At Grand Text Auto, Andrew Stern has reopened a discussion that ties into my last podcast. In that episode, I asked whether an NPC that ignored the player’s input but gave the illusion of reacting could still be called “interactive.” In his post, Stern asks a question about the other extreme. How can we allow the player to understand how the NPC’s mind works without blatantly exposing the logic?

The discussion is based on the assumption that players enjoy figuring out characters, and that the author of an interactive work should encourage that. Stern argues that breaking down the illusion by exposing the numerical “stats and sliders” that govern an NPC on the algorithmic level is a bad thing. I agree; in fact, I think this takes away some of the feeling of agency; instead of it seeming like you’re affecting the character, you can tell that the author anticipated your actions, and that you’re actually just increasing the “pugilistic” variable until the guy punches you out.

It reminds me of the character interaction in Morrowind, which I like to call the “Nice Hat/I Hate Your Hat” technique. You have two options: Admire or Taunt. There’s Intimidate and Bribe, but they’re for boneheads and suckers. Anyway, Admire has a chance of either increasing or decreasing the NPC’s opinion of you, depending on skill. Taunt has a chance of making them attack you (without you having committed a crime), but only if their opinion of you is high enough for you to care about what they say. This totally broke the illusion for me once I figured out how it worked; you can just play the numbers and, through persistence, skill mining, and maybe a spell or two, get whatever reaction you want from folks.

As for Stern’s actual question, which is “how to design an NPC… that allows a player to construct an accurate mental model of character behavior without ‘showing the data’ of the characters,” I have very little idea. I’ll have to think about it to increase my Cogitation score before I’ll have an answer.