Tit for tat

Seymore4Head Seymore4Head at Hotmail.invalid
Tue Apr 28 05:18:10 CEST 2015


On Sun, 26 Apr 2015 22:50:03 -0700 (PDT), John Ladasky
<john_ladasky at sbcglobal.net> wrote:

>On Sunday, April 26, 2015 at 6:41:08 PM UTC-7, Seymore4Head wrote:
> 
>> Richard Dawkins explains with passion the idea of game theory and tit
>> for tat, or why cooperation with strangers is often a strong strategy.
>> 
>> He talks of a computer program tournament.  I don't know what I could
>> say that would be more interesting than just watching the video.
>
>Well, I'm not sure sure what any of this has to do with Python -- but since I know something about the subject, I'll reply.
>
>That Richard Dawkins video is quite old -- it would appear to be from the middle 1980's.  Douglas Hofstadter's 1985 book, _Metamagical_Themas_, covered this exact same material.  A game called the "Iterated Prisoner's Dilemma" is played (I'll abbreviate it as IPD).  Humans can play, of course, but in this case it is played by algorithms.  An algorithm called "Tit for Tat" is surprisingly simple and robust.  When meeting a new contestant, Tit for Tat plays nice in round 1; and on every subsequent round, it plays however that opponent played the last time.  
>
>Evolutionary biologists like Dawkins point to the success of Tit for Tat in IPD as a model of how cooperation could emerge in a population of selfish organisms.  Now, in a round-robin IPD game, Tit for Tat wins pretty handily.  But in some other scenarios, as I recall, Tit for Tat is not a runaway winner.
>
>Suppose that instead of each strategy playing EVERY other, each strategy inhabits a "territory" in a space, and each strategy only plays its neighbors.  In "rough neighborhoods", Tit for Tat can lose out to more punitive strategies.  If Tit for Tat is around more cooperative strategies, it thrives.  The boundaries between good neighborhoods and bad are chaotic.  Tit for Tat more or less holds the borders, but usually can't clean out a bad neighborhood.
>
>This finding came out many years after the Hofstadter and Dawkins reports, so it's not covered in the video.  My reference to the idea is a 1997 paper entitled "The Undecidability of the Spatialized Prisoner's Dilemma," by Patrick Grim (http://link.springer.com/article/10.1023%2FA%3A1004959623042).

In the past, I have had some measure of success with the Toot for Tail
strategy.



More information about the Python-list mailing list