« The public face of computer gamers | Main | New York Times on Permadeath »

Mar 03, 2004



Several years ago I proposed that you could create agents mirroring playstyles and patterns and let them loose in your actual MMO database to attempt to model things like min-maxing of equipment, areas or zones that are underutilized (a very common problem), population centers, etc.

But it's a non-trivial amount of modeling. You'd need to have a baseline for comparison, which would presumably be an already running game, along with collections of data to match the experimental results to. Then you'd iterate your agent behaviors and models until they matches the real game to a certain margin of error. Then you'd attempt to test changes within the world on that agent-driven testbed.

Appying it to another game, or even to the same game after accumulated changes, would be risky, though. After all, games select for certain user types and behaviors. It's quite likely that a generalized model would be difficult to develop.

If you could get it to the point where the agents evinced boredom, it would be a hellaciously powerful tool for churn management and game longevity. ;)


"What if we have to take a test before we are allowed to play the game?"

Like the Gypsy questions in Ultima 4... But this time, no tests. The best way is to make it an integral part of the gameplay. Watch the user's actions and determine his chosen answers based on what he does. Put him in the hot seat, tight spot, corner, whatever and see what they do. It doesn't have to stop there, the VW should be constantly re-evaluating the parameters of each player as they go about their game.
Take it one step further and use ths data to tailor the content provided to that particular user, so while your VW might be huge, filled with thousands of locations, quests, players, etc. you bring those that are most likely to appeal to a particular user one step closer to him so they stand out, engage him, drive him to the kind of experience he prefers, always keeping an eye for a change of pattern.

Ah... The game that knows what I want even before I do and gives it to me from a huge repertoire. ... Where do I sign up?


Raph>Several years ago I proposed that you could create agents mirroring playstyles and patterns and let them loose in your actual MMO database to attempt to model things like min-maxing of equipment, areas or zones that are underutilized (a very common problem), population centers, etc.

Isn't this the same problem as designing good MMOG AI? The point being - not a trivial problem.


Oh, I didn't say it would be easy. It would, in fact, be monstrously difficult. :)


Yeah, that's what I am saying too! If we could simulate at that level, heck, we'd use the technology to actually build in game elements. And the problem is, modeling that kind of complexity is really hard.

I would if evolutionary algorithms are a possibility here? I think was once suggested to me by Fred Hapgood, a technology writer. Plants seeds for trees and folks and use simple rules to grow them generation by generation.


"Plants seeds for trees and folks and use simple rules to grow them generation by generation."

Looking at Linden Script this certainly seems doable in Second Life.


You could certainly use genetic algorithms to try to build the agents, but then you wouldn't be assured that they'd actually develop into something that served as an adequate simulation of a playerbase.


Are there devs or academics who have tried to model a complete MMOG abstractly? E.g. as a system of rules that can be analyzed/executed separately from the game implementation itself? Building sims against abstract models would be a heck of a lot easier.

I suppose there might two possibilities along these lines: devs who devs who abstractly model the game completely first, then test and build implementation; or academics who reverse-engineering an existing implementation. Or is everyone in such a rush, that the implementation is the game.



Nathan> Or is everyone in such a rush, that the implementation is the game.

Yup. This also falls out of the fact that, per Raph's comments, that a usable simulation may be as hard or harder than actually building the game. Thus iterative design becomes the norm and the dev spend time (hopefully) building a system that allows rapid changes to be made to the design and code-base.

Also, to reinforce Raph's comment on GA's, while GA's are really interesting and have numerous applications in game development, using them to actually grow agents that approximate human activity in a complex MMOG would be an extremely challenging project. It's scope would most likely dwarf that of any MMO to date.


One issue I take with this is that despite the handy "grouping" system developers have for players... very few can be classified as just "Killers" or just "Achievers"... most all players enjoy all of those various elements to one degree or another. This is why those labels are routinely applied on a percentage basis and not as just a simple label.

If you build your game with the idea that every player can be lumped into one type or another you're setting yourself up for failure.


Nathan > Can we model the vast-dimensional glory of players and their social networks in an MMOG?

Modelling the fullness of a living vaste game universe with vast dimentions of players and their social networks would be anthropological research in the ‘field’ of virtual worlds instead of in fields like remote islands, slum-areas (The Social Order of the Slum, that’s a classic) or specific groups of people in NY. PPl in the social sciences seem to be attracted to virtual worlds as test environments for studying human behavior in a setting that is very controllable and possible to design. (check out Emotions in Humans and Artefacts, ed. R. Trappl, P. Petta, and S. Payr, Massachusetts Institute of Technology, The MIT Press, 2002 – especially Bellmans “Emotions: Meaninful Mappings Between the Individual and Its World”) Im not a psychologist myself so I find it exhilarating to read about these methods.

But that’s not the issue here, right, but to model the possibilities and funding frameworks for players and social networks in all its complexity, wouldn’t that be totally cool. It would be easier to take the initial data from the player, but possible to get it out from the player in-game, as divineShadow prefers. Much more fun to play, I would prefer it as a player, but not as a dev who just wants to test the idea.

The thing is that most discussions about this are so much geared towards NPCs and autonomous agents. That’s fine for single player games, but with MMOGs... The more of oneself one put into the game as a player in the form of engagement the more fun it gets. If I don’t care how my family is doing in the sims or if I don’t care about how many foodthingies I can gobble when im pacman, the game isn’t much fun. Usually the triggers that makes a player find a game as fun are quite small – but they have to be the right ones. If one are able to catch the player and get her whole attention, one also have access to the enormous sofistication and complexity of the human brain. That we should use more.

Natanthan > […] the lens would peer upon the player more thoroughly.
Is this just a fantasy? And if not, do we really want to go there?

I might be able to check if it is a fantasy or not – at leas at some level. I just started a five year project where ill (if nothing stops me, or if I not somehow realize it’s a bad idea and do something else, I mean - who can ever know) build a semi-autonmous agent for the player to possess. I have no idea if we want to go there… really… But its too intriguing not to explore I think. The first iteration of it is described in a paper I wrote with a collegue last autumn. In short im building a mind module which is to provide a system with emotional output from the individual player character. The mind module performs computational operations on the input values, which come from virtual sensors defined at various levels of abstraction, and outputs in the form of emotional reactions and/or potential emotional reactions that in turn become inputs to the sensors of the mind modules of surrounding entities, or entities that in some other way are receptive (e.g. if they belong to the same social grouping) to the specific player character.


"To Mere MMO or not, that is the question."

There is a great opportunity to study social and micro-cosm cultures like never before using the MMO as fertile soil, and a global outreach.

We've already begun seeding this process by introducing the world's first "player accountability" system. This blended with a true form of role playing that is ever-changing rather than "persistent" will help to evolve the genre where it needs to go, in order to be more than just a game, and more than a source of revenue.

By persistently allowing new people with new ideas to direct the growth of the game over time, a mere MMO can evolve toward the "virtual world" that evades many products being developed today.

In our wilderness survival sim set in Great Basin country during the Silver Rush of 1859, we can virtually turn players loose into an environment where one culture says "we were here first" while another culture seeks "valuable leads." This kind of project demands the most interactive environment ever created for a computer game, and also requires support from educational communities in order to build in the neccessary tracking tools from the ground up, which could provide the direct and indirect feedback your looking for.

These kind of dynamics will afford a ton of new material for anyone who wishes to study or teach future MMO game theory. ;)

The comments to this entry are closed.