NEWS: University of Southampton team wins Prisoner's Dilemma competition

From: Chris Healey (chealey@unicom-inc.com)
Date: Thu Oct 14 2004 - 10:48:06 MDT


I thought this would be of interest to a number of people on this
list.

It would appear that a new strategy has been submitted, that can
consistently outperformed tit-for-tat in the Iterated Prisoner's
Dilemma(IPD). The strategy involves a predetermined authentication
sequence of initial moves to identify colluding agents in each
iteration.

Here a link to the IPD competition's results:
http://www.prisoners-dilemma.com/results/cec04/cec04.html

And the original press release...

----------------------------
Source URL:
http://www.soton.ac.uk/Press/PressReleases/Name,4551,en.php
Posted Date: 7 October 2004
----------------------------

News Release
>From the University of Southampton

Ref: 04/151
7 October 2004

University of Southampton team wins Prisoner's Dilemma competition
A solution that allows computer agents to collude, rather than compete
with each other, has won this year's recreation of Axelrod's classic
Iterated Prisoner's Dilemma competition.

The winning solution was devised by a team from the University of
Southampton's School of Electronics and Computer Science (ECS) which
met the competition's 20th anniversary challenge to resolve the
original dilemma in a noisy environment where moves could be
misinterpreted.

Professor Nick Jennings, a member of the ECS winning team said: 'We
developed ways of looking at the Prisoner's Dilemma in a more
realistic environment and we devised a way for computer agents to
recognise and collude with one another despite the noise. Our solution
beats the standard tit-for-tat strategy.'

The Iterated Prisoner's Dilemma has been of interest to the world of
computer science since the publication of Robert Axelrod's seminal
book in the 1980s. It presents a simple game between two prisoners, in
which their combined choices to co-operate or defect, determines
whether they are imprisoned or walk free. Both players make a choice
and then their decisions are revealed and both receive scores.

Up to this year, the most common strategy applied to the dilemma was
tit-for-tat which consistently outperformed every strategy entered in
the original competition. It starts out by co-operating, and then
punishes any strategy that defects by defecting on the next move.

The solution devised by ECS uses coding theory to enable agents to
recognise one another and to transmit messages reliably over noisy
communication channels. No outside communication is allowed so the
agents have to recognise one another by playing a specific sequence of
moves at the start of each game.

Dr Alex Rogers, another ECS team member said: "A key question in
future computing systems is how the different agents within the system
should interact and co-operate. Working on simple models such as the
Prisoner's Dilemma gives a fundamental understanding of how to tackle
these problems."



This archive was generated by hypermail 2.1.5 : Tue Feb 21 2006 - 04:22:46 MST