A metaphor I think is useful here is the story Sylvie and Bruno by Lewis Carroll.  In it a king orders the court cartographers to make maps of increasing accuracy until they finally have a 1-1 scale map of everything in the kingdom.  The hero’s quest is to convince the king he’ll win a Darwin Award if he unrolls the map.

There is a constant debate among those of us who assay life as to just how closely our models of reality have to agree with empirical data in order to be useful.  Unfortunately, despite George Box’s hopeful assertion, models are only ever useful for deceiving oneself into believing that natural law has anything to do with human behavior.

The first wrong model I used was a taxonomy  for classifying player types developed by Richard Bartle in 1990.  In Bartle’s taxonomy there are four basic types of players in games.  Achievers, Explorers, Socializers and Killers.  I was definitely the explorer he describes as players for whom “The real fun comes only from discovery and making the most complete set of maps in existence.”  I should’ve known better than trust the ramblings of the self-proclaimed “Wizards of MUDS” but it turns out Bartles accurately predicted my final score as an “Explorer” in the game of Freemium game design.

You can imagine what happens to characters like me when the game industry is dominated by people whose ludo type is Killer.  According to Bartle, “Killers use words like: ‘Ha!’, ‘Coward!’, ‘Die!’ and ‘Die, Die! Die!’ (Killers are people of few words).  “

It’s amazing how fast you find this out how dominated a strategy being an explorer is but what you don’t realize is how quickly you also learn to change your game.

The learning process in game play is nicely described by theories of Q-learning or re-enforcement learning models.  The problem is that those models assume a well defined Markov decision chain that every rational player would follow…and of course All players are rational.


At about the same time Bartle was deciding how people played Dungeon’s and Dragons, Nigel Howard was doing the same thing for international arms talks.  Only Howard didn’t assume rationality.

Howard’s Drama theory holds that as one learns the game they are only holding their definition of the game as a provisional assumption.  If they get enough evidence that the game isn’t what they thought it was, they’ll re-define the game to better suit the data.  Very Bayes.  The result of this was that players would fall into predictable Markov decision processes only so long as there was no divergence from expected outcomes.  As soon as something unexpected happened, the game would change.  And nothing’s more predictable than the unexpected in the game industry.

In 2011 Seattle hosted Casual Connect, a conference of casual game developers, at which everyone was told that any company without several statisticians on staff would be at a competitive disadvantage against teams that could see into the hearts of its players.

Everyone got the message and I truly believe that, but for that message, I would not have been hired by a game company.
The problem was that they were expecting this guy, while I thought I was there to make the most complete maps of player behavior in existence.

What happens in a 2-player game when each player is given the rules to completely different games?

If they have the expectation that everyone is playing the same game, the person with the fewest words moans “Lame” and struts to the next machine in the arcade.

The thing is, that because they are learning and adapting, the way they play the next game is different than how they would have played in the absence of analytics.

They’re haunted by the possibility that within the crystal ball of analytics there lies the super-power to make people like you…even if only for cosplay at a conference.