As a central aspect of all RPGs play behavior is surprisingly unexamined in the context of RPG theory. About three years ago I started trying to piece together a theory of RPGs from the basis of play, and in particular the idea that the underlying motive of RPGs is learning. My major goals were the following:

• A theory that is amenable to experimental validation.
• A theory that scales from single player to larps with hundreds of players.
• A theory that can be directly applied to RPG design.

While it is safe to say I have not achieved all of these goals with this theory, I believe that this approach serves as a foundation for a theory which can meet these goals.

This theory has three main portions: Play Content, Play Controls, and Play Constraints.

Play Content

Play content is the basis of play. It is a set of all the information and objects which may become active during play. Active content is content which is observably being present in play. When this content ceases to be active it become historical content. Some of this content never becomes active, remaining only referred from play, I refer to this as peripheral content, this is reasonably treated as potential future content. Content is not limited to imaginary elements, it incorporates physical space, player histories, and anything which can be brought into play.

In practice the entire set of play content is cumbersomely large. Active content, historical content, and peripheral content are often much more useful as structures to break down play. For example, an RPG book can be treated as a set of peripheral content, which over time is supplanted by a historical content, by becoming active during play. What makes this an interesting process is that the historical content is the recollection of the reference of the book, rather than the book itself.

The first foundation of PCon3 is axiom 1:

(1) The purpose of all RPG play is learning, whether discovery or reinforcement.

This leaves quite a bit of room, since you could be learning how to use a combat system, or reinforcing that you are the best roleplayer or storyteller at the table (whether or not this is true). What is important is what this excludes as a valid purpose, namely aesthetics. The advantage of excising aethetics from the study of RPGs is that this permits us to move towards observable phenomena as the foundation of theory, in this case active content.

So because of our first axiom we are safe basing our models on the active content, and it's relatives peripheral and historical content.

Experimental Aside - The major ways to test play content's relationship with player learning is to test retention. Likewise recording the active content can be very useful, although this approach can have its drawbacks, as the observer may unknowingly apply his or her view (see below) to the data.

Play Controls

Now that we have a place to look for developing a model, the next step is to describe how a player interacts with content. Since each player is attempting to learn from content, they will be activating portions of peripheral and historical content as well as reactivating portions of active content to achieve that goal. However learning is not just the presence of an object or datum within the active content, it is based on the patterns of the play content as different elements become active and then become historical. These patterns allow a player to learn from the play content.

Each player has a projection of play content called a View, which defines the area of interest for that player. The underlying motive of the player is to improve the ability to learn from their view. Clearly this will only work if the patterns of play are present in some way within this view. In particular, axiom 2 describes when play is comprehensible (and hence when learning may occur within a player's view).

(2) A comprehensible sequence of active play content has intermediate complexity (or high epsilon-complexity).

Essentially comprehensible play implies that the play is neither totally random, which produces very high complexity patterns, nor very structured, which produces very low complexity patterns. (Note: I'm using Kolmogorov complexity for this portion, which is simply the minimum algorithm needed to describe the pattern. Epsilon-complexity has the nice property of giving low values to both random and highly structured patterns, by permitting the use of probabilistic algorithms to describe a pattern statistically.)

Play controls are the ways we can model how a player takes their view and attempts to produce intermediate complexity. In essence play controls are simple strategies for choosing content to activate. Controls are innately tied to the player's view, and together they limit what we can know about the player. Thus we often must hypothesize based on activation choices a player's view and controls, and then attempt to test this hypothesis with continued observations. We can even consider testing this

Experimental Aside - Testing play controls is something that lends itself to cognitive science style simulation. By encoding a hypothesis into a simulated play scenario we can learn more about the effects of different controls. Controls are also somewhat amenable to game theoretic analysis, but the multiple conditions (views, complexity, and algorithmic simplicity) can easily make this intractable.

Play Constraints

With each player acting to influence play for their own optimization of learning, the sequence of active content can be hotly contested. As a result active content becomes the observable portion of a dynamical system, driven by each player. These dynamics are helpful, because they can help ensure that the complexity remains high, unfortunately that also runs the risk of making a character's view essentially random.

Over time these shared pressures typically constrain the dynamics of play. These are the play constraints. In essence these are the rules through which play manifests, although they cannot properly exist prior to play. Finding a way to influence play constraints using only peripheral content is the essential goal of game design.

Experimental Aside - Since constraints are empirical observing them is not simply possible, it is necessary.

This is the basis of PCon3. Much more can be discovered by indepth examination of play content patterns, of views and controls, and of constraints.

  - Mendel Schmiedekamp

Your set of criteria looks practical to me. I can see doing lab experiments using it.  I'm certain that it fits in the Big Model - or the Big Model fits in it.

I can see play content including all the written material of the game, but also a lot of unspoken information that players impose on the game to fill in the blanks. In Engle Matrix Games I provide basic scenario information and a suggestion on plot and then rely on human's abilities to mentally form it into a whole - as described in Gestalt psychology. So play content pulls in the sum total of what each player knows. The problem that can arrise is that players see very different things out of the same openning data set - much as happens in politics in the US now.

Play control sounds a lot like "Actual Play" which can be objectively observed. You could video tape game sessions and break down the behavior. You could get at the internal processes players were going through by stopping games and asking them - or during the debriefing session after the game.

Play constraints could be deduced by observing actual play.

Recently I put in an essay on the machine model of games that analyzed games for "static" "friction" "overload" and "waste". which might be qualities one could look for in observations.

While it's not been my impression that the Forge is collectively interested in scientific experiment, given my small ties to the academic gaming world I think applying the ideas to more than just game design is interesting.

Chris Engle
Hamster Press = Engle Matrix Games

Wormwood wrote:
This leaves quite a bit of room, since you could be learning how to use a combat system, or reinforcing that you are the best roleplayer or storyteller at the table (whether or not this is true). What is important is what this excludes as a valid purpose, namely aesthetics. The advantage of excising aethetics from the study of RPGs is that this permits us to move towards observable phenomena as the foundation of theory, in this case active content.

I think, Henry, that he's not saying that people don't play for aesthetic reasons, but simply that this model doesn't address aesthetics instead focusing on analysis by looking at RPG play through the lens of theory about learning.

Interestingly, the boundaries of content were what we were calling "Scope" in a recent thread. Also I think that a lot of this has been covered in some ways by some of John Kim's work, and others who have looked at RPGs in terms of the diegetic framework (a spate of it a while back aimed at clarifying what the SIS is and how it's created). This theory might be more coherent, though I'm not familiar enough with those other works off-hand to say for sure.

I like that these things are testable, too. But what I worry about is that it's going to be hard to discern all or even many of the things that players are learning. That is, what they want to learn, vs what the game seeks to teach. Are there generalized strategies that exist that can cover whole swaths of these things? And that leads to my last worry, which is that the learning theory presented will map over RPGs with any success such that the observations from that field will be of any use in this one.

Further this assumes that the field of study is well put together. I have no idea what the state of learning theory is (though I'm fond of the terms Algorithms and Heuristics). Let us in on what you know about that, Mendel.

I had a feeling we'd see something big from you with your return. :-)

Mike

I second what Mike says.

I can think of a few reasons to focus on behavior over aesthetics. The easiest is that it can be seen. That lends itself to experimentation. Aesthetics is internal. How can you know what is going on with it in an experiment?

A behavioral approach doesn't negate the validity of internal processes. It just operationalizes a study of what can be measured. I encounter this dichotomy in psychotherapy all the time. Do I use a cognitive behavioral approach or a psychodynamic one? Cognitive behavioral has the patina of being more scientific, but both are really arts.

A behavioral study of role playing would "legitimize" it in many people's eyes and would be useful in countering "You're worshiping Satan" positions. One could say - Here is what happens in a game. 50% of the time is used socializing, 20% was on making up events, 10% on sacreficing virgins (oops! not suppose to report that!) etc.

Combined with player's desires for GNS it could show how the machine/game works which could be very interesting.

Chris Engle
Hamster Press = Engle Matrix Games

Henry,

Why people play, and what players get out of a game are very complicated and varied. It is safe to say that players are not solely motivated by aesthetics, nor by learning. But since they are choosing to engage in a play activity it is very reasonable to say that every play has an underlying purpose of learning. Axiom one states that this purpose is sufficient to develop a reasonable model of RPGs, likely because the learning goals (meaning the view chosen under play controls) are influenced by the remainder of that player's motivation. For the basic analysis it doesn't matter why a player chose a particular view, merely that such a view exists. By doing this we can clean out the complex unobservables of play, and begin to focus on what can be observed from it.

Chris,

The experimental nature of PCon3 was an essential part of its development, as it has always been a frustration of mine with most RPG theories. I'm interested in hearing more about your machine model.

As far as the relationship of PCon3 with the Big Model, I was originally trying to make an experimentally amenable version of the Big Model, but after some time I was forced to break from that mold to ensure that PCon3 was able to function. One of the key differences is that as a theory PCon3 doesn't attempt to distinguish between "imagined" and "real" content. For example, a player may be learning about real things (social relationships between players, science, realistic tactics), but the individual content required for this learning may or may not be attributed to the real. From a learning perspective the ultimate reality of what is learned or of the content which enables the learning is irrelevant. On the other hand, the distinctions of SIS and Social Contract in the Big Model provide a context for much of the theory development, but can be artificial constraints on what is being examined with PCon3.  In essence, the two theories are different approaches to solving similar problems. I believe they can be mutually beneficial, filling in gaps of the other, but I wouldn't treat it as an inclusion-containment relationship.

Mike,

Admittedly I'm a distributed / dynamical systems person, rather than a learning theory person. And I've been working through the different theories related to RPGs as quickly as I can (around all the research I get paid to do...). But the way I see learning theory and dynamical systems relating in PCon3 is through models of play controls. A play control is an algorithm for learning within a view. Learning theory presents a collection of viable algorithms for this, usually based on more structured learning. But since we are dealing with multiple players attempting to control play content for their own goals, these algorithms need to be vetted through simulation and/or game-theoretic analysis, or at least have their dynamic interaction properties extracted and understood.

Interestingly enough, this theory also presents a good justification for why people choose to learn from groups of players. Simply put, people are chaotic, neither as structured as simple interaction/response, nor as random as dice or cards. This makes it easier to fulfill the ideal learning complexity, assuming some cooperative constraints emerge. As far as whether axiom 2 is actually a fundamental truth, all I can say for sure is that it is necessary for learning, but I it may not be sufficient. If we knew what is sufficient we would have a far better understanding of how humans think than we do at this time. Essentially, most learning theories attempt to produce high epsilon-complexity, but in different ways, perhaps there exist algorithms for producing high epsilon-complexity which are non-functional for some deep cognitive reasons, but at this point we can't really guess what those might be.

Ultimately I approach the problem as a complex dynamical system, with each player driving a common resource of play content using play controls. The fact that this system has deep structure becomes apparent from the play constraints we observe. Analysis comes from creating local models and testing them against the specific observations or by taking content structures and generalizing them to form heuristics.

  - Mendel Schmiedekamp