Go Nutty!

RSS feed

To talk back:
Send email to my first name at my last name dot com
-Keith Pelczarski

Random Nuttiness
Why go nutty?
Recalling London
Drink me
RageBoy Speaks!
More equal
Wonder Woman
Raging King, Fear Driver
31 Flavors
Refreshing?
On the edge?
Good Morning, Sunshine

Where is that post?

Go Nutty WWW

Interesting Voices
Randy Cassingham
Jonathan Coulton
Kevin Fox
Vince Hanks
Mark Hurst
Jason Kottke
Jeffrey Martian
Kerah Pelczarski
Max Pelczarski
Dave Pell
RageBoy
ResearchBuzz
Doc Searls
Tracy Sigler
Oliver Thylmann

Sites of Note
All Music Guide
Boing Boing
Financial Engines
The Fray
Google
Google News
How Stuff Works
Internet Movie Database
Memepool
Metacritic
The Motley Fool
Netflix
The Onion
Slashdot (/.)

:: Wednesday, August 21, 2002 ::

The Prisoner's Dilemma

Consider this -- you and a friend are busted and interrogated. You can either cooperate by staying quiet or defect by ratting out your friend.
If you both cooperate, 3pts each

If you both defect, 1pt each

If one defects and one cooperates, it's 5pts for the rat and 0pts for the sucker

This game can be used to study the interactions between people in many contexts. In fact, there's an interesting book about this very thing -- The Evolution of Cooperation by Robert Axelrod. In it, he reports on a Prisoner's Dilemma tournament that he ran wherein he solicited entries for computer programs to play this game against each other in an iterated series of prisoner's dilemmas.

The findings are interesting. The program that performed the best was a routine called TIT FOR TAT. It was a very simple program which always started by cooperating. It would continue to cooperate until the other player defected, at which point it would defect in kind, but only for one round. After that, it would go back to cooperating. It never beat any individual opponent, but it netted the best overall score in multiple rounds, even after all participants were given detailed analysis on how it won the first round.

Alexrod drew some interesting conclusions about this:
It pays to be nice.
TIT FOR TAT (TFT) always starts by cooperating, and it never defected unless the othe player did so first.

It's important to retaliate when provoked.
Whenever anyone tried to take advantage by defecting, TFT would punish its opponent by defecting on the next turn.

Be willing to forgive.
After retaliating with a defection in one round, TFT always demonstrated its willingness to play nice by cooperating on the subsequent turn.

Be clear about your position.
By using a simple strategy, TFT made it easy for other programs to recognize its willingness to both retaliate and forgive.
This model, while abstract, does have application to real-life situations. In WWI, opposing units in the trenches reached cooperative states where they would only shoot to kill after the other side broached the tacit agreement to not kill each other. (Oftentimes, these retaliations were two or three times the magnitude of the original defection.)

Another example can be found in the U.S. Congress. Representatives can accomplish more for their constituents by going along with other Reps' legislation, with the promise of future reciprocation.

It's important to note that being nice tends to pay off better if there's a reasonably high probabilty of future interaction between the players. If you're only playing the game once, defection is the safe bet. Luckily, many social systems not only involve repeat interactions, but also have a stabilizing affect which ecourages this type of behavior.

If this discourse hasn't put you to sleep, I'd recommend reading Alexrod's book for a more in-depth analysis of the Prisoner's Dilemma and its implication for biological, business, political, and social situations.

If you're not inclined to read the book, just remember this -- be nice, retaliate in kind, forgive the occasional trespass, and be clear about your position.

:: Keith 20:44 [link] :: ::