The kind of RPG I would like to see

[Ken_Oh]

That's very interesting. Some sort of drama manager (haha, DM) might be called for, but I'm personally interested in the exact opposite of that.

Take the murderer example from the first post. I wouldn't want any undue clues to solving the case. I also wouldn't want aspects of the case to be necessarily driven by player action. That's one thing in traditional RPGs that breaks the fourth wall for me. If your murderer is intent on killing everyone in town, one villager at a time, why not let him if/until the player (or other NPCs) don't manage to stop him? I'd rather not know or feel like an invisible hand is guiding me towards a goal; rather, I would want to know success depends on myself and perhaps a little luck.

Still, very interesting read and good food for thought. Thanks. However, I didn't find the game. Is it available?

[MCP]

No it is not available, but I played it. I had to fill out some agreements so I could play it.
It's a Zelda clone, and I beat it in 5-6 minutes and 15 minutes of total time with forms and stuff. She was saying 30-40 minutes to do this??? Do people suck at games?

Pre-first, I completely ignored what I was supposed to do and ran ahead looking for anything. It turned out I didn't need to know what I was supposed to be doing to know what I was supposed to be doing(lol).

First, it was too easy, health potions regenerated when you left the room, so obviously I abused that and got like 20 of them. They can fix this easily.

Second, I went straight for the magic sword I read about in a note and beat the [censored] out of the enemies (monkeys).

Third, I escaped with the pick axe I got from the prisoner(why didn't he get out of jail with the pick axe?) and the candle from the monkey king.

If I re-did it I could probably beat it in 2 minutes easy, assuming the magic sword, the monkey king, and the prisoner don't randomly move to different rooms.

[Jetrel]

Automated content generation is a pretty tough field. In particular, once you get to the stage of having to dynamically generate conversation, it's going to be very complicated, and very difficult to make convincing. Good dialogue is going to have to be hand-crafted.

I think this will be an area of development, but it would be very hard to do. I think anyone trying such a thing should aim for small evolutionary improvements over past games rather than setting their goals too high.

For one thing, it's quite arguable that in order to create convincing procedurally generated dialogue, you may have to write a fully functional AI. And I don't mean the kind that controls the bad guys in wesnoth - I mean the kind that perfectly simulates a person.

No one has ever succeeded at that. Nor do I expect they will until we're fully aware of how the human mind functions, which will take decades or centuries to unravel.

[e7th04sh]

@Ken_Oh: The murderer's case is a good adventure for small version (if it was possible to do that). Then, i think that in a full blown version you should be able to fail at each single adventure, and actually do nothing interesting if you are bad/unlucky at what you do. Anyways, that's what most of us do with their lives.

@Jetryl: I believe it doesn't have to pass Turing's test with a psychologist, Jetryl. It will be just enough, if it passes it when talking with me. What i mean, is we can invent totally different scheme of reasoning, emotions, that just tries to simulate ours. It can be done that without science, but with introspection instead. Perhaps psychologist(s) on dev's team would help.

[Ken_Oh]

In case anyone is wondering, I'm planning to make baby steps on making something like this but I've figured out that it's not my next step. Before I make NPCs that are the face of towns and villagers to players, I need to make the villages. I'm working and I hope to make great progress soon on a system where villages and cities (groups of village tiles) grow and interact organically. Units will spawn from villages at this stage only for economic or military purposes (invasion, raiding, trading, supplying, defense, retreating, etc.). I hope to have something that I can just leave on, let the AI continuously take turns and get interaction between villages and nations without any human input. With all that in place, I can think about spawning NPCs that will interact with players.

[markm]

Like umpteen other people I too have been working toward this for decades. I reached a point where I figured I had to do it top down instead of bottom up. That is, instead of starting with individual characters in an inn or tavern or village trying to gradually invent a larger and larger world around them like human-run D&D campaigns around here tended to when D&D first came out, start by generating a world of the technology level that you want.

For example, run Freeciv in AI mode for however many turns necessary, either by guessing the endyear you want or by using its scrpting to set endyear to the current year as soon as someone develops the technology at which you wish to start play. (Bronze? Iron? Unfortunately, as I found in making a "Quest for Fire" ruleset for Freeciv, settlers have agriculture and roadmaking and fluent speech right off the bat so there are limits on how primitive you can start if using Freeciv until someone separates out the flags for roadbuilding and irrigating.)

Now you have a world, with existing nations that have developed various relations between them. Time to zoom in. Looking at the map of Wesnoth I counted eight huts in Weldyn, off the cuff lets make a wild arbitrary guess that maybe it might work to guess that means city of size eight. Even if Weldyn isn't size eight, our randomly generated world might have a city of size eight so lets figure how we want to represent that in Wesnoth. In Freeciv it means typically eight tiles near the city are being worked on any given 'turn'. Maybe we can represent worked tiles by placing villages on any Wesnoth maps we make of those tiles. Would it work to put 8 villages on each such map to representing a size 8 city is working those tiles? Mind you there are likely also villages even when a tile isn't being worked. So maybe 8 on each map worked or not, plus 8 if being worked that Freeciv-turn? Then how many on maps that are on the central Freeciv-tile? Size squared?

Plugging in actual numbers will need tuning by people excruciatingly familiar with Wesnoth-map-based operations.

Meanwhile Freeciv has 'units' moving around. I figured each such 'unit' should have a recruiting list of which Wesnoth units it gets to recruit when a Wesnoth 'leader' of a portion of that unit appears on a Wesnoth map. That is, I would prefer that the Freeciv units have more variety when zoomed in to Wesnoth scale than simply being a homogenous horde of one unit-type. A "Knights" unit should probably be able to have some horsemen (squires?), mounted sergeants, whatever. (In Chivalry and Sorcery a "lance" of Knights was one knight, his squire, and a pair of mounted sergeants, but we'd probably want to also have it able to recruit spearmen, archers, whatever.)

So now we have NPCs moving around, even if we only know from decade to decade the huge region (100km or 100mile tiles or thereabouts) they are concentrated it. Its a start, it gives facts the chatty bartenders can refer to, specific hordes of troops moving around in specific areas, exerting threat or discouraging threat over a particular area of influence and zone of control.

To zoom down even further, I had been thinking Crossfire.

But the basic point here is building it top down instead of bottom up.

-MarkM-

[In_My_CrossHairs]

Your asking for the non-linearness (I guess that's a word) of a RPG say, Morrowind, the hyper-intellegence of the AI from F.E.A.R., the vastness of Civilizations (the game), and a dozen other games that I can't think of at the moment. It'd be a vast and time-consuming undertaking to make a beast like that, to say the least. I guess it could be possible to see such games coming out in the future. I kind of doubt most game developers have the machines that could make such a game. I bet you the entire game would take up an entire desktop computer hard drive, if said game was made . But who knows.

[markm]

I figured an online game, not a game the entirety of which resides on a player's own machine. There are more and more providers offering the kind of huge computing resources that it would need, a few I can think of offhand being Sun's Darkstar, Amazon's computing cloud and Google's similar offering, albeit two of those have a specific language restriction currently. It does seem though that the hardware is out there, begging for the software...

-MarkM-

[In_My_CrossHairs]

Wait. So your saying the game would be downloaded from the main website of said game, probably just 100 megs or so, and the main client would be on the developer/s computer/s? Now that could be very possible. And yes, graphics cards, motherboards etc. seem able to run such a game, but I don't think a great majority of people would have such a computer. It kind of sounds like a Crysis scenario, where everyone thought it looked fun, but knew it wouldn't run on their computer. Maybe it be a wiser solution to wait a few years when most people will have such a computer, and then make it. I don't know, its just my opinion of the idea. It DOES sound cool, though.

[Jetrel]

@Jetryl: I believe it doesn't have to pass Turing's test with a psychologist, Jetryl. It will be just enough, if it passes it when talking with me. What i mean, is we can invent totally different scheme of reasoning, emotions, that just tries to simulate ours. It can be done that without science, but with introspection instead. Perhaps psychologist(s) on dev's team would help.

Emphasis mine. The human mind is so complex that you can't "fake it", as you suggested, by scripting together a series of predefined emotional responses. No matter how many rules you made, you'd still end up with something as wooden and predictable as the combat AIs in most videogames. You have to go straight down to the root, and simulate the low-level "laws" that make intelligence work. Unfortunately, that's a bit outside the purview of science right now. I don't doubt we'll get there eventually, but it's going to be a long and hard road.


In other news, it appears that Blizzard is finally taking us up on some of this stuff. Diablo has always had one nice element of randomization, in that everything except the villages and boss-levels was randomly assembled. However, in both of the games out there, the quest list has been totally fixed. Apparently they're randomizing quests in D3 by not just allowing them to have random starting locations, but also allowing them to randomly chain together (like their levels) if certain criteria are met.

We'll see if they manage to actually ship something interesting, but at the least, it's nice to see a well-funded team of engineers doing a big experiment like that.

[MCP]

Hey, keep at it.
But in the mean time I may be playing the next in the series of the Heroes of Might and Magic:
http://hommkingdoms.uk.ubi.com/

It's an in browser game HOMM MMO. There are a lot of these types of games on the net, but this will have that professional quality behind it.

[daniel.santos]

No sense in letting this thread die. =) I should start by saying that I really like your idea Ken_Oh! =) Too bad I didn't discover this thread sooner.

@Jetryl: I believe it doesn't have to pass Turing's test with a psychologist, Jetryl. It will be just enough, if it passes it when talking with me. What i mean, is we can invent totally different scheme of reasoning, emotions, that just tries to simulate ours. It can be done that without science, but with introspection instead. Perhaps psychologist(s) on dev's team would help.

Emphasis mine. The human mind is so complex that you can't "fake it", as you suggested, by scripting together a series of predefined emotional responses. No matter how many rules you made, you'd still end up with something as wooden and predictable as the combat AIs in most videogames. You have to go straight down to the root, and simulate the low-level "laws" that make intelligence work. Unfortunately, that's a bit outside the purview of science right now. I don't doubt we'll get there eventually, but it's going to be a long and hard road.

Actually, I've done a lot of preliminary work designing just such an engine. I haven't touched it in 2 years or so, but I still have my notes and analysis around here somewhere. Enter the "Personality Engine".

I originally came up with this several years ago when I was disgusted with the behavior of NPCs in Everquest. Either they were not "agro" or they were. If they became agro, they intend to kill you or die trying and they don't care about you otherwise. The only exception is that some NPCs will attempt to flee when they get below 10% health or so, only to try to kill you again when they reach 11%.

But "intelligence" has very little to do with it. If you want to understand human behavior, you have to peel back the human persona and mind one evolutionary layer at a time, from the mammalian to the reptilian. So what we are analyzing isn't human behavior, but the behavior of animal organisms. By properly understanding and analyzing this we can design, and subsequently create, an engine which will, simply put, simulate life. It is the level of detail that should be addressed when discussing feasibility. Further, a good design should be able to functionally simulate the most rudimentary models while allowing development to continually expand the depth of these models. Thus, it can be used before it's really "complete" (if such a thing there be) and would probably evolve (pun optional) for some time if it proves useful.

So what does a reptile want? In short, to survive. If it gets that, it would like to reproduce. So for the reptile, we pretty much have:

• Safety (this mostly comes into play when threatened)
• Food (when hungry)
• To reproduce

I put safety first for a reason, but it certainly doesn't ursup the need to feed all of the time. Part of safety is also preventive.

We know that most of the genetic code between known life forms is common. I don't remember the numbers, but I'm pretty sure that more than 50% of our DNA is common to a mushroom's. Less sophisticated life forms reproduce, but never form an attachment to their offspring or visa-versa -- their young are able to begin fending for themselves as soon as they are hatched, born, etc. Then, life forms began to form attachments to their young so that the mothers didn't have to carry them all the way until they became mature. Instead, the offspring would be born, hatched, etc. and then the mother would care for the young until they were mature, but they did not have to carry them in their bodies -- thus leaving them more able to hunt, graze, etc.

Somewhere up the evolutionary chain, we began to form attachments beyond youth, teaming up with others or our kind in order to achieve our goals of survival and ensuring the propagation of our species -- which is quite handy when those you team up with want the same thing. I personally think that this was a huge evolutionary leap that introduced the very need for the intelligence that humans eventually came to posses -- to manage social needs. Thus, herd mammals and birds evolved by sticking together and this can be found in our DNA (or at least the resulting brain wiring). A study was done a few years ago where subjects were isolated in a room alone with a computer where they met other subjects (online) and played an interactive game while monitoring key areas of brain activity. After an initial time interacting with other subjects, each was (falsely) led to believe that the other subjects did not want to play with them anymore. The rejection registered in the area of the brain that processes the type of pain that effects us (i.e., not the stubbing your toe and ignoring it area) and was equivalent to being poked in the eye! That is because we have evolved to be social creatures and being included in the social group was life or death, that is, those who bonded to a social group are the ones who survived and reproduced.

So I affirm the claims that it's no trivial task, but I reject claims that it's impossible. To make it work, it would definitely need the aid of at least one behavioral psychologist type, but you would be amazed what types of people you can meet on a college campus and what they might be willing to impart. Heck, maybe somebody can be convinced to do their PhD or PsyD work on this! So a very rough summary of human needs would probably look something like this (from primitive to complex):

• Safety (avoid danger -- or when in danger, use fight, flight or freeze responses in attempt to survive it)
• Food (when hungry)
• Avoid danger (find/create shelter, avoid predators, etc.)
• To reproduce
• A social group (for protection, increased chances of food supply, etc.)
• To meet the needs of the social group to ensure it's survival (as a means of ensuring survival of self)
• To ensure needs are met via preventative measures (this is "seeking comfort" -- being well fed, well protected from danger, not exposed to elements which can be harmful)
• To ensure needs of social group are met via preventative measures (we can also call this "compassion", "caring", "sympathy", "love", etc.)
• To learn new things (curiosity)
• To understand and fulfill one's purpose or meaning in life (spirituality)

I'm certain that some smart people can further simplify and refine this list by reducing items to their lowest common denominator.

Now stress is the primary factor that determines which area of this that we focus on at any given point in time. When stress is the highest, we only care about surviving. Part of this evolutionary path has led to a lot of health problems because when we are stressed at a modern day job, our brain chemicals, hard-wired responses (i.e., DNA-driven behavioral responses, not learned) and physiological responses are enabling and moving us to utilize a flight, flight or freeze response -- either kill the threat, run away, or stay still and hope it doesn't notice us. I guess the correct way to say that is that our modern life styles are inconsistent with our genetic programming. Our stressful situations are long lasting and the stress response was intended as an "emergency use only" mechanism -- doctors have well documented the massive toll this has taken on our health. But the stress response, I believe, may be the most important factor in a personality engine to determine what programming the NPC will be accessing at any given point in time. We get stressed when we haven't eaten. We get stressed when we feel threatened (or our attachment to something or someone feels threatened). We even feel some stress from just being uncomfortable. When stress is lower, we are able to consider a larger portion of our needs. In general when stress is higher, only the most important needs are visible to the psyche.

To synthesize the behavior of a living organism, one must track all of these needs and how they are being met as variables. The Sims is a good example (although very simplified) of this type of of simulation.

Next comes associations, which are learned. When I was originally thinking this out, I figured that the code for a game would define parameters for the randomly generated life forms and a simulation of their life would be run feeding events and impressions that were randomly generated within the provided parameters. The idea of this is to shape the individual personality by simulated life experiences. This is how the biases and prejudices are formed which will affect their impulses and the decisions they make them through out the simulation.

But how do you get this back from "life sized" down to something that can reasonably be implemented? That's where the dumbing-down comes in. So far, everything I've discussed is completely abstracted from any specific implementation of anything (game, science experiment, etc.). To make this usable, it needs filters to encapsulate content (i.e., what the author of a "village", "scenario" or even an entire video game will write) into hand-written behavioral modules that are easy to write and reasonable to understand. I'm going to skip over some of the implementation details of the engine right now and jump straight into some fairly abstract (from an implementation format standpoint) examples:

Event Name: Injury
Imprint level (severity in this case): 5-12 (maybe a formula with standard deviation instead of just a range)
Frequency: 10-20 per year
Effects: Safety -5, Anger +3, Fear of loss (life) +3, Fear of loss (health) +4, Comfort -8, Social Need +8
Subjects, Percentage & Imprint adjustment: orc 0.75%/+40, dog 2.5%/+25, peer 10%/+8, self 87.25%/0

Event Name: Aided when Injured
Imprint level: 8-10
Triggered by: Injury
Frequency: 35-50%
Effects: Safety +4, Anger -2, Fear of loss (life) -2.75, Fear of loss (health) -3, Comfort +4, Social Need -7
Subjects, Percentage & Imprint adjustment: mother 50%/0, father 20%/0, clan adult 28%/-1, magic mushroom man 2%/+15


Ok, so this fake snippet says that our life form will get injured 10 to 20 timers per year and that the severity will be between 5 to 12 points each time (maybe on a bell curve or some other type of curve). This will usually have no special subject (to associate the memory with) but expose them to pain. 10% of the time it will be due to a peer and that will add 8 stress points. 2.5% of the time it will be a dog and rarely, 0.75%, it will be an orc attack. If this happens to the subject during the formation simulation, they will have a strong prejudice against orcs, associating a huge loss of safety and comfort with orcs and a desire to belong to the social group. Next, we have an event that's triggered sometimes (35-50%) when the subject is injured. Most of the time (50%), the mother will be the target, 20% the father, 25% a clan adult and rarely, a fabled magic mushroom man will rise from the ground to come to the subject's aid (or replace with any revered species, like elves of the Tolkien world). If this happens during the simulation, especially after an orc attack, the subject will associate mushroom men very highly with safety, comfort and social belonging.

So that's the basic gist of the formation. Next the game implementer will need to build ties between the emotional world of the simulated organism and the real events that can occur in the game. These will need to be done by hand.

Species: Human
Response: Attack
Influence: Anger divided by Restraint times 1-2, Fear of loss (life) times 8, Fear of loss (health) times 4, Comfort times -0.5, Envy (of target's of belongings) times Greed divided by Restraint

This snippet defines what has to happen before a human will attack another creature. The language is a bit ugly above so let me put it into C-style terms (complete with a few fake functions I pretend are implemented elsewhere).

Code: Select all

influence = anger / personality.restraint * randRange(1, 2)
    + fearOfLoss(LOSS_TYPE_LIFE) * 8
    + fearOfLoss(LOSS_TYPE_HEALTH) * 4
    + comfort * -0.5
    + envy(targetBelongings) * personality.greed / personality.restraint;

Now if our subject who got attacked by an orc has a child sees an orc later on, this formula will determine rather or not it will drive them to the point of attacking, which you will implement in your game as you see fit. All of that just to kill a damn orc grunt!

Anyway, that's the rough idea. I gotta get going.

EDIT: Ok, so I had time to re-read it and edit it (clean up typos, etc.).

[Ken_Oh]

Very interesting read. Your talk of emotions will be very useful if/when I attempt a project like this. I've got too many things going through my head right now to make a coherent comment, however, I think we can all agree that any steps away from the simple "agro" system would be improvement.

[Darkmage]

I'm not really familiar with code, but i hope this may help to keep clear some ideas about how city building could be.

You may use civ builder-like code, like new land or the empire builder, in wich you can "step" on the the buildings and then take the actions(the way wesband works) and it looks very wesnoth like stile but in means of the game, if you imagine some kind of orcish/undead raid, i still wonder how could them destroy a building(maybe steping on it, but it seems to me quite quick).

On the other hand you can give buildings HP, this may ease the "raid" problem, but sincerely, aestetically talking it is too rough and confusing to my eye(all those health bars on a city(mean citizens and buildings) it may be not attractive or relaxant) and the way to interact with buildings may be something to develop, but after all this way would look quite different from original wesnoth.

Personally i think the first is quite ok for the idea, and maybe the correct one, but somehow the second one has something i like...

[Ken_Oh]

Yeah, I was working on something like this already. You had about the same thoughts as I did.

http://www.wesnoth.org/forum/viewtopic.php?f=15&t=18567

Villages would be created and grow organically and then would be destroyed by normal actions. However, I'd probably do something to keep certain units on villages when they would normally leave (like, Walking Corpses would stay on villages to produce Walking Corpses from the villages population).

There are a lot of details I'd want to work out, and this would by no means be the easiest part of this whole concept. For example, I'd have to work out things like where the blacksmith lives, and then what happens when the blacksmith dies? Would someone be able to replace him or would the village not have that kind of talent? How would a whole village act when attacked? When would they stay and fight and when would they pick up and run away to the next town?