The Prisoner's Dilemma Competition

 

If you have arrived at this page by mistake you might like to go to the home page or read more about the prisoners dilemma

Frequently Asked Questions

This page lists some questions that we have been asked, which we hope answers any questions/queries that you might have

 

Q1: What will the tournament look exactly like? If it is a round-robin like tournament and the winner is the one player with the greatest sum of payoffs?

Yes - this is exactly it.


Q2: How are final scores calculated? Is it a total/average score of all rounds played against every single competitors? Or is it round robbins format playoffs in which we need to win to advance further in the playoffs?


The final score will be the summation of the scores after all your games (not an average). There is no concept of having to win to progress.


Q3: How many games are to be played by each player?

You will play against every other player.


Q4: Will a player play against himself as well, or just against his opponents?

Each player will play against all the opponents as well as themselves.


Q5: It says on web page that there will be default strategies in the population for the competition. Are they proportionally equally represented?

They are likely to be represented only once (and maybe not at all) and we reserve the right to change this!
This is not meant to appear as if we are skewing the competition. This is done on the basis that you can make no inferences about the other strategies in the population.
Also, if we put in (say) TFT, many other people might do as well, so they would be proportionally higher anyway.
So, you cannot make any assumptions about what strategies will or not be represented and in what proportions.


Q6: Does a strategy have to be purely deterministic, or can the decision-making-process be based upon propabilities?

You can do whatever you like. The only thing we need is a discrete cooperate or defect (for the "traditional" game) and a value between zero and one (which we will round if necessary) for the other game.
The only other thing we will enforce is a time limit.


Q7: What about "noise" in the data: are there equal noise propabilities for mis-interpreting Cooperate and Defect, or different ones? In what range will the propabilities lie?

The probabilities will be equal and will be around 0.1, though not fully decided yet.


Q8: What will be the noise probability in competition 2? How will it be used, that is, will a probability of noise of 0.1 mean that in one out of ten times, the strategy's bid will be *reversed*, or that it will be set one time out of ten to defect only?

If it's 0.1 (which is prooably will be), it will mean there is a 0.1 chance of seeing the wrong signal (i.e reversed) NOT a 0.1 chance that you will see defect.


Q9: Is the number of rounds per game known in advance ?

No but it'll be a mean of 200 (which mimics Axelrod) with a large standard deviation!


Q10: After the pay-off matrix for competition 3, the webpage states that "each player is allowed to vary their level of cooperation between 0 (no cooperation) to 1 (full cooperation)". Does this mean that other values than the ones in the matrix are allowed? If yes, what kind of numbers are allowed?

You can specify what values you like betwen 0 and 1, but we will round to the nearest legal value. If you want to be sure - then just use the allowed values.


Q11: The description of competition 3 states that "varying numbers of players" are allowed. What does this actually mean? Is it meant in the sense of http://plato.stanford.edu/entries/prisoner-dilemma/#Players related to the tragedy of the commons? And will the population of participating strategies be constant, or will it change, and if the latter, based on what? If the population does not change, what is the difference to the first two competitions with regards to the number of players? On the poster, this is written differently, namely "strategies that have more than one player", but then how is their overall performance evaluated, and what constraints are imposed on these strategies?

Okay . Let me tell you how it will work and we'll see if that answers your question.
The population will be constant throughout one run and you will know how many players you are playing against. What you won't know, at the very start, is how many players there is. You will only be given this information at the start of the game (i.e. so your strategy has to be able to cope with that).


Q12: Related to the previous question, but concerning all three competitions: is it possible to specify how many copies of one's strategy should take part? If yes, how is the performance of that strategy evaluated? That is, is the average of all copies of a strategy counted for the competition, or is it the individual copy that gets the most points that determines the competition's winning strategy? Also, what is the maximum number of copies of a strategy? Or is the number of copies decided randomly? If yes, in what range? Is the distribution uniform, if not what is it?

You will not be able to specify how many copies compete (as we are not designing the software in that way) but you can enter more than one strategy. We have a proviso on this (in case somebody enters hundreds or thousands) and we will say that we have the right to limit the number.
But we don't want to do this as it will be interesting to see if people can come up with strategies that cooperate with themselves within the whole population.
In any event, we will propobaly just run a competition where only one of each type of startegy is implemented just to see what happens.


Q13: Are there time-outs for strategies? That is, will a strategy be stopped if it takes too much time to compute? If yes, what is the allowed time, and what are the available computing resources one can expect?

For the web based entries there is no concept of a time limit, as the moves are straightforward (just a look up).

For the Java based entries we need to impose a time limit else a submitted strategy could "hang" and stop the tournament.

Of course, we could get into debates about what "what is time and how are we going to measure it?" However, this would be too complicated so we are simply going to say that we will impose a time limit of 2 seconds for a move. If you have not responded in that time then we will terminate that game. You will receive zero points and your opponent will receive the point sthey have accummuated thus far.

We are also aware that some people want to "link" to the outside world (to link to a faster machine, for example). We have said that this is allowable but you need to tell us if your strategy is going to do this (as we'll have to make sure that the competition machine is linked to a newtork).
In these cases we will allow 20 seconds per move (and that is open to negotiation) but we may have to run a separate competition when these oppoenents are involved as we may have to run on a desktop macine at one of our institutions as we may not have a link to a network at CEC. In nay case, we will run these strategies against ALL the others to give a fair competition.


Q14: Are the stategies allowed to use outside computer resources, e.g., can a strategy communicate with some, say, super computer via http?

You can do whatever you like - but bear in mind there will be a time limit for your move (see question 13).


Q15: Does the strategy know against which other strategy it plays at the moment, i.e., can it know for instance that it plays against itself by looking up the name of its opponent?

Probably not - this is related to above where, if you put in multiple copies you have to have some protocol to decide if you are playing against yourself - else you'll just do mutual cooperation (or one defect and one cooperate). We've become wise to this strategy :-)


Q16: I would like to participate at the Prisoner's Dilemma Competition at CEC'04. But my program is written in C/C++.
Is it only possible to submit Java programs or do you also allow C?

Sorry - but we can only accept Java entries.


Q17: Will there be some sort of global space to record the actions of every player against every other player? What I mean is that will it be possible that a player "C" can see what actions did player "A" did against player "B"?

No, this will not be possible. Each strategy should be self-contained.


Q18: What machine will the tournament be run on?

At CEC, it is likely to be a desktop PC with a spec of about 1.5GHZ.


Q19: How many tournaments will actually be run?

We will run the three competitions as advertised, but this has been increased to 4 for the second competition.. For the web based entries we will write a converter so that these strategies are converted into Jave based entries so that we can run them more easily (of course, this will be transparent to to the users).

However, we will also run other competitions in order to provide greater insights to the IPD.

For example, some strategies have asked us if they can link to an outside machine (super computer?) - see the question "What is the Time Limit for each Move?". We have said this is allowable and we will run these strateagies along side all the others. But it might be interesting (if we get enough) to run just these strategies against each other. And it would be interesting to exclude these strategies from the tournament.

Another example. (also see questions 11, 12 and 15) We are also aware that some people are planning to submit multiple (possibly duplicate) strategies. We know this as we have been told and have also been asked if the strategies will know which strategy they are competing against at any given time. So, given constraints outline above, we will run the tournament will ALL submittes strategies (even duplicates) but we may also remove duplicates and just run with one copy of each strategy.

If you have any ideas for additional competitions, please let us know.


Q20: What publications are you planning?

Please see the page devoted to this topic


Q21: How will the multi-player game work?

At the start of the game you will be told how many players are in the game. At each decision point you will have to decide on your level of cooperation. This will be compared against every other players choice and the payoffs calculated. You will then be told what levels of cooperation each other player used, before another round starts.


Q99:
dear Grayham.
my name is ????? ??? and me sponsor prisoner compettition and winner will get two fried ducks with rice sauce and lemon.i can also put bok choy in if wish you.
if you have a good idea for my sponsor then we can discus this on the email net.
p.s. if you like not this idea for me sponsor competition, then we can meet up when you free are at a daytime. Please wrtie back as i am keen very to sponsor compettition.

This was a real EMAIL we received. We, unfortunately, had to decline this very kind offer, but we thank the sender anyway.