I have been playing the Diplomacy board game online for roughly a year now over at Playdiplomacy.com. Roughly half-way through the year, I started using a Bayesian framework to help predict what my competitors were attempting to do. I have had some success using Bayes theorem during plain vanilla Diplomacy games, but I have had even more success with another version of Diplomacy.
To illustrate I will offer up my Diplomacy ratings for a few different versions of the game:
- Under the General rating category (which I believe combines all versions of the game), I am ranked number 131 out of 16,125 rated players. This puts me in the top 0.8% of players. Not too shabby.
- Under the Classic rating category (which counts just the standard plain vanilla version of the game), I rank number 337 out of 15,981 rated players. This puts me in the top 2.1% of players. Still good, but I am obviously better at other versions of the game.
- Under the Chaos rating category, I currently rank number 8 out of 1,186 rated players. This puts me in the top 0.68% of players. This is pretty good and is also the reason why my General rating is so much higher than it would be otherwise.
So, what is it about the Chaos rating category that allows me to excel?
First, it is probably helpful to describe what a Chaos is. From the Play Diplomacy website, “Chaos is a house variant that scatters each players’ starting three supply centers at random across the 34 centers on the map. Winter 1900 and Build Anywhere rules are included.” However, I don’t believe it is this feature that necessarily gives me an advantage.
My advantage is gained because Chaos games tend to also have “fog of war” and “gunboat” features. Fog of war means that you can only see what is in territories you control or territories next to your units. Gunboat means that the players do not speak to each other. In other words, information is extremely limited in most of Chaos games. This is where the Bayesian framework provides tremendous insights. In other words, I most likely have an information advantage over my opponents in these games because of how I analyze what I can see. Using a Bayesian framework, I can better estimate what is really going on and I can capitalize on the information. It is counter intuitive but, I find the behavior of my opponents in Chaos games more predictable than in the Classic games.
Another development is that I am no longer doing the actual mathematical calculations of Bayes Theorem in my games. I picked up this habit after reading Superforecasters by Peter Tetlock. I am just using the intuitive nature of the framework to figure out what is going on. It takes less time and it is still effective.