Thursday, November 10, 2011

Axelrod Tournament: KPC Style!

I run an Axelrod-type tournament among my students each year. (This software is not exactly the same as the tournament we will run, but it is useful for demonstration purposes). And some background on THE EVOLUTION OF COOPERATION.

This year resulted in a particularly strange outcome: one of the student entries thumped the bejeezus out of Tit-for-Tat, and also did well against other strategies. It wins consistently, in various trials and contexts.

Is this an important new discovery? Well, the new strategy is....

HEY, WAIT A MINUTE. A hallmark of science is replicability. So let's throw this open to the smartest folks in the world, the real 1%: THE READERS OF KPC! If the new strategy is really that good, then let's prove it against the first team!

My main man David Sparks and I will run a wide open tourney. You can submit your strategies here. But READ THIS FIRST... Okay, you went to the web site, and had no clue what to do, and now you are back. I TOLD you to read this first. Jeez.

Instructions: Here is a useful planning sheet for writing out your entries before you send them in. You have two options for determining your strategy: You may

(A) choose your first round strategies and responses for each of the 4 second-round and 16 later-round combinations of your and your opponent's two most recent histories, or

(B) choose a play for each of the 25 rounds against each opponent. Choose one (1) method, and plan accordingly. Each play must consist of a number between 0 and 1, inclusive. 0 indicates that you will always defect, 1 indicates that you will always cooperate, and some decimal p between 0 and 1 indicates that you will mix with probability p of selecting cooperate.

In other words, A-type strategies have a maximum memory of two periods. You load the lags in the first two entries, and then after that go forward by specifying your response to any possible combination of your play and opponent's play in the previous two rounds.

B-type strategies allow you to specify a non-contingent stratgey, simply saying here is your play (possibly mixing) for each of the 25 rounds.

Each entry will play a tournament (25 consecutive rounds) against the same opponent, for all opponents who enter, plus a replica of your own strategy, plus 50 doves (All Coop) and 50 hawks (All Def). In other words, the tournament is round-robin, where everyone plays against everyone, plus 100 automatons (50 Doves and 50 Hawks).

The top five winners will get a free "I fought Tit-for-Tat, and I WON!" t-shirts in your chosen size.

And, not least important by any means, you get to name your strategy. Here are the best five name entries from the recent Duke PPE student tourney:

1. "Megan Fox (Great tits, Bad tats)
2. "Remember, kids, ALWAYS use protection!"
3. "Suck on my Rawls"
4. "Munger's Favorite Little French Fry"
5. "Fielder's Tits, Pujols's Tat(ers)"

(And, yes, friends, two of those beauties were submitted by women, so don't get all snooty...)

My own entry: "Hey, Euvolunteer DEEZ!" It did not do very well.

Remember, the actual strategy has to be entered here. If you have questions, post them in comments here and we'll answer them as a kind of ad hoc FAQs section.

Deadline: All entries put into the web site in usable form by Monday November 28 will be eligible for the fabulous prize.

BTW: Check out some of D. Sparks' other good work, which has been featured in KPC before.

3 comments:

Anonymous said...

What?


The use of PWNage in society is a brilliant name. Top 5 easily.

Scott said...

Such a great article it was which A-type strategies have a maximum memory of two periods. You load the lags in the first two entries, and then after that go forward by specifying your response to any possible combination of your play and opponent's play in the previous two rounds. Thanks for sharing this article.

Alison said...

Nice post which resulted in a particularly strange outcome one of the student entries thumped the bejeezus out of Tit-for-Tat, and also did well against other strategies. It wins consistently, in various trials and contexts. Thanks a lot.