« Evolving Virtual Organisms to create Virtual Life | Main | Altered States »

Jun 02, 2007



Overall it would seem such simulations would be overly simplified and not model the more interesting aspects involved in manufacturing and production.

On the one side you often have buisnesses that are the sole supplier such as pet food production, or you have multiple manufacturers but they produce slightly different products using different distribution models. In the latter case usually consumer themselves drive the market, preferring ipods over zunes or internet shopping over brick and mortar stores.

In addition in todays global marketplace you have a host of regional effects that help this such as local rules/regulations (pollution/taxes), cost of living (taxes/wages), employees, etc that provide a cost differential in product costs.

So a more realistic simulation would be more encompassing that the simple distribution scheme outlined above.

I would argue that an existing game such as WOW already exhibits many of the interseting characteristics already, such as the fundamental supply/demand. How many producers are there and what is the demand for an item, and how does that cost vary overtime. Are there ways to manipulate the price, such as increasing the number of components to make an item (or decreasing their cost)?


Phenomena like oligopoly are rich, complicated and are not well suited to be studied in isolation. Oligopoly itself is a manifestation in a mature market typically. That is, it takes time to form, is not formed by random participants, and rarely if ever occurs instantaneously. Any oligopoly created by instantiation in a VW would be a poor representation of the actual phenomenon.

As a former chemist, I greatly appreciate the desire to have fixed inputs in experiments. But the greatest potential for research in virtual worlds is the ability to perfectly study (within the confines of the VW physics) human beings in situ. However to maximize the potential therein the virtual world has to become more expansive and less controlled.

Studies can still be conducted with strict entrance criteria. The researcher simply has to find the VM residences that meet those criteria, which can be done automatically through the game data itself.


Landon puts his knuckle right on the node. There are two kinds of science: analytic (chemistry, with apologies to Landon), and historic (geology, yay!)

An analytic scientist might want to isolate one person and add other players as variables to see what happens. An historic scientist would just look at what actually *did* happen, and try to explain it reasonably. I wonder if TN ppls would vote on which kind of science they are into.


A significant factor in Oligopolistic behavior is based upon the strategic actions of the oligopolists' vis-a-vis game-theory. As mentioned above, such equilibria do not arise spontaneously, but are the result of a process of moves and counter moves. The system in equilibrium state is itself inherently unstable and prone to defection and tit-for-tat punishment.

The problem I see attempting to study such a system in a VW is that to eliminate this dynamic property (rigidly enforce the equilibrium state by fiat) would render the results irrelevant. However, allowing dynamic strategic behavior will suffer from meta-gaming, which in a VW environment is very different than in a RL environment. In RL, oligopolists meta-game by doing things like hiring the same lobbyists and sending managers to the same MBA programs. In a VW oligopolists will meta-game by scripting, botting and other direct-manipulation techniques.


Nice example project! And, given my biases, I think it's a nice example because it illustrates why VW's might be a good platform, and a laboratory is a good environment, but the glovatar approach probably is a hybrid which inherits more weaknesses than strengths. Now, it may be that there's some carefully-designed interplay between the instance-game and the VW which would make it useful (just as a scientist might want a glovebox which lets in and out visible spectra but not higher frequencies, or which lets air in/out, but not a claw).

First, here's something which I don't think is much of a problem:
>> What if one were a newbie and had no idea what they were doing, or had played the quest many times and therefore were ringers?

This is a potential problem with laboratory experiments, too. It's always quite likely that different subjects have different relevant experience / expertise. (This is particularly a problem when there's a core of "professional subjects"). Usually it seems like the problem is solvable by putting subjects through the experiment just a few times, with the differences shrinking quickly. Of course, this approach is also usually done with pretty simple environments. Anyway, what I'm saying here is that this isn't much more of a problem than in standard lab experiments.

Now, on to problems. Here's one obvious thing that could happen:
- Joe enters the glovatar, and find out about the experiment in question. He IM's Betsy and Jens to see if they're in the same experiment. They chat about how the experiment looks. If they like it, they stay and collude via IM. If they don't like it, they quit and either try again or go kill some orcs.

This scenario describes several of the problems with doing VW experiments which don't happen in standard lab experiments. To be fair, the problems could well occur with any internet experiment, not just VW ones. In any case, there's communication outside of your control and self-selection in your subjects. (I suppose the experimenter should feel lucky that Joe isn't using a mule on a second computer!)

(This is completely ignoring the lack of statistical independence of your samples, which I think is an insurmountable problem for VW research.)

So, the largest experimental problem isn't that you're giving up control over geography/production functions/whatever. It's that you're giving up the control that you get from having subjects sitting in a lab in front of you.

And I don't see that much is gained from doing the project in a VW rather than the lab. Manufacture in VW's has "production functions" in much the same way that a family-operated business has a production function. You could argue that there is one, but nobody would ever be able to identify it unless (for instance) one could quantify how much fun it is to be killing imps, or standing out back of the shop chatting and smoking cigarettes. Travel costs are inherent to VW's with meaningful travel, but again, you can't identify exactly what they are.

If you do try to make things like demand, travel costs, production functions, etc. concrete, then you're making a laboratory embedded in the game. It's going to be *essentially* very different from the broader game. And it's going to have control problems the laboratory doesn't have.

On the other hand whomever (above) and Landon point out that VW's have rich properties naturally occurring (I need to practice using "in situ"!) which could make study of oligopolies interesting. You don't identify a research question, but for many questions you could establish just enough constraint--on the whole world--that the study would address your question. Any issues of collusion which arise are then part of the experiment, just as they could arise IRL.

That approach has huge costs and drawbacks as well. The huge cost comes from needing to have several entire servers with a fixed ruleset specific to one experiment. Those rules could change with time, but it's still a problem. However, I think that's the kind of problem that's worthwhile.


There are a lot of options in your post up above:

“The simplest models would also make assumptions about where the businesses are located (snip) More complex models would allow the businesses to choose…”

“It could simply be a quest...”

“Would there be one other player or 10?”

As a result we all post in reaction to the options we find interesting, either because we think they might succeed or that they might fail. But we can’t really tell what you are proposing so our thoughts and discussion may be on the mark or they may not.

Timothy points out that you don’t identify a research question.

Your also pretty vague about who would play the game. Grad students? Players already in WoW or a similar game? Business people?

Why would they play the game?

Will you build a game or study an existing game?

There are at least two (and probably more) ways to approach the question.

One is to follow the standard academic research model of designing the experiment and recruit subjects either by recruiting them as volunteers or paying them.

Another would be to identify a market need and fill it with knowledge gained from a VW environment, either from an existing game or from a purpose built game.

If you want to study oligopolies find a market that needs to know more about them. In the US there are a great many policy makers grappling with the healthcare system. A VW environment that helped them make better policy decisions could be very valuable to them.

There is great potential for research in these environments because they are very rich and generate wonderful data sets. I find myself having the same problem that your having, the possibilities are so great that it is hard to define the project, the problem or the value, not because there are none but because there are so many.

I do think it is possible to build an experiment that is compelling, and that careful thought and design can avoid many of the red flags that others are raising. I also think that will be expensive.

Keep posting, as you may have guessed I really enjoy the topic


@randolfe_: Well put.

@CherryBomb: Well, I know the same distinction under a different set of terms -- experimental vs. exploratory science (and, to answer your question, I'm all about the exploratory :-) ). For a thread that (as the comments unfold) reveals some positions by TN'rs on this, see here (the conversation turns to this issue around here).


Hmmm... I was actually concidering something like this for my honours project.

As is mentioned before by Timothy, "This is completely ignoring the lack of statistical independence of your samples, which I think is an insurmountable problem for VW research." I feel that doing research in VWs are unique to other types of experiments in that, we are interested in non-independent samples.

If we look at virtual worlds, there's always a way to communicate with other players about the game, whether in-game or through forums or websites. They discuss methods to advance through the game faster, collect special items, etc.

In the real world, we also discuss things similar in nature to the virtual world. We talk with other people about what cars are better, where to eat for good italian, etc.

VWs are great for producing this sort of data. The problem that was mentioned about newbies and ringers is easily solved when these two groups come together and discuss ideas about the game.

An instance type setting was also mentioned by Robert as a way to control the number of people participating in an experiment but included an issue about a player with a lot of money/skill/capital goods that might interfere with the experiment. Given the name of a "glovatar", the idea is similar to Sony PS3's "Home" which has mini-games within it, allowing users to play mini-games with their friends. These mini-games could have set input values that limit them to a certain amount of resources.


This sounds like a nice improvement of the proposal in Virtual Worlds, Real Rules, 1 N.Y.L. SCH. L. REV. 103 (2004), available at www.nyls.edu/pdfs/v49n1p103-146.pdf

Note, though, that to get meaningful results you'll need to control who the players are so that different shards/iterations have at last similar populations in terms of nationality, age, wealth etc.


Turbine has built this mini-game into LOTR-online. Each player has 3 "professions", only two of which relate to each other. The odd profession requires input of raw materials from other players to be able to complete the "quests" and level up skill. Since LOTR has a such a large RP population, crafting is, for many people, the end game. Some (according to the forums) only fight/kill to get materials. The end-game for them is the business of crafting, and being able to produce end-game quality products. . Crafters guilds are cropping up, with finished products provided to the players-at-large, or to the Auction House for a profit.

The comments to this entry are closed.