Even then, the PD is wildly overused, and gives exactly the wrong intuitions about human behavior in most cases. Essentially the PD is like blood letting: useful in treating 1-2 rather rare conditions, but widely applied to all sorts of situations where it does a lot more harm than good.
The PD is a model of social behavior that predicts the opposite of the social behaviors we generally observe. That's a problem, especially when it is constantly used as an example of why you need mafia bosses and state coercion. (Not to mention the non-coercive solutions rarely get brought up alongside those others... apparently it is all stick and no carry. I wonder why...) It is far more useful to see how humans deal with these issues than to point to the issues and say "And therefore we need mafia bosses or we can never do any of the good things we want to do that we already do!"
And frankly, if you want to give better examples of irrational self interest being damaging to society there are many better options. One could point out how individuals voting benefits for themselves at the expense of the state all end up paying lots of money for everyone, with only the state coming out ahead due to their cut of the transfers.
I think the attraction of simplified logical extremes is that they remove the need for active decision making. Just do X and things work every time. That's powerful and appealing, in the same way that "All the problems of alcoholism can be solved with more alcohol" is appealing to the drunk.
The PD is even more appealing if you are the type who wants to design the rules, or hope to be one of those in power to enforce them. People always see themselves as the rule makers in their mental utopias as opposed to those whose face is under the boot. Odds are, it's gonna be their face on the ground, just by simple arithmetic.
I think the impact of reputation on reproductive fitness explains why PD doesn't generalize. Reading this sparked a thought, maybe obvious to everyone but me, but I feel compelled to share. The social credit system is an attempt by the managerial elite to rigidly define reputation. They recognize that reputation is key, and think they can simply define it to achieve their ends. I see distinct parallels for how they think they can set prices, really all central planning. Just another organic factor that arises from free humans interacting being co-opted by tyrannical asshole busybodies.
Yup, I think that is a big part of it. Make the only reputation that really matters your reputation in the eyes of the state and you remove many competing loci of power. It is the same move with corporate cancel culture: forcing everyone to use the same scale for reputation means that whomever defines the scale rules. Normally your power company or bank wouldn't care whether or not you vote for the right party so long as you pay your bills, but now suddenly it all uses the same number, and voting pattern becomes more important than paying your bills.
Elinore Ostrom wrote a lot about that sort of thing, how different overlapping centers of power and reputation balanced out things and kept certain power centers from becoming too overbearing because people could shift to another if it got to be too much, etc. By centralizing reputation one can do away with that; centralize the money such that people can't even do business properly and you have pretty much wrapped up society at that point. All that is left at that point is the underground economy/society, which, if the USSR is any indication, can be quite expansive, but then the USSR was pants at really controlling things. The modern world holds many more potential terrors in that direction.
Hi Doc Hammer. I am sorry you have been having a bad week. Perhaps this paper about Functional Decision Theory by Eliezer Yudkowsky and Nate Soares will cheer you up. It's a fun read, assuming that you find mathematics (and logic) fun. And it is another way to discover that PD isn't all it's cracked up to be.
Thanks! I will check that out! I wonder if Nate was the guy at the conference... I am pretty sure it wasn't Eliezer since the presented was about 6'1 or 6'2 and Eliezer is... not. Had the beard and the glasses, but frankly that describes about 90% of white male economics grad students at conferences at the time.
I had to look up "smoking lesion problem" and... I wish I hadn't glanced at the images that turned up. I had parsed "smoking lesion" to look something like the entry/exit wound of a laser or light saber. It does not.
I just remembered that when Nicky Case talked about Functional Decision Theory instead of the Smoking Lesion problem he used the PMS-headache-chocolate problem. It goes like this: A woman gets Premenstrual syndrome every month. It gives her headaches, and a craving for chocolate. Should she give up eating chocolate? If you are a Causal Decision Theorist, the answer is an obvious No. Evidential Decision theorists end up concluding that they would give it up, and unlike when they walk off with the extra money in the Newcomb's paradox, they do not feel good about the outcome. This says to me that the CDT and EDT theorists both really want 'give me the decision tree that causes me to win' rather than have any fundamental unserving loyalty for or against causality.
Hah, yea it's ok, and yeah that is a much better example... less icky when you search for it. I have been known to come up with some questionable examples in my time, but that one probably tops 90% of them :D
That's really interesting though with the different CDT, EDT and now FDT business... I hadn't come across those before, the literature I have seen always defaulting to CDT it seems. It makes me worry a little that I should start asking "how serious about causality are you?" every time I read about decision theory from here on out. It also makes me want to insist on seeing computer testing of any theory someone proposes, so I can see that things were set and done before the decision gets made. None of that "Well, it must be there because the oracle predicted I would decide it would be there and pick it." I don't like it when thinkers don't seem to be sure the world exists whether or not they are there to observe it :)
Enjoyed this and it led me back down the rabbit hole I first entered by reading The Evolution of Cooperation by Robert Axelrod. Computer-generated strategies for PD-style games often seem to advantage ‘be nice but make it clear that you aren’t going to be anyone’s schmuck’, but even that gets complicated.
Sorry life’s chucking shit on the windshield. Same here at the moment 🤷♂️
Funny anecdote about Robert Axelrod: there's another big computational social science guy at GMU named Robert Axtel. No one can keep them straight, constantly coming up at conferences saying "Oh, I really liked you paper on ____" when it was the other guy's paper. They just gave up correcting people and got in the habit of learning each other's recent papers so that when people bring them up they can just roll with it; no one has noticed them doing it yet. Rob Axtel told us that if one of them wins a Nobel Prize they are going to flip a coin to see who actually will show up to accept it, just to see if the committee catches on.
Sorry to hear life's taken a hard turn to hell for you too. Must be something about late September... the air is turning crisp, the days are getting shorter, and Fate says "You know what? Fuck that guy in particular."
Great discussion of this (as someone perhaps guilty of overusing this game myself)! A couple of things:
1. I'm not sure the criticism of this specific model in it's application to the real world is much different to that of any model from the social sciences. In all cases you have to be aware of the nuances of reality and which of your models best fits to describing that situation and the limitations of that application.
2. In terms of the practical/political implications of the equilibrium outcome, it can be interpreted in a beneficial way for society and against government control. If the players are firms trying to arrange collusive pricing, their failure to cooperate is good, and also provides less justification for strong regulation and state interference.
1: Yes, most models in the social sciences are crap, because they are not calibrated to the real world in a manner that lets them be applicable in the "Ok, here's the right model, input the relevant values for the variables, and... here's what will happen." Most models are play models, where you can show things like "This can happen, if the conditions are right" but don't say anything about whether the conditions you obtain in reality are actually right. That's ok! I've made one of those models myself, tried to get it published and everything. The models are useful for showing what CAN happen, so long as you always remember they do not imply that it WILL happen in any particular case. Typically the points of failure are those uncalibrated variables, not to mention the bajillion other variables the model excludes and you hope are not important.
So I agree with this point, but I fear you don't take it far enough :) The limitations of application are immense, much more so than most social scientists grasp, much less convey to their students.
2: I agree, although with the same caveats: humans are really good at figuring out ways around those problems of cooperation. As Bryan Caplan puts it "If you are looking for evidence of price collusion, look for evidence of successful industry enforcement of rules." It does make for a good case of why regulatory capture is such an issue, and so appealing to businesses, since the normal non-coercive solutions to PD type issues don't generally apply to businesses.
On the first point, let me put it a different way. If no such thing as the PD game existed, would your own thought about the issues you write about here be so precise?
You discuss in great detail scenarios which are more complex, nuanced, where assumptions fail etc. But I put it to you that you would not have been able to conceptualise these ideas nearly so well if you did not have the mental framework of the PD game to work with in the first place.
You might be vastly underestimating my ability to nitpick the ever loving shit out of things :D
I mean, yea, modeling is useful, but not nearly so much as social scientists make it out to be. A screwdriver is also useful, but not if you think it is going to let you build a functional car. Game theory takes this problem a little farther than even most economics, as all but the simplest models are unsolvable analytically and even then rely on assuming variable values to get outcomes, yet are then treated as if they can be ported right into the real world. Very much a case of spending too much time staring at models and not enough time staring at the real world, in my opinion, so much time that they forget to look at the real world, perhaps.
A fair point, but as a final rejoinder I would point out that economics as a discipline has moved in a much more empirical direction. Pure theory is much rarer these days, much harder to publish than applied stuff.
Sorry to be such a Negative Nancy, but empirical economics as done these days is possibly a worse mistake, at least in the sense of "spending your time doing something useful". Effectively most empirical work seems to be picking your end result, misapplying statistical theory to questionable data, futzing around till you get a p < .05, then calling it a day.
What should economists do? I think a much higher value activity would be more descriptive: going out and seeing how people, businesses and markets actually work, what they are actually doing. For example, Arnold Kling frequently points out how many economists never have heard or credit default swaps, yet those are critical in how the Federal Reserve does business. Personally, when I hear most economists talk about businesses, and especially supply chain, I want to yell at them "THAT ISN"T HOW IT WORKS!" I think economists like writing papers based on things they don't need to leave the office or talk to people to do, but that is limiting their understanding. They should be more like biologists, going out in the field and seeing what they subjects get up to and figuring out how that fits into their theoretical models.
You highlight what I think is a huge issue in many peoples' thinking (and here, economists specifically). That is, people tend to treat COMPLEX systems as if they were merely COMPLICATED systems. Like biological systems, even the smallest sectors of "the economy" are complex systems. Models can trend toward accounting for or approximating all of the possible variables in a complex system, but can never perfectly map them. This is why drastic interventions into complex systems like markets or human bodies often have huge, unintended consequences, whereas drastic interventions into complicated systems like machines tend to generate outcomes that are within a foreseeable range. Models are often helpful for increasing our knowledge of complex systems, but should rarely, if ever, be relied upon to fully explain or predict complex systems.
This reminds me of a conference I attended several years ago where the keynote discussion involved an economist who studied firm location decisions and a guy who worked for firms specifically to scout out locations for their factories/offices etc. Despite it being the major topic of his life's research, the economist admitted he had no idea there was someone with the specific job this guy had until that day.
Even then, the PD is wildly overused, and gives exactly the wrong intuitions about human behavior in most cases. Essentially the PD is like blood letting: useful in treating 1-2 rather rare conditions, but widely applied to all sorts of situations where it does a lot more harm than good.
The PD is a model of social behavior that predicts the opposite of the social behaviors we generally observe. That's a problem, especially when it is constantly used as an example of why you need mafia bosses and state coercion. (Not to mention the non-coercive solutions rarely get brought up alongside those others... apparently it is all stick and no carry. I wonder why...) It is far more useful to see how humans deal with these issues than to point to the issues and say "And therefore we need mafia bosses or we can never do any of the good things we want to do that we already do!"
And frankly, if you want to give better examples of irrational self interest being damaging to society there are many better options. One could point out how individuals voting benefits for themselves at the expense of the state all end up paying lots of money for everyone, with only the state coming out ahead due to their cut of the transfers.
I think the attraction of simplified logical extremes is that they remove the need for active decision making. Just do X and things work every time. That's powerful and appealing, in the same way that "All the problems of alcoholism can be solved with more alcohol" is appealing to the drunk.
The PD is even more appealing if you are the type who wants to design the rules, or hope to be one of those in power to enforce them. People always see themselves as the rule makers in their mental utopias as opposed to those whose face is under the boot. Odds are, it's gonna be their face on the ground, just by simple arithmetic.
"People always see themselves as the rule makers in their mental utopias as opposed to those whose face is under the boot."
Some people don't think this way. The problem is that the people who don't think this way rarely become the rule makers.
Well, fair enough. I was using modal logic there, but the mode for my readers is probably pretty far from the mode for the human species overall :D
I wasn't criticizing your writing or logic. I just wanted to find an entry point for getting on my soapbox.
I think the impact of reputation on reproductive fitness explains why PD doesn't generalize. Reading this sparked a thought, maybe obvious to everyone but me, but I feel compelled to share. The social credit system is an attempt by the managerial elite to rigidly define reputation. They recognize that reputation is key, and think they can simply define it to achieve their ends. I see distinct parallels for how they think they can set prices, really all central planning. Just another organic factor that arises from free humans interacting being co-opted by tyrannical asshole busybodies.
Yup, I think that is a big part of it. Make the only reputation that really matters your reputation in the eyes of the state and you remove many competing loci of power. It is the same move with corporate cancel culture: forcing everyone to use the same scale for reputation means that whomever defines the scale rules. Normally your power company or bank wouldn't care whether or not you vote for the right party so long as you pay your bills, but now suddenly it all uses the same number, and voting pattern becomes more important than paying your bills.
Elinore Ostrom wrote a lot about that sort of thing, how different overlapping centers of power and reputation balanced out things and kept certain power centers from becoming too overbearing because people could shift to another if it got to be too much, etc. By centralizing reputation one can do away with that; centralize the money such that people can't even do business properly and you have pretty much wrapped up society at that point. All that is left at that point is the underground economy/society, which, if the USSR is any indication, can be quite expansive, but then the USSR was pants at really controlling things. The modern world holds many more potential terrors in that direction.
Hi Doc Hammer. I am sorry you have been having a bad week. Perhaps this paper about Functional Decision Theory by Eliezer Yudkowsky and Nate Soares will cheer you up. It's a fun read, assuming that you find mathematics (and logic) fun. And it is another way to discover that PD isn't all it's cracked up to be.
https://arxiv.org/pdf/1710.05060.pdf
Thanks! I will check that out! I wonder if Nate was the guy at the conference... I am pretty sure it wasn't Eliezer since the presented was about 6'1 or 6'2 and Eliezer is... not. Had the beard and the glasses, but frankly that describes about 90% of white male economics grad students at conferences at the time.
I had to look up "smoking lesion problem" and... I wish I hadn't glanced at the images that turned up. I had parsed "smoking lesion" to look something like the entry/exit wound of a laser or light saber. It does not.
Terrible example. And here I was, trying to improve your mood! But that one they did not make up, so not _their_ fault.
I just remembered that when Nicky Case talked about Functional Decision Theory instead of the Smoking Lesion problem he used the PMS-headache-chocolate problem. It goes like this: A woman gets Premenstrual syndrome every month. It gives her headaches, and a craving for chocolate. Should she give up eating chocolate? If you are a Causal Decision Theorist, the answer is an obvious No. Evidential Decision theorists end up concluding that they would give it up, and unlike when they walk off with the extra money in the Newcomb's paradox, they do not feel good about the outcome. This says to me that the CDT and EDT theorists both really want 'give me the decision tree that causes me to win' rather than have any fundamental unserving loyalty for or against causality.
Hah, yea it's ok, and yeah that is a much better example... less icky when you search for it. I have been known to come up with some questionable examples in my time, but that one probably tops 90% of them :D
That's really interesting though with the different CDT, EDT and now FDT business... I hadn't come across those before, the literature I have seen always defaulting to CDT it seems. It makes me worry a little that I should start asking "how serious about causality are you?" every time I read about decision theory from here on out. It also makes me want to insist on seeing computer testing of any theory someone proposes, so I can see that things were set and done before the decision gets made. None of that "Well, it must be there because the oracle predicted I would decide it would be there and pick it." I don't like it when thinkers don't seem to be sure the world exists whether or not they are there to observe it :)
Enjoyed this and it led me back down the rabbit hole I first entered by reading The Evolution of Cooperation by Robert Axelrod. Computer-generated strategies for PD-style games often seem to advantage ‘be nice but make it clear that you aren’t going to be anyone’s schmuck’, but even that gets complicated.
Sorry life’s chucking shit on the windshield. Same here at the moment 🤷♂️
Funny anecdote about Robert Axelrod: there's another big computational social science guy at GMU named Robert Axtel. No one can keep them straight, constantly coming up at conferences saying "Oh, I really liked you paper on ____" when it was the other guy's paper. They just gave up correcting people and got in the habit of learning each other's recent papers so that when people bring them up they can just roll with it; no one has noticed them doing it yet. Rob Axtel told us that if one of them wins a Nobel Prize they are going to flip a coin to see who actually will show up to accept it, just to see if the committee catches on.
Sorry to hear life's taken a hard turn to hell for you too. Must be something about late September... the air is turning crisp, the days are getting shorter, and Fate says "You know what? Fuck that guy in particular."
This whole reply made me laugh - thanks 😊
I’m actually fine - just caught up in bad shit happening to people around me 👍
Great discussion of this (as someone perhaps guilty of overusing this game myself)! A couple of things:
1. I'm not sure the criticism of this specific model in it's application to the real world is much different to that of any model from the social sciences. In all cases you have to be aware of the nuances of reality and which of your models best fits to describing that situation and the limitations of that application.
2. In terms of the practical/political implications of the equilibrium outcome, it can be interpreted in a beneficial way for society and against government control. If the players are firms trying to arrange collusive pricing, their failure to cooperate is good, and also provides less justification for strong regulation and state interference.
Thanks, glad you liked it! To your points:
1: Yes, most models in the social sciences are crap, because they are not calibrated to the real world in a manner that lets them be applicable in the "Ok, here's the right model, input the relevant values for the variables, and... here's what will happen." Most models are play models, where you can show things like "This can happen, if the conditions are right" but don't say anything about whether the conditions you obtain in reality are actually right. That's ok! I've made one of those models myself, tried to get it published and everything. The models are useful for showing what CAN happen, so long as you always remember they do not imply that it WILL happen in any particular case. Typically the points of failure are those uncalibrated variables, not to mention the bajillion other variables the model excludes and you hope are not important.
So I agree with this point, but I fear you don't take it far enough :) The limitations of application are immense, much more so than most social scientists grasp, much less convey to their students.
2: I agree, although with the same caveats: humans are really good at figuring out ways around those problems of cooperation. As Bryan Caplan puts it "If you are looking for evidence of price collusion, look for evidence of successful industry enforcement of rules." It does make for a good case of why regulatory capture is such an issue, and so appealing to businesses, since the normal non-coercive solutions to PD type issues don't generally apply to businesses.
On the first point, let me put it a different way. If no such thing as the PD game existed, would your own thought about the issues you write about here be so precise?
You discuss in great detail scenarios which are more complex, nuanced, where assumptions fail etc. But I put it to you that you would not have been able to conceptualise these ideas nearly so well if you did not have the mental framework of the PD game to work with in the first place.
You might be vastly underestimating my ability to nitpick the ever loving shit out of things :D
I mean, yea, modeling is useful, but not nearly so much as social scientists make it out to be. A screwdriver is also useful, but not if you think it is going to let you build a functional car. Game theory takes this problem a little farther than even most economics, as all but the simplest models are unsolvable analytically and even then rely on assuming variable values to get outcomes, yet are then treated as if they can be ported right into the real world. Very much a case of spending too much time staring at models and not enough time staring at the real world, in my opinion, so much time that they forget to look at the real world, perhaps.
A fair point, but as a final rejoinder I would point out that economics as a discipline has moved in a much more empirical direction. Pure theory is much rarer these days, much harder to publish than applied stuff.
Sorry to be such a Negative Nancy, but empirical economics as done these days is possibly a worse mistake, at least in the sense of "spending your time doing something useful". Effectively most empirical work seems to be picking your end result, misapplying statistical theory to questionable data, futzing around till you get a p < .05, then calling it a day.
What should economists do? I think a much higher value activity would be more descriptive: going out and seeing how people, businesses and markets actually work, what they are actually doing. For example, Arnold Kling frequently points out how many economists never have heard or credit default swaps, yet those are critical in how the Federal Reserve does business. Personally, when I hear most economists talk about businesses, and especially supply chain, I want to yell at them "THAT ISN"T HOW IT WORKS!" I think economists like writing papers based on things they don't need to leave the office or talk to people to do, but that is limiting their understanding. They should be more like biologists, going out in the field and seeing what they subjects get up to and figuring out how that fits into their theoretical models.
You highlight what I think is a huge issue in many peoples' thinking (and here, economists specifically). That is, people tend to treat COMPLEX systems as if they were merely COMPLICATED systems. Like biological systems, even the smallest sectors of "the economy" are complex systems. Models can trend toward accounting for or approximating all of the possible variables in a complex system, but can never perfectly map them. This is why drastic interventions into complex systems like markets or human bodies often have huge, unintended consequences, whereas drastic interventions into complicated systems like machines tend to generate outcomes that are within a foreseeable range. Models are often helpful for increasing our knowledge of complex systems, but should rarely, if ever, be relied upon to fully explain or predict complex systems.
This reminds me of a conference I attended several years ago where the keynote discussion involved an economist who studied firm location decisions and a guy who worked for firms specifically to scout out locations for their factories/offices etc. Despite it being the major topic of his life's research, the economist admitted he had no idea there was someone with the specific job this guy had until that day.