Business Insider: “They Finally Tested The ‘Prisoner’s Dilemma’ On Actual Prisoners — And The Results Were Not What You Would Expect”
Article at http://www.businessinsider.com/prisoners-dilemma-in-real-life-2013-7#ixzz2ZxwzT6nj, seems revelant to a lot of the discussion here.
There’ve been studies about people who consider themselves to be relatively successful are less cooperative than people who consider themselves relatively unsuccessful. The study referenced in that article seems to bear this out.
So if you want the other party to cooperate, should you attempt to give that party the impression it has been relatively unsuccessful, at least if that party is human?
Since prison culture often emphasizes punishment and retribution for not cooperating with fellow prisoners, the fact that they often cooperate is not surprising. My guess is that if one did this with people who had just been convicted of crimes but had not yet been in prison for a long time, one would see less cooperation.
I already said this on Facebook, but I might as well paraphrase it here. Social feedback loops in American prisons are supposed to be pretty tight: unless they went to some trouble to blind this after the fact (and it doesn’t look like they did), it would be obvious whether participants had cooperated or defected. Especially in the non-iterated version (smaller result space), which is the opposite of how it’s supposed to work.
On top of that, the social dynamics are a little different from most PD-like problems we encounter in that there’s an adversarial relationship baked in: usually you’re pitted against someone with whom you have no special kinship, but here we have outsiders setting members of a ready-made insider group against each other. (The original formulation of the problem does have this feature, but it’s usually been ignored in analysis.)
What they should have done is partner them against people from other prisons.
Before reading this article, and based only on its title, I predicted (on PredictionBook, with confidence 80%) that the result would be that prisoners cooperated “surprisingly” often, based both on the phrase “not what you would expect” and based on vague general things I guessed about prison culture. Thanks for the calibration exercise!
The calibration exercise aside, I don’t think this is particularly relevant to PD discussions on LessWrong, which I thought were more about “true” PDs (e.g. us vs. aliens, us vs. a paperclip maximizer). The incentives in a PD between humans are not easily deducible from the setup of the experiment alone.
Link to the original article.
Maybe you should give the impression that you’re both relatively unsuccessful.
I don’t think so. It seems more likely to me that the common factor between increased defection rate and self-perceived success is more consequentialist thinking. This leads to perceived success via actual success, and to defection via thinking “defection is the dominant strategy, so I’ll do that”.
It’s my understanding that, in a repeated series of PD games, the best strategy in the long run is “tit-for-tat”: cooperate by default, but retaliate with defection whenever someone defects against you, and keep defecting until the original defector returns to cooperation mode. Perhaps the prisoners in this case were generalizing a cooperative default from multiple game-like encounters and treating this particular experiment as just one more of these more general interactions?
Well, to be precise, researchers found tit-for-tat was the best, given the particular set-up. There’s no strategy that is better than every other strategy in every set-up. If everyone has a set choice (either “always defect (AD)” or “always cooperate (AC)”), then the best strategy is AD. If there are enough TFT players, however, they will increase each others’ scores, and the TFT will be more successful than AD. The more iterations there are, the more advantage TFT will give. However, if all of the players are TFT or AC, then AC will be just as good as TFT. If you have an evolutionary situation between AD, AC, and TFT where complexity is punished, “all TFT” isn’t an equilibrium, because you’ll have mutations to AC, which will out-compete TFT due to lower complexity, until there are enough AC that AD becomes viable, at which point TFT will start to have an advantage again. All AD will be an equilibrium, because once you reach that point, AC will be inferior, and an incremental increase in TFT due to mutation will not be able to take hold. If you have all AC, then AD will start to proliferate. If you have AC and AD, but no TFT, then eventually AD will take over.
Thanks for that explanation. The complexity factor hadn’t occurred to me.
Tim Harford has some relevant comments: