves, now House goes some random place and gets random inspiration, and comes back and saves the patient." Perhaps you can throw in some random changes to change up the formula, but eventually, the players exhaust all the tricks and 'oh, it's a X-type episode' or whatever.

At a certain point of diversity, that's really okay. Cuz you can do that with TV, but that's where I think the bulk of the work and headaches will come in.

Accounting for those divergent cases, even if you don't have a town that's empty, you'll still have to account for quest chains that have knights throwing pies in one quest and it turns out their goal is to rid the world of food (so why use pies in that previous quest?). Part of that will be fixed by constraining the mad libs into story frameworks, but you still have those outliers. What those will look like is almost anyone's guess, but you'll have to develop constraints and 'grammar' for things you probably always assumed. That'll just be part of the iterative process, even post-launch, I'll bet.