Civilization veteran Soren Johnson, one of the foremost designers of strategy games and AI today and certainly one of the best writers on the subject, often remarks that the theme of a game is not to be confused with its meaning (slides here). Diplomacy may cast its players as the great powers of pre-1914 Europe, but it’s about simultaneity. StarCraft may put you in charge of Heinlein-esque space marines and alien civilizations, but it’s about asymmetry. If the theme and mechanics harmoniously cohere, then the mechanics can shed light on the theme in the way that art sheds light on the world. Pre-war Europe is an intriguing setting for Diplomacy because in all their backroom double-dealing, the empires didn’t take turns. Aliens are a good fit for StarCraft because you can map anything onto aliens, be it the collectivist swarm-by-numbers ethos of the Zerg or the judicious high-tech investment of the Protoss.
I am partial to this view, predominantly for reasons of aesthetics. If we are to conceive of game design as an art form, it does not suffice to decompose games into the artistry of constituent parts—the music, the models and sprites, the cinematic sequences, on rare occasion the writing. The aesthetics have to come from the specific properties that make something a game, whether it is played with a board and dice, a deck of cards, or a mouse and keyboard—and those properties come from the mechanics.
But that’s neither here nor there; I won’t elaborate today. Instead I want to turn to my favourite of Johnson’s examples: the evolution game. For your fill of Darwinian game mechanics, look not to Spore (which Johnson worked on), a game that is nominally about evolution from microbe to intergalactic juggernaut, but is actually about special creation. Back when I first played it, I wrote, perhaps a tad generously:
Let’s not bury our heads in the sand: by placing creature design into the player’s hands instead of leaving it up to random mutation, Spore inherently owes a lot to intelligent design. There’s still room for a real game about evolution in the Darwinian sense, where you set certain environmental constraints and preconditions, let a species run loose, and see if it survives in an ecosystem full of other models—kind of like how some engineers pit robots in mortal battle, but with adaptation.
Spore is a lot more creationistic than I gave it credit for; consider that the functional components of your custom-made species—the mouths, the horns, the flagella—are interchangeable parts from a specified, modular set, which is precisely what we would expect from a designing agent but not at all what we would expect from natural selection. But never mind all that. The evolution game exists, says Johnson, and it’s called World of Warcraft.
I would contend, however, that the Darwinian features Johnson ascribes to WoW are equally prevalent in most games with competitive and highly interactive player populations, provided there is sufficient strategic depth worth talking about. WoW is an evolution game because its core mechanic is community. Where there is a community of players and a developed metagame of optimal practices, strategic decisions are memes that compete for survival. Let’s call them strategemes.
Strategemes include everything from chess openings to Scrabble vocabulary: they are transmissible units of knowledge that players learn, study, and adopt—and crucially, copy. Copying them is not seen as unfair, but as an advantageous and often essential behaviour. They leave room for mutation, and we can perceive a frequency distribution of variations over a population of players and games.
But where does natural selection come into play? Let’s look at the exemplar we get from Johnson: the WoW talent tree.
World of Sporecraft
Every player in World of Warcraft belongs to one of ten character classes—Mages, Warriors, Priests, and so on—and every class has a range of “talents” (bonuses and special abilities) that fall into three specializations. As players rise in level, they gain points that they can allocate to configure their characters however they like.
Here’s the rub: all but the most basic talents are only available once you have already assigned points to the same specialization tree. To make your character as powerful as you can, the strategy is to allocate most of your talent points in one tree and distribute a few residual points in the other two; otherwise you will be a jack of all trades and a master of none.
The ramifications of specializing are most pronounced in classes like the Paladin, which can perform all three of the major roles in the game’s group-oriented content: healing, “tanking” (damage absorption), and dealing damage to enemies. For Paladins, these correspond to the Restoration, Protection, and Retribution trees. Within these trees, you can further allocate points in a manner optimized for the content you are playing through, be it fighting other players or raiding (killing bosses in coordinated groups of up to 25 people).
Who determines what is optimal? Not the game’s designers at Blizzard Entertainment, but the players themselves. Every few months the designers shift a few percentage points around to address balance issues, and the players involved in the game’s most challenging content scramble to adjust.
So far, this doesn’t sound evolutionary in the least. There is no heredity at work, and players can change their specializations at any point for a fee in the game’s virtual currency. Consider the following comment-box response to this article about Johnson’s recent theme-versus-mechanics presentation at the Game Developers Conference:
Soren Johnson knows nothing about evolution and [should] probably just never talk about it until he’s taken a biology course. His “Paladin Natural Selection” is completely off the mark. What he is describing is much closer to variable phenotypic expression where certain traits are expressed in an individual based on [its] immediate environment.
The commenter’s objection would be absolutely right were we to speak of the player characters themselves as individually evolving organisms. From this point of view, WoW characters do not evolve any more than an Englishman living in Boston evolves by speaking in a Bostonian accent. But it still makes sense to think of the English language as evolving, and look at traits of regional accents and dialects as they statistically appear in a population. Similarly, in WoW we can look at the player population as a whole and, with the aid of tools like Armory Data Mining, observe measurable changes in the frequency of talent allocations.
The underlying reason for this is in the way that talent choices spread. If you think WoW’s talent system gives players considerable freedom to customize their personal experience, you would be wrong. In reality, whenever the game is rebalanced, a few expert players proficient in mathematics run simulations and tests to crunch the numbers and figure out the best theoretical allocations under ideal circumstances. (Very few boss encounters present ideal circumstances, and almost no players have the ideal weapons and armour; but for the range of players serious enough to keep abreast of the theory, these findings usually suffice.) The experts debate about the data on discussion forums, update their FAQs, and release new versions of their simulation spreadsheets. Over time, more and more players adjust their characters to account for the theoretical recommendations. Players trying out a new character class will often model their choices after high-performance players whom they respect without comprehending the underlying rationale.
Talent trees can be replicated with perfect fidelity: a player can copy them point for point. They are unambiguously quantifiable, and it makes sense to think of them as genotypes. But they are only optimal provided the player behaves in certain patterns and within particular constraints, using their abilities in specific sequences. There we see phenotypic variation, which sometimes feeds back into talent-tree mutations within a tolerance of a few points; some character classes are inflexible while others are less rigidly specified.
Critically, a “cookie-cutter” talent setup is not necessarily what is best for every encounter, or even every player of that class and specialization. The fitness of talent choices—their capacity to replicate throughout the population—depends on social factors as well. Players copy talent choices that correlate strongly with success. This is observably the case for player-versus-player combat in arenas and battlegrounds, which involves considerable improvisation and is less susceptible to accurate mathematical modelling. As for raids, to adopt a theoretically endorsed talent allocation is to demonstrate that you stay informed, and minimize the risk introduced by your inclusion on the team. The raiding guilds that tackle even mildly challenging content will not invite players who deviate from the prescribed talents unless they provide an expert justification for why their modifications make sense.
In short, certain sets of talent choices have survival advantages over others. The fitter variations spread throughout the population while the frequencies of unfit ones decline. Players that do not adopt the most frequent choices are excluded from much of the high-level group content, and they (along with their decisions) are less likely to remain active in the game.
In this respect, World of Warcraft differs significantly from most video games. What does it matter what other people think, so long as you are having fun? When I play Super Mario Bros. I don’t care if the speediest route through the game is via the Warp Zones in Worlds 1-2 and 4-2; I can take the long road and clear every stage if I want. When I play Civilization I don’t care if early global conquest yields the highest scores; I like to win by constructing a spaceship to Alpha Centauri without ever building a city beyond my first. In adventure games like The Legend of Zelda the pleasure lies in figuring out the puzzles yourself; to look up the solutions in a guide is to admit defeat and cheat, and is held in lower regard.
Or look at Spore. In Spore there is abundant community interaction through an online portal where players can share their creations and randomly “seed” them on the worlds of others. Players can choose to preserve the creations they like, and the game tracks the most popular ones—which, for a time, was a roster headed by a model of Charles Darwin. In one sense this would appear to set up a competition to make the most creative and amusing models: the better your creature, the likelier it is to attain a high rate of incidence on other worlds. Why not consider the Darwin model a strategeme as well? Because the fittest creatures in the online community are not the ones that lead to the greatest mechanical success in the game itself. There is no selective pressure directed towards accomplishing objectives in the game, so players continue to make their in-game decisions independent of everyone else.
Granted, some WoW players ignore selective pressure as well: they impose various challenges on themselves and configure their characters specifically for the task. This can involve everything from controlling multiple characters at once to defeating high-level challenges in entry-level equipment. (Lisa Poisso’s 15 Minutes of Fame column routinely puts the spotlight on some of the most fascinating specimens.) But these activities are usually individual—rarely do they take place in large groups—and the special strategic decisions they involve do not tend to spread. From the population’s point of view, these innovations die out.
The dominant culture in WoW is one where copying strategic templates is not regarded as cheating, but is an expected precondition for accessing most of the game’s group content at all. This is not limited to individual characters and talent points, but extends to the behaviour of a raiding guild—who does what and who stands where. For all but the very best players, the raiding game is an exercise in fine-tuning locally appropriate variations on top of a broadly established strategy copied from somewhere else. Here, too, we see strategemes competing for dominance: successful variations spread to other guilds, and the most widespread ones become part of the orthodox strategic package that much of the player base comes to adopt.
World of Warcraft is an evolution game as a consequence of being a community game. But not all games with communities lead to evolutionary trends. What makes my concept of a strategeme unique among memes is that it applies to games where social learning, often in the form of direct imitation, is advantageous to the pursuit of absolute objectives embedded in the game’s design—so overwhelmingly advantageous, in fact, that it affects one’s inclusiveness in a community and the accessibility of the game’s content. There is feedback from the game mechanics that tells you if one set of choices is performing better than another. In WoW it may be your binary pass/fail success in defeating an encounter, or a metric of your damage or healing output. In chess it is your ability to win games (perhaps further broken down into your success as white or black, against certain opening lines, or facing various strata of opponents); this also generates social feedback as expressed in your Elo rating, which estimates your chances against other rated players and unlocks your access to upper echelons of competition.
Putting it generally, we see evolutionary change occur in games when the benefits of copying others far outweigh the qualitative value of making your own choices from scratch. This is a jarring philosophy of play when you consider that figuring things out for yourself is the central pleasure of games and puzzles that are free of social selective pressure—which includes most of the games people play in their ordinary experience. A game with no predesigned capacity for strategic evolution quickly develops one with the emergence of a community, so long as the mechanics are deep enough to reward the exploration of an immense variety of choices. An evolutionary account gives us the tools to explain how a player population prunes those choices, favouring some for widespread imitation while others are left behind.
The rest of the story
In future instalments, I will be looking at some other games that cultivate different kinds of strategemes, anticipate some reasonable objections to using analogies from biological evolution, and look at what strategemes may tell us about life outside of games (if such a thing exists). Depending on interest, I may also provide a technical supplement to this first part for readers who are comfortable with World of Warcraft‘s esoteric lingo and want to see some specific examples and hard data; I do have those at the ready.