If Only Game Designers Were Scientists

Today’s New York Times Magazine featured an article entitled “Where Do Dwarf-Eating Carp Come From?”. The article describes the quirky aspirations of Tarn Adams, the programmer behind the underground computer game called Dwarf Fortress. Like other simulation-based games, Dwarf Fortress allows players to engineer a society (in this case made of dwarves) that faces adversity (in this case hordes of marauding non-dwarf creatures). But if the article is accurate, Dwarf Fortress represents a very different kind of simulation game.

Besides describing Adams’ fervent dedication to constantly expanding the game, the article points out a number of features of his creation that make Dwarf Fortress very different from most computer games. The game is rendered not in visually-realistic three-dimensional graphics but simple ASCII symbols of various colors. There is no interaction with other players, as the entire world is commanded by a single player. There is no end-goal of the game, as the only real gauge of success in the game is playing for an extended period of time: in the end, your fortress will be destroyed and your game will be over. And most notably, there is absolutely no story line that Dwarf Fortress follows because the world in which the game takes place is generated by algorithms underlain by stochastic processes.

I am intrigued. It has always been my contention — especially after working with Dylan Moore on the fieldTest project — that computer programmers who work on games are far ahead of ecologists and evolutionary biologists in the complexity of the worlds they simulate. There is a growing community of scientists who use individual-based (or agent-based) models to explore system-level questions. But most of these models are pathetically simple, because most modelers do not have the programming support to create more complex models. Sometimes the simplicity of these ecological/evolutionary models is embarrassing: I have reviewed two submitted papers featuring models that took place on torus-like spaces with wrap-around edges reminiscent of early video games like Pac-Man. So it would be really cool if those of us in the business of creating agent-based biological models could learn a thing or two from computer game design.

While this sort of cross-pollination sounds good in theory, in practice it rarely happens. In part this is due to the fact that most biologists work in academic isolation from other disciplines. But there is also the issue of conflicting goals: game designers want to create worlds that people enjoy exploring (and will pay for that privilege), while researchers using simulations seek to understand the properties of these worlds. As a result, most simulation-based games really do not offer much opportunity for discovery because they are ‘canalized’ to maintain a particular narrative and to keep the game fun: there is some room for random happenstance in the games, but most of the game is deterministic in nature. A perfect example of this sort of canalization in simulation-based games is Will Wright‘s Spore [1, 2], ironically designed to be a game about evolution. As others have pointed out, Spore is a terrible model of how evolution works because it makes a fatal assumption: that life evolves along a predictable path. While the development of one’s particular world of life may vary slightly from game to game, the progression is always the same: not surprisingly, if you play the game well, you manage to evolve a world of sentient, social, cooperative creatures who eventually colonize space. Whether Wright is aware of Stephen Jay Gould‘s famous “tape of life” argument (Gould 1990) regarding the limits of convergent evolution or not, his game clearly ignores this argument. Rather than allow a world to unfold in a free manner based on its own algorithms (which is how I conceive of the evolutionary process), Spore enforces a very limiting convergence on players. To me, this makes the game scientifically uninteresting.

And here is where Dwarf Fortress is intriguing. Based on the article in the Times, it seems like a game of Dwarf Fortress is much more open. As game creator, Adams has given up a fair amount of control. He even comments that a particular play of the game sometimes develops in ways he never could have anticipated (hence the dwarf-eating carp). That the game is still interesting to play despite this open-endedness is fascinating.  I am curious just how much control he has exerted on the game. It has to be the case that many alternative versions of Dwarf Fortess would be interminably boring to play. For instance, if the enemies of the dwarves were capable of simply overrunning their colonies, this would be a very short and very boring game. To make an interesting game, it has to rest on the edge of two very boring worlds: one in which survival is nearly impossible and one in which survival is nearly inevitable. A great number of factors in combination or in isolation could drag the game into either of these unacceptable realms, so the algorithms that rule the Dwarf Fortress world must be tweaked exactly in order to make a successful game. In the world of modeling, we call this ‘exploration of parameter space’, and frequently life as we know it only can exist within the boundaries of a small set of parameter values. I am really curious how Adams addresses this issue in his continual expansion of Dwarf Fortress, because it seems likely to me that the addition of new features would frequently make the game too easy or too hard to play. There could be a lot to learn from Dwarf Fortress or similar games about system stability. It makes me wish that Tarn Adams would decide to take on an ecological or evolutionary simulation world, because there might be a lot to learn from his ‘strictly algorithm’ approach.