This is not Prisoner’s Dilemma. In order for the PREFIX strategy to be workable, a trusted third party has to guarantee that the source code passed to each agent is actually what the other agent will execute. A real Prisoner’s Dilemma has no such trusted third party. If there is no trusted third party to guarantee the source code, then seeing the “source code” becomes useless.
Of course this is not PD. This is a different game model that happens to have an interesting and non-obvious solution.
Inventing such models is worthwhile because one of them might someday turn out to be a closer match for the behavior of real-life agents than the classical PD. Real-life agents do have some mutual information, so the natural first step is to construct a model with complete mutual information. The next steps… are up to you, dear readers. :-)
You did not phrase your article as if it was obvious to you that this is not PD. Throughout your article, you referred to this as a “PD tournament” and “playing the PD”.
So you’re trying to model real-life agents in PD-like situations. Real-life actors generally don’t face PD, but face a variant PD+I, where I is infrastructure that tries to induce agents to cooperate. Such infrastructure generally works through incentives, disincentives, probabilities of discovery, etc. The justice system, traffic police, even religion are examples of such PD+I infrastructures.
Each of these infrastructures works differently; in the case of religion, radically differently. I guess religion might be the closest real-life example to your source code inspection proposal—people do appear to cooperate more with others who share their religious beliefs. Other infrastructures, such as the justice system and traffic police, do not appear to be captured by your model.
Yes, that’s completely right. The justice system might be a less interesting example of I than religion—it just tweaks the payoffs in the cells. It would be really interesting to bring scenario 1 closer to real-life religion by somehow making information incomplete, but as of now I have no idea how to do it formally.
This is not Prisoner’s Dilemma. In order for the PREFIX strategy to be workable, a trusted third party has to guarantee that the source code passed to each agent is actually what the other agent will execute. A real Prisoner’s Dilemma has no such trusted third party. If there is no trusted third party to guarantee the source code, then seeing the “source code” becomes useless.
Of course this is not PD. This is a different game model that happens to have an interesting and non-obvious solution.
Inventing such models is worthwhile because one of them might someday turn out to be a closer match for the behavior of real-life agents than the classical PD. Real-life agents do have some mutual information, so the natural first step is to construct a model with complete mutual information. The next steps… are up to you, dear readers. :-)
You did not phrase your article as if it was obvious to you that this is not PD. Throughout your article, you referred to this as a “PD tournament” and “playing the PD”.
So you’re trying to model real-life agents in PD-like situations. Real-life actors generally don’t face PD, but face a variant PD+I, where I is infrastructure that tries to induce agents to cooperate. Such infrastructure generally works through incentives, disincentives, probabilities of discovery, etc. The justice system, traffic police, even religion are examples of such PD+I infrastructures.
Each of these infrastructures works differently; in the case of religion, radically differently. I guess religion might be the closest real-life example to your source code inspection proposal—people do appear to cooperate more with others who share their religious beliefs. Other infrastructures, such as the justice system and traffic police, do not appear to be captured by your model.
Yes, that’s completely right. The justice system might be a less interesting example of I than religion—it just tweaks the payoffs in the cells. It would be really interesting to bring scenario 1 closer to real-life religion by somehow making information incomplete, but as of now I have no idea how to do it formally.