Emergence and Game Based Learning
In game design there is a commonly made distinction between “emergence” and “scripting”, but the distinction is often poorly explained. Emergence is often treated as some kind of ‘magic’ that just happens (or fails to happen) when a system is complicated enough. Or else is just a term used to explain anything unexpected in a game or unintended by the designer. We are only beginning to understand reliable ways to engineer emergence deliberately, with specific goals in mind.
Rather than speak of ’emergence’, I see a more useful distinction between ‘endogenous’ and ‘exogenous’ variables in economics. The exogenous variables are those whose value is imposed from outside the system, while the ‘endogenous’ variables arise from the system itself. So, for example, in an economics problem the supply and demand curves are often exogenous (externally imposed) but the price is endogenous (the outcome of balancing supply and demand).
Because these two words look so annoyingly similar, I prefer to use ‘intrinsic‘ (endogenous, internal) and ‘extrinsic‘ (exogenous, external) instead, which have roughly the same meaning.
Consider jumping in a platform game. How far does a jump take you? This may be intrinsic or extrinsic depending on the game. A simple platformer will have a discrete ‘jump’ action which will move you through a prescribed arc. In this case, the jump distance is extrinsic — it is a number externally imposed by the designer. A more complex platformer might have a richer physics model which is used to plot the jump. The result will then depend on a number of interacting variables in the system — i.e. it is intrinsic.
Two levels of interpretation
Often there is a layer of abstraction between the extrinsic and intrinsic variables of the game. The game simulates some aspect (eg physics) at a fine grained level (velocities, forces), but the player observes and interprets it at a more abstract level (character movement). We recognise patterns at the higher level, but these patterns are not explicitly represented in the game rules. We call this ’emergence’, but it is really a feature of our abilities as pattern recognisers than a feature of the game. When the game is too complex for us to grasp at both levels simultaneously, it can seem like magic.
Where these two-level of representation exist, the game can feel more ‘open‘ — that is it can provide a large possibility space for action with subtlety. The opposite it a ‘closed’ game, one where the outcomes of actions are extrinsic – press the jump button and you do the jump action, press the punch button and you punch. Many ‘street fighter’ games work this way, they have rich ‘combo’ sets, but each combo is extrinsically defined – you press this precise combination of buttons and it plays that scripted animation.
Some kinds of systems are more amenable to this kind of two-level representation. Physics, for example, is easy to simulate at a low-level but computer graphics enables us to observe the results at a much higher level. Notice that this is a property of both the simulation and the representation. The same calculations presented as a table of numbers would not feel ’emergent’ as the patterns would be obscured by the representation. Similarly, if the results were rendered as text (“The ball bounces off the table”) the patterns would be codified and made rigid, and the subtlety is lost. The important thing is that the graphical representation allows us to see both levels at once: the patterns and the detail.
One of the main failings of computer games has been representing social interaction without this two-level approach. Dialogue in games feels clunky because it is a single-layer system and our choices are all extrinsically defined. The failure is not for lack of effort, but we simply do not know the ‘equations’ that represent the low-level physics of social interaction, nor do we have the ability to input or output them with subtlety. Language is capable of subtle shades of meaning and a real dialogue game would be about playing with those details — an opportunity which current day dialogue trees completely lack.
Story sits in an interesting place in this discussion. Many games these days have extrinsic narratives – i.e. stories imposed on them by the designer, but often these narratives are at odds with the intrinsic narratives that arise out of the gameplay. For example, a character may tell you that a certain mission is desperately urgent, but the game does not apply any time limit to completing it and the result is the same if you take minutes, hours or days to do so. This doesn’t have to be the case. The game can create intrinsic urgency as well by varying the outcome based on the time the player takes.
Again the problem is one of levels. The extrinsic narrative is written at the high-level of “story events”. The intrinsic narrative is written at the lower level of game mechanics. We don’t yet know how to make these two levels mesh together, so they tend to sit side-by-side in our game and only coincide more-or-less by accident. Making these two levels talk to one another is, in my opinion, one of the key challenges of narrative AI.
Intrinsic outcomes in games are often considered more interesting than extrinsic ones. I believe this is because extrinsic outcomes betray the hand of the designer and break the suspension of disbelief.
Games based learning
How is this relevant to games-based learning? I believe that games are best suited for familiarising the learner with processes, rather than teaching them facts. A game allows the player to directly interact with a system, to play with it and observe the results. This allows the player to ‘get to know’ the system rather than just ‘learn about’ it. So, for example, playing with a market model and observing how and why prices go up and down can provide a better intuition for markets than a text-book description of supply and demand curves. The Kolb learning cycle starts first with concrete experience, which leads to reflection and abstract conceptualisation, which informs experimentation leading to new experience. A game can facilitate this cycle in learning by providing a space for concrete experience with a system.
This means that a learning game should be most effective when the concepts learnt about are intrinsic properties of the system. Otherwise the game is nothing more than a textbook in (poor) disguise. Contrast Trivial Pursuit and Monopoly. The gameplay in Trivial Pursuit teaches you nothing about its topic matter. It is just an arbitrary competition with extrinsic answers. The gameplay of Monopoly on the other hand, gives the player an intrinsic familiarity with wealth accumulation and bankruptcy, because these things are emergent properties of the system. It supports reflection and theorisation because there is a complex system on which a theory could be built. Trivial pursuit supports no theorisation because the answer is only ever right or wrong, and has no other foundation.
This creates a design problem for serious game designers: the level of representation. If the level of representation is too flat, if the abstract theory is extrinsically encoded in the rules, then there is no meat for the players to chew on. Such games can often feel like propaganda: they impose a certain point of view rather than presenting an argument for it based on first principles. But representing real low-level systems is hard. A bad approximation made yield to a completely different high-level representation, leading the player to false conclusions. When designing for emergence, the designer gives up some of their control over the result. Often this leads to unexpected high-level behaviours. In a game for entertainment, this may make the game more or less fun. In a game for learning it can be worse: it can completely mislead the learner.