Formal Abstract Design Tools
In the lecture preceding Doug Church's, Marc LeBlanc sought out a different approach to game design. Under the label of Formal Abstract Design Tools (FADT), LeBlanc, Church and others are trying to establish rules, models, techniques and, most importantly, a shared vocabulary to improve the understanding of game design as a craft. The ideal FADT, according to definition, is "well-defined," "abstract" (i.e., applicable across genres), has day-to-day utility, and works in a well-understood application context. During another GDC discussion, Warren Spector pointed out the unexpected resistance game designers exhibit when facing the idea that their work and thinking could possibly be described and discussed in more formal ways, and Marc LeBlanc was quick to pre-empt concerns by assuring that FADTs are not meant to be a "Swiss Army Knife" of game design.
His presentation started with the battle cry "down with fun," which he elaborated upon as he created a taxonomy of "fun" to illustrate both the fuzziness of the concept and its limited applicability. "Fun" as applied to games covers everything from simple sensual pleasure to make-believe, from drama to the satisfaction of solving intellectual challenges, from social interaction to submission, from exploration of another person's invention to self-discovery. It is hard to find precision in such versatility.
Personally, I suspect that a good share of the FADT efforts are trying to reinvent the wheel. The problems and challenges in finding a common vocabulary for a craft have been encountered in other professions, and the recent decade has brought a variety of solutions. In my limited experience, Christopher Alexander's work on "design patterns" seems the most promising avenue (more about this in a upcoming Gamasutra article), which uses an approach that has been successfully applied to fields as diverse as architecture, workflow management, and software engineering. (You can also see Zach Simpson's collection of game programming patterns as an example.)
Marc LeBlanc also looks to other disciplines for inspiration. In his GDC talk last year, he introduced feedback loops as a design concept. This year he ventured into "complex systems," which quite naturally contain feedback loops: the rules governing the state change take the current state into account. "Complex systems" (much like "fractals," "nonlinear dynamics," or "emergence") has become a popular science buzzword over the past decade, and the concept has suffered from this. Marc LeBlanc focused on the observation that the behavior of a dynamic system often cannot be easily predicted from its set of rules, even if those rules are deceptively simple. In reverse, mere observation of a sample system will not allow you to deduce of those rules.
The example commonly used to illustrate complex systems and their dynamics is that of cellular automata (for a reference I suggest Steve Wolfram's book). The most popular of the cellular automata is John H. Conway's Game of Life, which LeBlanc used to demonstrate properties "emerging" from a set of rules. He then compared this 2D cellular automaton with 2D board games like Chess or Go, where attack and defense tactics and strategy are only implied by the game's rules. He also pointed to card combos in Magic: The Gathering and emergent properties like Trains, Kiting, and Killstealing in EverQuest (the latter makes emergent properties look more like a problem than an asset).
LeBlanc tried to make a case for emergent complexity as a possible source for "fun." In his view, such emergence creates larger spaces to explore, offering the player more features to discover and more challenges to meet. He tried to distinguish systems with simple elements from those in which the constituents are quite complex . The one word he did not use was "combinatorial" – a good deal of the complexity in cellular automata is due to the fact that, while the number of possible states per cell is finite, the number of cells is quite large, and the resulting state space of the entire system is blown up beyond human comprehension by combinatorial explosion.
There is little to be learned from cellular automata that could be applied to game design, thus his presentation quickly moved on to examples such as sports simulations, which are well advised to replicate emergent properties of the real world by replicating the underlying dynamics. (Ted Zuvich's GDC lecture on physically accurate vehicle dynamics for Need For Speed was a case in point). For most games, differential equations will rule the player's world, not the discrete counting rules of transition that govern cellular automata. On another level, The Game of Life is indeed a telling metaphor with respect to game design: one detail LeBlanc did not elaborate on was that they are fully deterministic systems, ticking away like clockwork, driven solely by the laws of behavior and the initial conditions as defined by their design. If game designers ever get serious about abdicating authorship, these two devices -- initial conditions and laws of dynamics -- will be the essence of the tools left at their disposal. Random initial conditions, as contemplated by LeBlanc, will deprive the designer of even the ability to set the stage, before leaving it for good.
The lecture made the point that complexity is not accomplished by creating lots of rules. In fact, the common way to implement a game simulation is to write an expert system -- a database of rules of thumb and special cases, patched and hacked to accomodate short-term needs during game development. The properties created by rule-based simulations don't always make sense, and might even be contradictory. On the other hand, properties emerging from perfectly consistent simulations usually come as an unpleasant surprise. For instance, once players discovered rocket jumping in Quake 2 (whereby players fire the the rocket launcher at the floor and use the reactive force to propel themselves skyward), they began to use that trick to cut corners in ways the map designers had not anticipated. Warren Spector gave another example in a recent interview about his game, Deus Ex. He explained that the interplay between the game's AI and the sound caused by a single bullet casing hitting the ground exposed sniping players to such a degree that, in the interest of preserving the plot, the game simulation was hacked to accommodate this emergent property. (Caseless ammunition was seemingly not an option).
Marc LeBlanc pointed out that degenerate player strategies indicate a flaw in the simulation of resource exchange. He described game dynamics as a process of transportation or conversion (i.e., a flow defined by sources, sinks, and transducers), and he described player exploits as "energy spikes." In my physicist's eyes, this is just a way of saying that game simulations usually fail to model the conservation of energy correctly (as an example, multi-body simulations used in astrophysics perform error correction based on energy checksums). Degenerate player strategies as LeBlanc describes them are simply perpetual devices discovered by the player -- or at least a vast heat reservoir to tap into.
In other words, the solution to many such problems might be found in modeling the economy of transactions accurately. This is a hot topic for massively multiplayer games; this year's GDC offered Zach Simpson's lecture on In-Game Economics in Ultima Online. LeBlanc himself settled for a different solution, namely dampening the system dynamics to prevent spikes. From a physicist's experience, adding friction can only slow down every process, without addressing the real problem. His advice that designers should understand and tune exchange rates is certainly to the point, as is his recommendation to prototype early and test often games that exhibit emergent behavior. He also pointed out that introducing feedback loops into the system can have undesirable consequences: positive feedback (combined with an infinite energy supply, I would add) creates extremely unstable situations, while strong negative feedback can make the system too stable. Friction that overwhelms every other force will bring any system to a grinding halt.
All in all, the discussion of game dynamics in terms of nonlinear dynamics certainly opens an intriguing area of discussion, but I would advise game designers to proceed cautiously. The majority of popular science publications on nonlinear dynamics use the jargon without comprehending the underlying concept, and even physicists have applied them in sloppy and reckless ways. Beyond metaphor, there might be little practical use for this.
LeBlanc concluded his presentation by reviewing ways that various flavors of "fun" might emerge from toying around with a complex system, and the dreaded issue of "narrative" came up again. His claim that complexity gives you the proverbial "infinite number of monkeys" does not convince me, mathematically nor as a professional storyteller. Marc LeBlanc proposes "embedded (authored) narrative" as a complement, and wants game designers to restrict themselves to the major story arcs. While I consider "emergent narrative" a meaningless concept, abdication of authorship alone will not suffice.
This uneasy marriage of simulation-driven and scripted events fell apart when Marc LeBlanc made a case for "limited non-interactive moments." Certainly letter-boxing in-game cinematics and cut scenes is a visual cue to the player that he has entered a "hands off" sequence, but this technique does not remove the tension between the designer's desire for authorship and the player's desire for control. Game play is never "largely" emergent -- good narrative inevitably has long-range correlations and convergence points that impose constraints on the results of interactivity. I visibly annoyed one game designer by pointing out that interaction that is irrelevant to the chain of events at large is shallow and meaningless. Many game designers are painfully aware of this, and the workaround commonly used is a "Paris and the Golden Apple" hack -- at the end of the game, the player gets to choose one of two or three possible outcomes. (Incidentally, this kind of choice does not have in-game consequences.) A good example is System Shock 2, and some might be generous enough to include the cheesy ending of Half-Life as well.
An example mentioned during the discussion put the conflicting forces between authorship, emergence and exploitable features in a "mini narrative" nutshell: in System Shock, the last bullet in a clip did double damage. It is easy to see how such a simulation patch could be turned on its head once the player became aware of it.
LeBlanc offered valuable practical advice based on his experiences at Looking Glass: simplify game elements but use them in conjunction with each other, focus on interaction instead of element complexity, and settle for a two-tiered system architecture with a few solid foundation systems that are expected to survive the development process unchanged. It is as difficult for the game designer to predict the outcome of a set of rules as it would be for the player, since complex systems easily outgrow the human mind's ability to control and adjust. The slides of Marc LeBlanc's GDC 1999 and 2000 presentations are available as Powerpoint files.