Lot of talk, but what's your plan for implementing this? Peregrine City was my attempt at just this sort of thing, but I had to drop it because I couldn't invent new content fast enough. How do you program a computer to create content that's not just a mix of the parts you put into it? Pretty soon players learn to recognize how each new quest is just one part from group A, plus mission type from group B, NPC from group C, and reward from group D.

I've lately taken a renewed interest in Dwarf Fortress, and I can't help but feel that Tarn Adams has some lessons for us in terms of dynamic content.

In DF, you can build your fortress into actual interesting architectural designs. I can't help but think that's a great concept for RPGs in general, letting you not just build personal wealth but do something with it. And architecture in turn is a form of wealth. A secure stronghold might prove to be tempting to potential thieves and enemies, perhaps even make the builder feared by the local powers that be. Or a trader could use the benefit of their long experience forging profitable alliances to corner the market on a valuable commodity, or even parlay the wealth of a trade empire into political power.

Of course the problem is that characters are people, and their stories are not never-ending. The very best stories have a firm arc and an ending, and many times the chief actors don't go on to experience a sequel--the next story is played out by the next generation. I feel that everyone wants, on some level, to participate in their own epic story; at least this is true of RPG players.

I don't know if you can really achieve a good artificial story arc, but I do know that in a world of dynamic content, you can definitely have NPCs react in realistically intelligent ways. Powerful characters can become jealous of other characters who obtain power, or try to befriend someone they think can help them. Thieves look for opportune targets--they will look for richer characters to plunder. Mythical beasts will hide in the deepest, most abandoned caves and dungeons. Areas newly opened to civilization will be taken over by anyone with the power to hold the land and dwelt in by anyone willing to build there, while areas that have gotten a little too civilized could fall into political disarray, war, and even to ruin. Some of these things happen on a short time scale, and others on a large scale. I think if an RPG doing this is to realistically keep reacting to player ambitions, it also needs to provide incentives and disincentives that push the player into going with the flow. In time of war a player can make a living as a lone thief or rogue only if they get near to the action; in time of peace they have to stay away from the law. In this way the game can keep feedback loops in place, preventing players from throwing the balance too far out of whack but still using their input to drive new events.

To what extent this is doable, I don't know. Designing a good reactive world is just like designing any economic game writ large; the economy simply has more elements, some emotional and abstract. If it's done well enough though, a lot of classical economy-game problems fall right out of it, like prices adapting to the rarity of an item. If the town is swimming in goblin-made short swords, then an NPC attempting to haggle for one by asking prices of various people can eventually come across an NPC who's absolutely desperate and willing to sell at a low price. Word of the sale disseminates through both parties, and the generally expected price of swords goes down. I think the challenge is for NPCs--and by definition also any quest generator--to recognize that some tasks need doing, and that they have needs and wants, and various ways of filling those needs and wants. It's not trivial, but if that core mechanic can be mastered then all you have to do is decide on an algorithm (even a simple one will do) to make NPCs choose between the options they're aware of. As part of this by the way, communication is crucial, so another thing that falls out of it is that towns can gossip about the exploits of any given player (or NPC) who does great deeds in their territory. That in turn gives the world an even greater feeling of depth.

Sorry, in typical Internet fashion I failed to read every detail of your excellent post while I puffed up and ranted nonsensically. Here's one thing I wanted to add two cents on.

Lummox JR wrote:

I don't know if you can really achieve a good artificial story arc, but I do know that in a world of dynamic content, you can definitely have NPCs react in realistically intelligent ways.

[...]

I think if an RPG doing this is to realistically keep reacting to player ambitions, it also needs to provide incentives and disincentives that push the player into going with the flow.

[...]

I think the challenge is for NPCs--and by definition also any quest generator--to recognize that some tasks need doing, and that they have needs and wants, and various ways of filling those needs and wants. It's not trivial, but if that core mechanic can be mastered then all you have to do is decide on an algorithm (even a simple one will do) to make NPCs choose between the options they're aware of.
As part of this by the way, communication is crucial, so another thing that falls out of it is that towns can gossip about the exploits of any given player (or NPC) who does great deeds in their territory. That in turn gives the world an even greater feeling of depth.

It's definitely not trivial, but I have heard of one solution that produces a streamlined answer.

Basically, you have the objects in the game advertise their needs, and the NPCs pick up on that. It's how The Sims series and Dwarf Fortress both work.

You basically turn the whole thing on its head. Rather than the NPC thinking about how to interpret each element of the world, each element of the world thinks for itself and the NPCs are instructed by them.

I've experimented a little with this, and the really nice thing about it is this falls into an object oriented format rather nicely.

Introduce a new aspect to the world and you don't have to tell the NPCs anything. Instead, the aspect you're introducing knows all about itself, and competes with any other aspects previously introduced to goad the appropriate NPCs into doing its bidding.

(This is how The Sims is able to introduce expansions which bring new behavior to the Sims. The behavior is in the objects, not The Sims' AI. All the Sim decides is which object of the ones which are advertising to it will fill its needs.)

Of course, this is just one solution, it's not the only solution, and even in this solution it takes a great deal of niggling experimentation and paperwork to iron out the exact behavior you want.

This is also not to trivialize the extreme significance of what you're suggesting with dynamic NPC behavior as a part of dynamic content. I want to realize that, too! I'm just saying that this is the literal direction I'd currently go based off of the methods I heard about.

I know what you're going through. At one point, what I had was a hierarchy of units. (Just the framework, not a working model.)

The idea is that every unit in the game has a master and subordinates. The master instructs its subordinates, but the subordinates let the master know its needs.

How I thought to clear up the communication issue is that each unit only performs certain interactions.

The townsfolk might recognize, "the rabbits are eating my cabbage!" They consider the viability of taking out the rabbits for themselves. If it turns out they can't, what they do is send a notification up to their superior, "monster problem here."

The superior's AI is programmed to handle certain things such as "monster problem here" and amongst its subordinates are the town guards. It sizes up the threat versus the capability of currently available guards. (The bold part being an implicitly handled task queue.)

If it is deemed the available guards can handle the task, away they go. Otherwise, the villager's superior escalates even higher: perhaps to the king and his armies! His majesty receives "reinforcements for monstrs needed" and knows how to handle that. He instructs one his available subordinates, a general of a company of his army, to take care of the rabbits: problem solved.

The trouble at this point isn't that the AI can't figure out how to solve its problems. The trouble is that the player seems redundant! This game is intended to be played by players, right?

What actually ends up happening is error states. What if the AI doesn't know how to handle a problem (perhaps it tried and failed) or no means materialize to solve the problem (the king's men are busy)?

What's taking those guards so long to come back from the cabbage patch? Something needs to be done, but the AI has no idea what. The whole thing falls apart. The cabbage is eaten, the surfs starve, the kingdom collapses: unhandled cabbage eating rabbits are now in danger of destroying the world.

Guess what the heroes of your adventure, the players, are? That's right: glorified troubleshooters bridging the gap of things the AI is unable do! Powered by human brains, the unstumpable nature of your players makes them valuable indeed.

(It's creepy just how revealing this is: in your average D&D campaign, your player characters are just solving problems everybody else is too dumb or lack the means to do.)

Once the hierarchy was set up, all you had to do to set up an epic "save the world" story was keep randomly spawning rabbits until they showed up in the wrong cabbage patch at the wrong time.

The main problem now: Suddenly, there's fires everywhere, and the players are sick of putting them out. They ignore the problem, and so the unthinkable occurs: the world ends.

Who knows - maybe that's a good thing. If the world ends because the hero of fate stayed home, isn't that the only way we know he had a real purpose?

The "putting out fires" approach is an interesting analogy. I think the problem might be simpler to handle though in that it can be looked at as the game trying to give players the opportunity to participate. The rabbit problem for instance can manifest in a few ways. Villagers who can afford traps might be buying them, driving up the price. Villagers who can't deal with the rabbits themselves could ask for help, but since small animal trapping isn't necessarily the guards' domain, the local ruler can institute a bounty. At this point the player has the option of seeking that bounty, but if no one steps up for a while, a few NPCs might take it on themselves figuring it's a sure-fire way to feed their families.

As far as pushing things up the chain of command, the meat of that seems to be the question of "Is this something I can deal with, and if not can it be handled by anyone else locally or is it too big a problem for us all?" Obviously the coming of a dragon falls into the latter category. A pending invading army does as well. A band of thieves causing problems, that's work for the town guards and it also opens up potential bounties and quests as well.

On the subject of AI cheating: I've written about this before, how in SotS II the NPCs have a specialized system that treats them like real players in terms of giving them a more realistic form of memory. For a game like that, that's pretty important. For a simulated medieval town, it's probably quite unimportant how you handle communication and observation except when it comes to combat.

The fundamental problem with any system such as those being discussed here, is that in order for any of them to come into existence, they need to be designed and programmed...

At first glance, that is a very glaring "Thank you, Captain Obvious!" statement, but the reason it needs to be kept in mind is that for many of these types of systems, that programming (heck, just the planning behind it!) can quickly grow to be a prohibitive amount of effort...

If anyone really wanted to put forth the time and effort, they could create a virtually flawless real-world simulation (or fantasy-world) down to the minutest details... This would require the design and representation of virtually every single aspect of reality (think, The Matrix, with or without the fancy visuals and other sensory outputs) Behind a fully functioning simulation of the world lies a framework of an incredibly huge number of rules (again, equal to the number of rules our own reality is based upon)

Unfortunately, though, no one's up to that challenge... It's just far too daunting of a task from any angle, and at any stage... Yes, it theoretically can be done, but it likely never will...

So, we're forced into limiting ourselves further and further until we get to a level where we're comfortable with the effort required to pull it off...

Thus far, though, that level always seems to fall short of expectations... As it turns out, even getting the level of sophistication to make a passingly complex and nearly-real world is just beyond the threshold of effort we're willing to put forth...

I see lots of talk of these sorts of higher-level AI systems, economic systems, etc., but none of them are likely to come to fruition, because when you get right down to it, they require a hell of a lot of work... They sound amazing, and would likely be amazing if ever realized, but there's very little chance that they ever will...

And this isn't just directed towards BYOND developers, this includes commercial developers, too (for instance, even Spore falls far short of the general hype and expectations, simply because to really make it what people wanted it to be would have taken an exponentially greater amount of effort)

Not that anyone should stop dreaming, or theorizing, or even attempting... It's just that we shouldn't let these pipe dreams keep us from setting (settling?) more realistic goals and actually getting something done with them...

So rather than trying to flesh out an AI system or economy into something that nearly simulates real life (down to NPCs communicating with each other to solve problems, set prices, etc.), just take a step back, and fudge things a little... As long as from the outside it looks like something sophisticated is going on behind the scenes, it doesn't matter if it's just a man behind a curtain... Players will pay him no attention...

It's what all of the pros do... There's no harm in trying to reach for something a little bit more, but you have to keep your feet on the ground...

The way I look at it, if I've got the idea really well worked out in my head, it simplifies itself to the point where it's no harder than most of the things I do.

This cellular AI approach sounds super hard, but it's actually a lot easier than a centralized AI. Rather than having a routine that goes through all the units in the game and tries to figure out what to do with them, the units simply manage themselves and only escalate to a centralized AI when they encounter a problem they can't solve on their own.

It's the same with any technology, really. The workings of a flashlight seem painfully trivial now, but the early pioneers no doubt invested considerable effort developing the technology we take for granted now.

A couple things I could say about that.

First - how many possibilities are there in the average RPG, anyway? Computer RPG, that is. Attack, drink potion, cast spell -- ever play Final Fantasy XII? They actually gave you a simple table you set up in their Gambit system and the game pretty much plays itself.

Second - patching the gaps where I've yet to get around to telling the AI what to do in this specific situation is where I was thinking player involvement could pick up the slack.

This all sounds a whole lot more complicated than it is.

I wouldn't call it a holy grail if it were easy to find, now would I? ;)

However, the tricky thing about incredible artifacts is they're not nearly as special after you've found them, toted them home, and tried to figure out a good way to display them on your mantelpiece.

Go back 20 years and tell somebody you've created a game that uses 32x32 pixel tiles displaying tens of thousands of colors consisting of multiple layers while networking to allow multiplayer interaction from anywhere on the globe, and they'll tell you that that's impossible and an attempt to create it would inevitably fail too.

Dantom delivers us an engine that gives us all that, and we no longer sweat the details. We didn't even have to pay for the privilege to use it.

All I'm doing now is the kind of brainstorming that Dantom did earlier, but towards a different ends. I'm not saying I'm already there, but do you honestly think it matters that there's unresolved details? I'll just get to them when I get to them. If I don't, somebody else will, but maybe I'm not that patient.

The main reason why we haven't seen them yet is because not enough people have got around to doing it and releasing it in a game we've found and played. It's not commonplace feature, yet. However, games like Dwarf Fortress or The Sims show us that it can be done.

Let me put it this way.

We've all the materials we need to build this house, and the challenge is in putting them together correctly. This much you've said.

However, we can't build the house all at once. We can only build it beam by beam.

Focus on the incredible audacity of a dynamic content generation system that includes cellular life, and it's very, very, quite impossibly, hard.

Focus on the individual steps in doing so, including a step in which you conceive a plan of steps, and it becomes much more approachable.

Clearly, my problem isn't that big concepts daunt me (though I do tend to get snagged on little details just because I enjoy pondering the possibilities so much). My problem is I'm easily distracted and physically tired. That gets into that next blog entry. ;)