There’s going to be a major change in my life at the start of the new year. I don’t know exactly what I’m going to do, but it seems probable that I’ll go back to school and finish my doctoral thesis. My graduate thesis was about a learning system that made recommendations on studies to students, so artificial intelligence is sort of my thing. I’m going to continue in that vain. However, its in the nature of advanced studies that I’ll have to go deeper this time. So, what do I know well enough to do a thesis on? Games.
Adversarial reasoning is decision making where there is another party or parties who are also capable of making decision, which affect (or should affect) your process. We all do it all the time (probably not enough, though). I’m always reminded of a story McNamara tells in The Fog of War about how he met some Vietnamese minister long after the Vietnam war. They were discussing the reasons for the war and McNamara tells they were forced to go to war because they didn’t want Vietnam to join China. The answer was something along the lines of “Didn’t you know anything about our history? We’ve been fighting the Chinese for a thousand years”. So, basically the whole was fought over nothing. Understanding the opposition is therefore very important.
There’s plenty of research on adversarial reasoning, but it always seems to be either military or security focused. I’m more interested in a more dynamic situation where you don’t necessarily make decisions beforehand on who your enemy is or your alliances can shift. For this reason, I’m going to go with games, because they are a fertile ground for such research with clear rules of engagement and a lot of free and hidden information.
Since I don’t necessarily want to spend all my time coding the games, I’ll go with fairly simple ones, but ones where allegiances matter. I think Bang and maybe Werewolf are the easiest to simulate, maybe Resistance. I might also try to simulate something like Diplomacy, but not whole game, but some subset of it. I hope I can figure a way to put MtG in there somehow, probably through commander, but its clear I can’t make a full simulation of that game alone, so if I do it, it will be very rough approximation of how people choose who they want to take down in the game first and who they want to work with to do it.
Right now I’m inspired, but I also know there’s going to be a lot of work. I haven’t even discussed this with anyone at my department, or applied for any grants to do this, but we’ll get there.