Wednesday, June 18, 2008

Science in World of Warcraft

Eleven million people play Blizzard's World of Warcraft game. Blizzard is watching them, and they are recording what they do. Almost everything is recorded. Even obscure things like what is said in chat channels is stored.

In other words, Blizzard knows what you're doing and what you're saying. They keep these records for a long time, too, so let's assume that they can track what you've done over the entire lifetime of your World of Warcraft account.

I'm going to try really hard to ignore all privacy issues here. Feel free to be worried about this, but please be reasonable. This is a game. This isn't even a game marketed to businesses for meetings. Don't spill your corporate secrets in a chat with your coworkers. Anyway, if it is a problem, it's a problem for another blog to investigate.

Instead, I want to look at what sort of science Blizzard could look at by analyzing all this data that they've gathered. There is an enormous amount of information in there that might have some relevance to social science.

I should be fair, though. Blizzard isn't alone. Academics don't have a lot of real world data to play with. Netflix released some of their data, and they ended up getting a lot of free research done for them. Demand for real data is high.

Let's look at some example applications of Blizzard's data.


Blizzard has already expressed a desire to help out two epidemiologists at Rutgers. They looked at an unintended virtual disease that spread through the game a few years ago. They got excited about learning more about how diseases spread by studying this effect. Blizzard was at least talking to them.

In addition, Blizzard is now talking about some new diseases that will come out in their next expansion. They may still be collaborating with the scientists!

How might this work? Somehow, Blizzard has to create a disease that spreads through some interaction between characters in the game.

They have to create a penalty system for the disease that will allow the scientists to gather useful information about how real diseases spread due to these social interactions.

I don't know much about epidemiology. I vaguely understand that they could eliminate some variables by comparing the behavior of one study group with another. There would be a lot of things to eliminate if they want to understand people's behavior in the real world based on what happens in a virtual world, though.

I will never tell someone that an ambitious and worthwhile project is impossible just because I don't see how to do it, however. I love those projects! I wish them luck, and I'm eager to hear what happens.

Social Networks

Duncan Watts, over at Columbia University, does a lot of work on how social networks interconnect. I'll admit that I haven't talked to him about his research in about 10 years, but he's the only social scientist that I know.

He looks at all the social interactions between people in a group and how they connect together. He's done some work on how viral marketing works, for example. Go ahead and click on the previous link. Trust me, it's interesting.

Blizzard's data would be a huge help for him. Blizzard can make extremely accurate measurements of how much two players interact. They can measure almost every interaction between every pair of players in the game.

Given Blizzard's data, Duncan could look at a viral marketing campaign in minute detail. He could analyze every interaction and see why it was successful or not successful.

Furthermore, Blizzard's game is divided into hundreds of almost completely separated servers. Players on one server don't interact much with players on other servers. He could run experiments on multiple servers and get essentially independent results.

The best part is that these interactions are similar to real world interactions. They're in a game, but that's the largest difference. The epidemiologists have a much greater difference to figure out.

Quantifying Fun

I'd be shocked if Blizzard isn't working on this one. Their goal is to make fun games. The only question is how much they're willing to rely on quantitative analysis.

They've got an immense database of how people interact with their world. They should be able to measure the fun that people are having and see what affects it.

The first step would be to make up a metric. If I were them, I'd try to estimate the amount of effort a player is putting into the game at each minute. They might not store all of that information, but they should be able to make some good guesses.

For example, if a player is casting spells as fast as their avatar is able to, then they are likely pushing themselves at least a bit. If a player is motionless for several minutes in between actions, then they are probably not paying close attention.

I'd try to argue that the first person is having more fun than the second person. I'm sure Blizzard can come up with a better metric.

Now look at each activity in the game. Is a player, on average, more active or less active during those activities?

There are literally thousands of activities that could be measured. A variety of correlations could be found.

Would any interesting science come out of this?

I honestly have no idea. I'm certainly not a psychologist, but I have to suspect that, again, the enormous amount of high quality data is going to be better in many ways than a conventional study.

I'd bet that some interesting revelations would be found.

Does This Happen?

Blizzard isn't going to answer that question. They can't afford the public relations mess that would result.

Blizzard's employment website has a few "data analyst" positions, but I don't see anything more specific than the mention of the existence of a data warehouse.

All of the examples that I came up with should be valuable to Blizzard.

The epidemiology study would be some science that they could work on without scaring their customers much. Publicizing research like that would be an enormous public relations feat.

Learning about social networks in their game might allow them to try to influence relationships formed in the game. I'd be willing to bet that a slight increase in the number of friendships created in the game would dramatically increase customer retention.

Making an accurate measurement of the amount of fun people have doing various things in the game would directly impact Blizzard's ability to model the value of new features they're working on. They could tweak old features more accurately as well.

I suspect that Blizzard works on some of this. At the time of this writing, there are four data analysis positions on Blizzard's job site. That's more jobs than Blizzard's new mobile development team has.

Surely a large group like that is analyzing something more interesting than their customers' zip codes?

No comments: