Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/5/84; site rtech.UUCP
Path: utzoo!linus!decvax!decwrl!sun!amdahl!rtech!jeff
From: jeff@rtech.UUCP (Jeff Lichtman)
Newsgroups: net.singles,net.women
Subject: Re: Win/Win Players vs. Win/Lose Players
Message-ID: <166@rtech.UUCP>
Date: Fri, 21-Mar-86 22:52:18 EST
Article-I.D.: rtech.166
Posted: Fri Mar 21 22:52:18 1986
Date-Received: Mon, 31-Mar-86 04:59:34 EST
References: <10844@amdcad.UUCP>
Organization: Relational Technology Inc, Alameda CA
Lines: 48
Xref: linus net.singles:10049 net.women:9425

> 
> In article <698@hounx.UUCP> kort@hounx.UUCP (B.KORT) writes:
>>Has anyone worked out the dynamics where one player is using Win/Win
>>and the other is using Win/Lose?  Seems to me the outcome is generally
>>Lose/Lose?  Does anyone have a strategy  for dealing with a Win/Lose
>>partner?
> 
> There was an interesting article in Scientific American about a
> computer simulation of various strategies. The most generally
> successful strategy was one called "tit for tat". You assume
> cooperation until proved otherwise, but you remember the bad guys and
> treat them according. A population of such players would take over a
> population of bad guys if introduced and would dominate if
> contaminated with bad guys.
> 
> It's a satisfying result, I think.
> -- 
>  Phil Ngai +1 408 749 5720

The tit-for-tat strategy works as follows: start out cooperating.  If, on
a particular turn, the opponent competes (i.e. does the opposite of
cooperation), then compete on the next turn and go back to cooperating.
The only thing one has to remember is what the opponent did the last time
around.

The Scientific American article described an elimination tournament in which
the worst player would be dropped every n rounds.  The article said that
"tit-for-tat" was the best strategy; that is, it would end up surviving the
longest, even when competing with much more complicated strategies.  One of
the conclusions of the article was that keys to longterm survival were
non-aggression (never hitting first) and forgiveness (not carrying a grudge;
hitting back if hit but then forgetting that it ever happened).

I wrote a program to run a similar tournament.  My brother and I played around
with strategies, and found one that worked better than "tit-for-tat" in the
mix of strategies we were using.  We called it "progressive punishment".  It
would never hit first, just like "tit-for-tat".  It differed from "tit-for-tat"
in that it would hit back once the first time a particular opponent hit it,
twice the second time, thrice the third time, and so on.  It should be noted
that the results of such a simulation depend entirely on the mix of strategies;
it's possible that "progressive punishment" would do very badly under some
circumstances.
-- 
Jeff Lichtman at rtech (Relational Technology, Inc.)
"Saints should always be judged guilty until they are proved innocent..."

{amdahl, sun}!rtech!jeff
{ucbvax, decvax}!mtxinu!rtech!jeff