The Iterated Prisoner's Dilemma Competition:
Celebrating the 20th
You and a friend have committed a crime and have been caught. You are being
held in separate cells. You are both offered a deal but have to decide what to
do. But you are not allowed to communicate with your partner and you will not
be told what they have decided until you have made a decision.
Essentially the deal is this.
- If you confess and your partner denies taking part in
the crime, you go free and your partner goes to prison for five years.
- If your partner confesses and you deny participating in
the crime, you go to prison for five years and yor
partner goes free.
- If you both confess you will serve four years each.
- If you both deny taking part in the crime, you both go
to prison for two years.
What do you do?
The dilemma you are faced with is that it is better for you to confess,
UNLESS you both deny taking part in the crime. BUT, if you think that you
partner will deny taking part, then you CONFESS and go free. See the dilemma?
To read another (better!) description of the prisoners dilemma see the Stanford
Encyclopedia of Philosophy entry and, to play a game, visit http://www.princeton.edu/~mdaniels/PD/PD.html.
Another way of looking at the Prisoners Dilemma can be found here. This is the definition used in this competition.
We are planning a series of competitions (the first was held at the Congress
of Evolutionary Computation in July 2003, the second will be held in April
2005) to celebrate, pay tribute to and update the competitions run by Axelrod [AXE84]. Importantly we aim to provide the environment
and data to allow researchers to conduct investigations into the latest
developments surrounding the iterated prisoners
dilemma. The competitions we originally held were 1-3, below. For the second
set of competitions, we have introduced a new category (number 4, below).
- Re-run the original experiment of Axelrod. We aim to
see if a tit-for-tat strategy still dominates or whether somebody can
develop a better strategy taking into account that there has been 20 years
since the original competition (for example, tit-for-2-tats was claimed to
be better than tit-for-tat). In addition, we would expect many more
entries that the 62 received by Axelrod (largely due to the internet),
thus giving us access to much more data.
- Organise a competition that has
noise in the data. That is, a signal to cooperate or defect could be mis-executed.
- Allow competitors to submit a strategy to an IPD that
has more than one player and more than one payoff, that is, multi player
- This competition will mirror the original competition
of Axelrod. That is, the rules are
- We will only allow one entrant per competitor and we
will not allow collusion between different competitors.
- The only strategy that will be introduced by the
tournament organisers will be RANDOM.
- Each strategy will also play against its twin.
- Every tournament will comprise the same number of
moves, which will be unknown to the entrants. We will run each tournament
5 fives times using a different, fixed number of
moves. In the original (second round) tournament, Axelrod used a 0.00346
probability of ending the game at each move. In fact, he drew the random
samples before each game and then “fixed” the number of moves
at 63, 77, 151, 156 and 308. We will use a similar scheme, but will not
use the same game lengths.
- The payoff matrix will be the one used in the original
Axelrod competition. That is, both players will
be awarded 3 points for mutual cooperation, 1 point will be awarded for
mutual defection and if one player defects and the other cooperates then
the defector receives 5 points and the co-operator receives 0 points.
- We will impose a time limit on the amount of time that
a player can use to make a move. This will be set at 2 seconds. THIS IS
DIFFERENT TO THE ORIGINAL TOURNAMENT, but we believe it is necessary to
avoid infinite loops.
- We will insist on being given a description of the
strategy, and source code, for all the strategies submitted. The source
code will NOT be made publicly available. However, we have to insist on
the source code so that we can monitor for collusion
More details about the competition can be seen here.
Please note that we wish to make this competition open to as many people as
possible - not just the academic community.
Please note: If you submitted a strategy(s) to the first competition,
they will automatically be entered for the 2nd competition. In addition, if you
only entered a single strategy, this will automatically be entered into
competition 4, see above
Important dates can be seen here.
- This competition is being jointly organised
by Graham Kendall, Paul Darwen
and Xin Yao.
- It will culminate at The Congress on Evolutionary
Computation Conference (CEC'04) in June
- This work is being supported by the UK's The Engineering
and Physical Sciences Research Council (EPSRC)
hits (since 7th November 2003) :