Exclusive: Will Wright on Emergent Game Design (Part 1)
This is a multi-part post. Jump to:
One of the lovely things about working for Microsoft is that really cool people show up from time to time. Last week Microsoft Research brought Will Wright in to talk about Emergence, Dynamics, and Design. Unfortunately the talk is Microsoft internal, so I can’t post the slides or any video clips. I did however obtain Will’s permission to write whatever I like about his talk and to use “head shots” from the video, so without further ado…
There was an article put on Gamasutra a few weeks ago on Design Cognition, and on the concept of bottom-up vs. top-down design cognition. While Gilliard and Rafael are trying to touch on things on the meta-design level - how we think about game design - it’s interesting to note that few are the number of games actually produced in a bottom up manner. While the example of Doom is cognitively bottom up in the sense that the entire game exists as an exhibition for the features, I highly doubt it was actually designed in a bottom-up manner.
In fact, I doubt there are very many games designed in a bottom-up manner, Will Wright really being the only designer that comes to mind who does this on a regular basis. Will’s design philosophy stems greatly from emergence - a concept he claimed to learn primarily from playing Go, playing with cellular automata, and ants. This really struck a chord with me, being something of an Evolutionary Biology fan myself.
The concept behind emergence is that by creating some very simple rules and letting them interact with each other, you can get very complex pheonmena to emerge from this.
Ants are an excellent example of this, and much of the first half of Will’s talk focused on some of the particulars of the way ants behave, and how each individual ant obeys very stupid, simple rules, but these rules cause the colony as a whole to act in an intelligent manner. An example of this Will used was that ant larva need to be fed different things at different stages of their growth. To do this efficiently, they need to be sorted. Sorting is a rather advanced concept, but an emergent sorting algorithm occurs in ant colonies by the following mechanism. At the different stages, larva produce a different smell. When an ant comes upon a larvae, if the smell it emits is different than the surrounding area, the ant will pick the larva up. The ant will then wander around essentially randomly until it comes upon an area that smells the same as that larvae, where it will drop it. This simple rule applied across each individual ant in the colony will result in the larva being sorted into like piles.
This is just one example of an emergent phenomenon of several he gave (and if you’re interested in more, you should certainly read up on the fascinating little creatures). The question then is how does this come into play in game design. If you view a game as a possibility space, the act of playing the game is centered on the exploration of this space. Once the space has been explored to the extent the player is willing to spend their time on, they will burn-out on the game and cease to play. It has been incumbent upon designers over time to enlarge the possibility space as much as possible while retaining a high quality experience. This drive for high quality content has ballooned development budgets and staff requirements by orders of magnitude over the last several years causing relatively little increase in the size of that possibility space, and in many cases a shrinkage. Will views what I’ll refer to as Emergent Design as a method for creating extremely large possibility spaces without a comparable development cost.
The major problem with emergence is that it’s very difficult, if not impossible, to design for with any accuracy. The designer brings to bear several game mechanics and allows them to interact in various ways. For example, in conway’s game of life, there are only four rules:
- Any cell with less than 2 neighbours alive dies
- Any cell with more than 3 neighbours alive dies
- Any living cell with 2 or 3 neighbours stays alive
- Any dead cell with exactly 3 neighbours comes to life
From just looking at these rulesets, it’s nearly impossible to tell whether this game will be fun, what the emergent strategies or phenomena will be, or anything else about it. It behooves us to find a mechanism to determine this. In nature, successful genes are those that are able to maintain their existence in a competitive environment. There are no rules that apply to genetics to determine fitness, no fitness function. The only way to determine fitness in life is by allowing it to occur.
The hard part then becomes what the best way is to go about doing this. Upon asking him about this, Will said there were two key components: Smart Interns, and creating many prototypes. The interns create and playtest many, many prototypes in the possibility space of the games they could make. When making Spore, approximately 200 prototypes were made, 60% of which were complete garbage.
Even if it’s not possible to determine the end result in an emergent system, throwing darts at the wall randomly is not an incredibly efficient system. In his many years of experience, Will has noticed certain patterns that he’s adapted into a toolkit to refine the ’search space’ of possible games he wants to make, thus making the likelihood of any given prototype revealing fun gameplay more likely. I’ll share some of these insights in part two.
