Big Multiplayer Online Games

Game Play & Game Design
#1

What games/tech demos have been done, if any, that are multiplayer games that are targeted at mid-size multiplayer games. By this I mean no less than 200, and no more than perhaps 500 players? Could community unity be helped by smaller community sizes? Can a player feel that their role in the world is more important/significant if the player size is smaller, and more tightly focused on community goals? I think the answer is yes.

Another angle on the Big Multiplayer Online Games (BMOG) idea is huge numbers of AI peoples in a game world instead of players. For example, a god-game where perhaps 8-16 human players control a world of 10,000 or 1,000,000 AI entities each.

Any thoughts?

#2

I’m not sure how many computers could handle 10,000 bot AIs. Maybe modern PCs could deal with 10k very simple units. If that’s what you implied then I suppose a very massive RTS-like game of that sort could be cool.

I’ve heard nothing about what you call a BMOG. Seems to me that most games try to get all the players than can and either divide them into 64 player max games or whatever number of players they can stuff into one world.

I do think a smaller world might help players feel more important to the world. With around 200-500 people, and a very big world, you could have enough room for wandering heros. The story would still not be able to fully revolve around one character though, obviously.

What kind of responses are you looking for exactly? Do you plan on building such a game, or are you just asking to promote thought about current multiplayer game systems?

#3

I’ve never heard of a BMOG either…it’s just a term I came up with to distinguish it from other forms of multiplayer games. I think there are some unique gameplay angles that could be tapped in a BMOG, as opposed to other multiplayer games.

I’m just tyring to still up discussion on the topic. I realize that 10k complex AI units are perhaps unrealistic at this point. It will be something to look at in the next 5 years however. As you may have guessed, I’m not only interested in what I can do today, but in possible directions for games in the future.

I am planning on building a game around these ideas. The ideas are fairly vage at this point, and the potential hurdles that need to be overcome with this sort of design have yet to be discovered. This is one reason I’m putting the idea out there. You’ve already pointed out one major design hurdle, that of the large numbers of AI controlled units.

You make an interesting point, that even in a BMOG (or any multiplayer game for that matter) “story would still not be able to fully revolve around one character”. However, I am interested in what could be done to bring a stronger storytelling element into multiplayer games.

On your last note, I’m looking for any response of any kind. I’m hoping that this idea will get the gears in other people’s head turning. I hope the idea sounds interesting.

One thing I hope that BMOGs will do, if the idea every catches on, is a shift in game design, away from the MMOG ideal, which is to have the in-game population made up almost entirely of real players. I don’t think that for interesting interaction to take place, that you need real people. There’s plenty of bland, boring people in MMOGs. I think that through more clever and interesting AI characters, that a more true melding single player and multiplayer elements can be achieved. The multiplayer elements being the fun of playing with others. The single player elements including such things as reliable, interesting characters to interact with - those who don’t betray you, don’t take all the treasure of a team-killed monster, and those that allow players who like to “go it alone” to more fully enjoy a multiplayer experience.

There’s a lot there. Feel free to start picking at any of my comments.

#4

So basically, this would be runescape + many more AI bots?

#5

Never played Runescape.

However, it’s not about a particular genre, it’s the idea that interesting player/game world interaction can occur - unique interactions that that aren’t possible in single player games (because there’s no other people to mingle with), and not possible in MMOGs (because there’s so many people around that you just feel like a number).

Try not to think about a single implementation. Try not to think about a specific game type. I’m talking about what sorts of interesting player interactions are possible with a large group of people, rather than single player, or massive populations - in a general, theoretical sense. I think this idea has applications in any genre you want to throw at it.]

#6

Yeah, lots of “intelligent” NPCs would be nice… but to my knowledge, there is no AI algorithm available, which could really solve the problem.

AI is mostly done by scripting all possible actions of NPCs, that means you can define reactions to certain events. You can also script the NPCs behaviour for its general live of course, but that would be mainly standard procedures like going to work, eat when hungry or go to sleep.

If you take a closer look at this “actions taken at free will” you will notice, that they are also triggered by other events like time or the NPCs state (which again is changed by time or external events).

It would be really hard to let an NPC come up with an “idea to do something” that is not done in a routine way. Maybe you could implement some kind of trigger (time comes in again) that let an NPC end its daily routine and go for an adventure…

But basically, switching to “adventure mode” just means to switch the daily routine script. While adventuring the NPC might search and slay monsters instead of just “work” (you could even argue that slaying monsters is an adventurers job).

To summarize this: you can define different work patterns and even let one NPC switch between them. Is that what you meant by AI driven players?

#7

You could try something like the creature AI in Black & White… you would have the advantage that you don’t necessarily need the learning part to be in the game… it just has to behave like something that has already been trained so the (slow) training can be done off-line.

I’m thinking Markov models with a finite state machine could produce something reasonable.

#8

In my opinion training is nothing else then building a ruleset. The ruleset is used to define reactions to events. Wheter you use a finite state machine or a binary decision tree (or both) doesn’t matter. That would just be the technical implementation.

Real intelligence would mean, to be able to come up with an idea first and then match this idea to an event, or to compute an idea that was neither “trained” nor “scripted” (which in the end is both the same).

Of course, this doesn’t seam to be achievable by now. So there is the need for the best way to fake intelligence. I was thinking about an approach that is also used in robotics. Why not build an NPC like the human they represent? That means: give him sensors.

Sensors allone can’t make the NPC intelligent but they would let him look like he has more brains :wink: At least at first sight. NPC with sensors could for example be given commands like “guard that object” without the need to define waypoints or any further script/training.

Maybe a NPC could also have something like a character. This character could bring him to unforseen actions and these actions would again cause other actions that influence the NPC in return…

#9

That’s what I meant.

Markov models which will base actions on some sort of probability tables combined with input from the environment.

e.g. time since character was last in a fight make make them act more agressive to the next person than interacted with them. Hunger of character would affect probability of taking certain actions etc…

The probability tables would at least initially be computed off-line to get the characters behaving in a believable manner. If possible, the weights in the probability tables would be adjusted over time so that the environment can mold the personality of the character… that is similar to the creature in Black & White… but I think is is a faily complex thing to work out.

You could also use genetic algorithms to mutate probabilities in random ways so that the behaviour ultimately was not that of the ‘trained’ player but evolved to something of it’s own… Using environmental influences to make those adjustments and you coudl perhaps come up with some characters that seem very complex. even if they ultimately are simple autonomous state machines that effectively choose random actions but with the probabilities governed by a Markov model…

It is certainly not a simple thing to build the models in the first place, of course.

#10

Well some really massive multiplayer games exists, like Eve Online that sports a playerbase of around 10k people simultaneously connected in a huge universe. It will aqcuire the result in a surprisingly elegant yet simple manner (almost anything is player-based and the economy driven world is modeled inside a gargantuan database).
And it’s one of the most innovative and fresh MMORPG around the lot. No powerleveling at all, for example.

The matters about 1 million AI is a bit out of scope IMHO. If you’re thinking of so much AI you can use a statistical/probabilistical model to approximate individual behaviours and evolution in the behaviour itself. In either case the number is so huge that the single particle is not of the same magnitude order of the whole, so it shouldn’t be modeled as a single entity.

I think that you may find some answers to your questins reading something about full scale analisys of social and political behaviours to learn how to model a “model of masses”.

#11

Bots could be graded. Thousands of cheap dumb pawns, hundreds of automatic (mission) story-line generating actors, and a handful of key, scripted actors to give context and structure to large scale story lines written by admin types.

Game size limited not just performance based, but on available nof writers. Keep ratio of writers / players constant?

Complex, fluid and involving storylines are possible - keeping players interested, like following soaps or good books. A creative as opposed to technical problem that, tho, but BMOG offers more potential as a platform that could support it.

Possibly.

#12

The Markov models sound interesting. I will look at this closer. Does anybody have a good source for that?

[quote]The matters about 1 million AI is a bit out of scope IMHO. If you’re thinking of so much AI you can use a statistical/probabilistical model to approximate individual behaviours and evolution in the behaviour itself.
[/quote]
Statistical models would be useful to simulate an economy or some worldwide political events, but that is something “invisible” to the player. More like a concept. The background. It’s a different thing with autonomous NPCs I think. They should interact with players directly and therefore be as intelligent as possible to keep things interesting.

I don’t think it would be wise to hire story writers to substitute interesting players (either human or NPCs). It would be to much effort and wouldn’t that give the player a feeling of beeing manipulated? The story could only be a framework. Perhaps you could think of this story framework as the political events and simulate it with an appropriate statistical model.

What about this idea: we give all NPC some “goals” they would like to reach in their lifetime (like humans have them too). For example: to rule others, get rich, work as less as possible, be loved. And then let them look out for an opportunity to achieve their goals. All kinds of opportunities must be told (coded) to the NPC of course, so he could recognise them.

#13

As far as I remember, WETA has creating similar system for visualisation of LotR battles (called MASSIVE). Output is a model ready for visualization, but concept behind is a large number of independent agents, given some common goal, but with small variations in characteristics and possibly quite large in environment at given moment. From what I have read, every orc was calculating situation around it independent - some of them where more brave and thus run faster, some would start to retreat a second before the others, smaller ones would make a path for bigger one, etc.

Goals in MASSIVE where rather short-term (lay a siege, kill as many people as possible, stay alive for next few hours) and focus was mainly on animation part of output, but I think that this is a good prototype for system you are talking about. They have developed it for 5 years in large team… Even given the CG aspect which does not have to apply for game with such quality, it is a major undertaking - by major, I mean something probably not possible on current game budgets.

#14

[quote]As far as I remember, WETA has creating similar system for visualisation of LotR battles (called MASSIVE). Output is a model ready for visualization, but concept behind is a large number of independent agents, given some common goal, but with small variations in characteristics and possibly quite large in environment at given moment. From what I have read, every orc was calculating situation around it independent - some of them where more brave and thus run faster, some would start to retreat a second before the others, smaller ones would make a path for bigger one, etc.

Goals in MASSIVE where rather short-term (lay a siege, kill as many people as possible, stay alive for next few hours) and focus was mainly on animation part of output, but I think that this is a good prototype for system you are talking about. They have developed it for 5 years in large team… Even given the CG aspect which does not have to apply for game with such quality, it is a major undertaking - by major, I mean something probably not possible on current game budgets.
[/quote]
Sounds like simple flocking (c.f. Craig Reynolds work on “Boids” since 1986).

Let me just point out that “making lots of people run convincingly as individuals” is about 10,000 times simpler than “making lots of NPCs behave even slightly interestingly”.

Using CR’s work, you could easily do the core logic behind LoTR’s crowd scenes as an individual in fairly short time. It’s really really easy to make this convincing (in fact, we’re using exactly the same technique in our 50-100hr game survivor we’ve been doing for the Sun competition - it’s that easy :)). I would imagine they had to spend 99% of their time creating all the animations (making even just one person run convincingly is hard :frowning: ).

Anyway, I was unimpressed by much of WETA’s CG - other people have done much more interesting stuff in much less time with far fewer people IMHO. For instance, take a look at the Creative Assembly’s Total War games. These have had massive battles running in real time on home computers for 5 years where every individual - in fact, even down to each and every ARROW (pause the game, and you can move the camera round and inspect each arrow in flight) - on the battlefield is modelled separately (and typically you have thousands of combatants, even 5 years ago!).

So, they’re doing the same stuff, only it’s in real time on home computers and there’s actually game-logic and rules and hitpoints etc behind it all. And they’re “only” using it for games. The only thing WETA did that was impressive was to add extremely good animations and lots of very high detail rendering. Shrug. I think they get a lot more credit than they earned just because their stuff looks so pretty :P.

#15

PS If you’re really interested in this, you should join MUD-DEV (just do a google search and you’ll find it instantly). It’s where all the MMOG developers from Sony etc hang out. Most of the above has also been discussed at length over the last 7 years on MD, and you will probably find all sorts of interesting new thoughts and ideas on there, given the high concentration of specialists participating.

#16

I’ll check it out.