Hacker News new | past | comments | ask | show | jobs | submit login
Game Theory Explained: Why The Joker and Not Batman is Our Savior (thisorthat.com)
155 points by sthatipamala on March 15, 2012 | hide | past | favorite | 50 comments



This is an interesting article, but The Tragedy of the Commons is only 1 of 7 Nash Equilibrium possible. It was cherry picked for a funny thought experiment where Joker is the good guy. Now let's try the Volunteer's Dilemma.

In the Volunteer's Dilemma a group is faced with an inevitable negative, and the only way prevent it is for one member to assume the negative unto himself. The classic example is jumping on the grenade. Nobody wants to do it, but somebody has to or everyone dies.

1) Assume Cooperators and Defectors only. Cooperators will eventually volunteer, Defectors never will. There can be a balance where Cooperators significantly outnumber Defectors, but if this balance is disturbed the population will eventually end.

2) Add Jokers. Jokers in this case would be like Defectors and never volunteer. Worse, however, is that Jokers would probably be so interested in the destruction that they would actively prevent volunteering. There can be no balance in a group with even one Joker still alive, and the population will end.

3) Add Batmen. Batmen neutralize the Jokers' ability to prevent volunteers, and are willing to volunteer themselves. The balance is restored as long as there are enough of these individuals to restore moral and put the group first.

So now in my thought experiment, Jokers are harbingers of the end while Batmen are necessary for survival.


Sounds like the Volunteer's Dilemma is the scenario that played out in the movie The Dark Knight, actually.


I believe the scene with the ferries was an example of Prisoner's dilemma not Volunteer's Dilemma.


True. That scene was almost a textbook Prisoner's Dilemma case.

But the overall arc of the movie -- concluding with Batman's needing to sacrifice himself for the greater good -- seems pretty Volunteery.


Are you the one who posted this exact comment on Reddit? http://www.reddit.com/r/science/comments/qw5y4/why_joker_and...


There is also the possibility of a Joker pushing somebody (possibly multiple to be sure) else onto the grenade, thus saving their life and causing destruction. Which has 2 possible out comes in itself:

1) Joker pushed another Joker or Cooperator. Which might be uncomfortable for a time, but would pass.

2) Joker pushed a Defectors, in which case some Defectors would want some form justice.

I have no numbers to prove this, just my feeling for the archetypes.


Shouldn't Batmen also have the ability to neutralize defectors?


How? By throwing anyone who refuses to "volunteer" on a live grenade?


it's funny, because, I think Batman would be considered a "joker" as well, he regulates defectors in his own trying to rationalize it while the joker creates chaos. Obviously, the batman prefers no loss of life.


How is scenario 3 different from scenario 1?


Just re-read your post.

Scenario 3 includes a population of Jokers. That is the difference.

So your assertion... is that Batmen only provide a benefit where there exist Jokers in the population. At least in the Volunteer's Dilemma.


Each of the scenarios 1 through 3 builds upon each other. The outcomes of 1 and 3 are similar, but in scenario 2 jokers are introduced into the mix, and in scenario 3 batmans are introduced to counteract the jokers.

...I believe the author is saying that if you have jokers, you need batmans to neutralize their effect.


"Game theory explained" with too many assumptions. Such as that punishment is so costly that it creates another problem of the commons, as the punishers (who are a subset of the cooperators) get less our of it than the non-punishing cooperators. I'd say this is wrong: in repeated games punishment is not a cost but an investment. A threat must be carried out in order to be credible in the long run, and thus a cooperator who punishes a defector (or a 'joker' for that matter) is better positioned in subsequent games to be treated fairly.


The analysis doesn't seem to hold up very well in terms of US post-9/11. Jokers (terrorists) caused a ruckus, which the defectors (military-industrial complex) overplayed as a call to arms for cooperators (US populace). The cooperators then produce excess benefit for the defectors (control of oil-wealthy regions, superfluous security contracts).

I can't pin my finger on what makes post-9/11 US different. Were the jokers not destructive enough? Does the model not account for power structure, or psychological needs (i.e. security) and exploitation thereof? Is the effect of a single joker short-lived (which would correspond to the outpouring of national unity post-9/11)? Did the 9/11 attacks actually increase the cooperator/defector ratio in the long term in a way that is not apparent to me?


The terrorists were not the jokers. Jokers are supposed to destroy everything without restraint, these guys were trying to defend (the best defense is a good offense) themselves from an empire. The irony is that the defectors (terrorists) are an offshoot of the original defectors (military-industrial complex).

So who are the jokers in this picture? The socially excluded psychopaths and sociopaths who sit in prison. I think that the proportion of real jokers (completely destructive agents) is extremely small and their effect is vastly overstated. I think that in real life, significant jokers tend to be forces of nature like hurricanes and tsunamis, not people.


Could it perhaps be that a very simple thought experiment is not a strong model for the chaotic actions of hundreds of millions of people over several years?


Oh indeed! I'm just trying to figure out the most likely reason why :)


There is probably something to be said for network resiliency, as well, but I haven't the faintest idea what.


This may not be helpful, but does an issue arise when a large number of defectors believe they're cooperators?

And actually, as katovatzschyn kind of points out, this setup (citizens / military-industrial complex / terrorists) may not actually map very well at all to the cooperator / defector / joker model.


Or, what happens if cooperators find out about this theory and so decide to act as jokers when they think that defectors are winning, but turn back into cooperators at other times.

Also, defectors want the maximum gain for least effort, so can reasonably decide to act as cooperators for the minimum time required to keep their preferred level of sloth going.

Jokers could also decide, if they trusted this model, that to increase overall destruction they act always as defectors, as otherwise they are just ushering in a new round of cooperation.


Minor quibble on your final point...

Jokers achieve greatest absolute destruction when the lifetime of the system is indefinite--you have to have sand castles before you can kick them over. So, it would seem to me that they'd be more likely to become cooperators for the sole purpose of building things back up to burn them down again.


Good point, though this suggests to me that there might be more than one type of joker;

Loki: likes eternal chaos, so fits your model.

Ice Giants: likes Ragnarök, so only sleeps or destroys.

However, both Loki and the Ice Giants could conceivably decide to act as cooperators if they suspect that there are a large number of defectors who are currently acting as cooperators, with the aim of overwhelming the cooperators by switching back to being jokers when the other fake cooperators switch back to being defectors.

[edit] Also, 'Loki and the Ice Giants' would make a really good band name, or possibly a tumblr site (nod to the xkcd enthusiasts).


I like your choice of terms! :)

Hm....

Ice Giants would seem to suggest a fourth state, one of pure inactivity.

I would almost say that you could model these modes (as the paper does) by changing the damage value to be 0 (dormant, effectively, Ice Giants) for some chosen system conditions. Were you to assign a negative damage, you might be able to model jokers-as-cooperator.

The interesting (perhaps most interesting) thing about the Jokers is that they have no stake in the system. So, I wonder how to account/model that in a Jokers-can-become-Coopearators system. Hm.


I think your problem here is that the terrorists weren't out as Jokers--they weren't attempting indiscriminate destruction of resources and benefits (regardless of what the government/news might lead you to believe).

Somewhat thankfully, true destroy-all-the-things thinking writ large and competently executed seems to be a very rare thing in the species.


For a contrary view on game theory, I highly recommend watching the excellent BBC documentary series "The Trap" by film maker Alan Curtis.

http://en.wikipedia.org/wiki/The_Trap_(television_documentar...

In the first episode, Curtis examines the rise of game theory during the Cold War. It discusses John Nash's paranoid schizophrenia and shows footage Nash in his later years acknowledging that his paranoid views of other people at the time were false.


There are two sides of Nash's work. One is the actual math, which can be evaluated without considering the credibility of the author. The other is Nash's attempt to use it to describe the real world, which necessarily involves assumptions and simplifications. This second side is where the obvious problems are.

Game theory remains popular because people can easily take the underlying math and apply it with their own assumptions. Of course, if you make too many simplifications you risk achieving a near-useless comic-book view of reality (literally, as in the blog post here).


> Of course, if you make too many simplifications you risk achieving a near-useless comic-book view of reality (literally, as in the blog post here).

For a moment there, I thought you were going to invoke Rorschach as an example.


Game theory doesn't require that people be selfish bastards. For example, in the Prisoner's Dilemma, it's perfectly fine to assume that each prisoner places a high value on the other person remaining free. That changes the payoff matrix, and the resulting matrix can still be analyzed by game theory (which will then predict that the prisoners will cooperate, since in game theory terms it won't be the "prisoner's dilemma" anymore).


The common mistake people make on game theory is the assumption that people will always use a hill climbing algorithm to find equibrilium, instead of considering the global optima. Hofstadter discusses this with regard to the "hyperrational" solution of cooperation-cooperation in the Prisoners dilemma.


I did a little googling, looks interesting but I couldn't find Hofstadter's actual work on this...where do I need to look?


To answer the question, if there has to be a Joker to stop Defectors, who is the Joker?

Joker does not need to be a person; it could be anything that destroys benefits with some good success. You get where I am going with this; Joker is any disaster that causes loss. For living things, death does the job pretty well. However, in real life, people are not the only entities which gather benefits, there are organizations that live beyond the lifetime of an individual. In that case, the Joker could be paradigm shifts, weariness, anti-trust cases, scientific theory, or the occasional market crash (you get 1 point for finding each entity these destroy).


The actual paper is here:

http://arxiv.org/abs/1103.3257


Thank you, $39.99 for the paper on the link seemed steep. If it went to the authors I would have no problem paying for it but as we all know - it's not.


Just the day I _really_ read "Non-cooperative Games (1951) John Nash, Annals of Mathematics". If you're mathematically inclined, I can only advise you to look it up: a pure piece of powerful and concise theory.


The Prestige and Inception have been two of the most literary films of the last ten years. Christopher Nolan keeps getting better and better.

So I'd be interested to read a serious analysis of the new Batman films since it wouldn't surprise me if Nolan has something up his sleeve here at this point. That said, this post is not it, so if anyone has any recommendations please share.


One big problem I see here is a tendency for black & white thinking. People in the real world don't fit into a neat little "cooperator" or "defector" package. A person's tendency to cooperate can be dependent upon a huge number of factors, including what the subject of cooperation is, how they're feeling that day, current events, political, moral, and religious opinions in vogue, illness and injury, the size of the potential benefit, relationships to other people who may be defecting or cooperating, whether they are a leader or a follower, etc.

Running a simulation of simple, unchanging people doesn't produce much useful information for the real world. Real life cooperation relationships are a lot more intricate and complex.


Your complaints are not really a huge problems with this statistical population model. In the end, you took goods out of a system and/or you put goods back into it - everyone falls somewhere along this scale in their lifetimes. All the reasons behind their choices are largely irrelevant, and you can simulate the model based on it.


You can sometimes emulate the model, but not always. See the other post in this thread about post 9/11 America.

The problem is that a few well positioned people who behave in a way that is not predictable, or larger forces that the model cannot anticipate (socio-economic, the weather, etc.) can distrupt the model.

Not that the model doesn't 1. Do a good job of explaining a concept and 2. Possibly fits into a larger, more complex model.


I'm having trouble understanding who is a defector, who is a cooperator, and who is the joker. The author has copious asides where he asserts "the financial industry; most of the people you work with, and probably you, too" are the defectors. However, I'm having trouble understanding the leap. Clearly he wants to relate this game to an insight about society at large as judged by all of his asides. So I'm trying to understand his position.

In our society most people must perform work to earn money, that money is taxed, to provide public services to all of us regardless of if we work or not. Therefore, the people paying taxes have a cost and get a benefit of the public services which fits his definition: "Cooperators generate a benefit, shared by all, but they pay a cost to do so." If someone pays taxes that must make them a cooperator. Of course we can have shades of cooperation based on how much is paid because no one give all of their money to the IRS. We do have the other extreme where no money is paid.

Now, there are people who pay less taxes or none at all as much or at all can enjoy those same services less cost or no cost all. That's his definition: "[Defectors] sit back and enjoy the group benefit without paying the costs."

In the sliding scale we are all both cooperators and defectors, but to what degree our role alters the equilibrium of the system (ie causing more cooperators or defectors).

That seems pretty straightforward, but he seems have flipped even the definitions: not just the joker/batman role.


In real life, the financiers Lloyd Blankfein represents are not simply "defectors." I would argue that they contribute (or at least could). However, a less contraversial point, especially in light of the recent crash, is that they cause a whole lot of destruction, e.g. they function as "jokers" as well.


Well this is where I think the term of joker should be clarified. Jokes must not consume the public good, but they must destroy it. Lloyd certainly benefits and certainly destroys, but his benefits exclude him from being a joker. Defectors cause the tragedy of the commons by not paying a cost. Or maybe as steep of a cost as others because in the real world Lloyd definitely paid a cost, but it was a mere fraction of his wealth where others were wiped out.

It was hard to understand the role of the joker and how it creates an equilibrium, but if the joker is random destroyer then he destroys the cooperator AND the defector equally. If defectors out pace cooperators then the likelihood a defector will be wiped out is greater. By wiping out a defector that shifts the balance (ie the defector dies, or the defector turns to a cooperator from the joker's actions). That's the only way I can understand the joker as a force of "good". And jokers turn out to be a better solution to the problem of equilibrium.


Fun to read! Thanks for sharing! Although, i think the author talked to much about the ins and outs without explicitly stating for everyone that the joker allows everyone else to cooperate (to eliminate the joker) :P

who are the jokers of our time?


I'm not a game theorist (and I'd welcome an opinion/correction from one), but I don't think the original paper's conclusion is very interesting, or has much bearing on real life.

Their result (that jokers, defectors, and cooperators will cycle) depends on some bizarre features of the joker:

- Jokers don't damage other jokers. This is why jokers drive out defectors.

- Jokers don't benefit from public goods. This is why jokers don't arise when there are lots of cooperators.

- The public good benefits that jokers forgo are redistributed back onto the cooperators. This is why cooperators flourish in joker populations: they produce a benefit for a large population, which then gets focused back on their small population.

This last feature is the really weird one. No public good I can think of can be redistributed this way [1]; indeed, goods that _can_ be efficiently reallocated like this tend to be naturally modeled as private goods.

Consider a public good: say, clean air. In this model we would have cooperators, who go out of their way to keep the air clean; defectors, who prefer air to be cleaner, but save effort by polluting; and jokers, who are indifferent to air pollution, produce a large amount of pollution, and somehow transfer the health benefits of clean air back onto the cooperators. Thus, if we added a single cooperator to a population of jokers, the cooperator would get a massive health benefit from not polluting, because of all the jokers "not consuming" the benefit. This is clearly nonsensical.

In short, the paper's conclusions follow from its premises, but its premises have nothing to do with any real situation that I can think of [2].

[1] There are a lot of public goods that degrade as more people use them, but not in a way that matches the math in this paper. Can anyone think of an example where this paper's conclusion would hold?

[2] Of course one can say that all game theory is an abstraction, which is true, but it still proceeds from a simplified model of reality, rather than totally arbitrary assumptions. This paper in particular would be getting no attention if the "joker" strategy didn't have a compelling real-world analogue. Furthermore, the problem here is not that their model is too simple, but that it adds weird, artificial features without explanation or justification.

(EDIT: Changed "small amount of pollution" to "large amount of pollution", as jokers do more damage than defectors.)


It's like the cholesterol saga - based on real research - but 'storyfied' beyond recognition of it's humbly narrow thesis.

What is really the relation between good writing and exploiting automatisms of human minds?


I stopped reading when I got to "tragedy of the commons." I'm having trouble deciding if I think the definition is condescending or just plain ignorant. Is the rest of the article as ignorant/condescending?


I get down voted for not agreeing with the article? Or did you think the definition of tragedy of the commons was accurate?

From the discussion it seems like the article may be interesting but based off of the definition of tragedy of the commons it seems like madness lies ahead. If the tragedy of the commons definition is indicative of what is to come reading the rest does not seem like a good choice for leisure reading.

I genuinely am interested in knowing if the definition of tragedy of the commons is indicative of the rest of the work? (Sorry the edit button is gone from my original post)


You were downvoted for relatively shallow and reflexive criticism. You're not disagreeing with the article; you didn't read the article. If you genuinely want to know if it's a good article, read it. You're not saving anyone any time by instead initiating a conversation about how you didn't like a sentence in the second paragraph.

There was a great article recently on being a good reader. I can't seem to find it again, but the general point was that you have to assume some level of good faith and try to see through inevitably fuzzy wording to assess the fundamental points that constitute real value and insight. In reading, you win if you learn something, not if you defeat an argument.

(Or maybe there's a broader problem if an article making the front page of HN doesn't count as sufficient endorsement in itself...?)


You are right, the article's definition of tragedy of the commons was poorly and bizarrely written, but you were downvoted because your criticism wasn't clearly expressed. You complained about the article but didn't specifically cite or correct its error.


Is the rest of the article written in the same manner?


I would be interested in how it changes the model if you assume that everyone is a continually varying mix of Joker, Cooperator and Defector, with also a value for how impressionable/stubborn they are.

Fixed roles and actions with a narrowly defined intents seem to be very common in these kind of thought experiments, mainly, I suspect, because it makes them easier to model. However I think it robs them of any real utility, and worse can give an illusion of insight that is entirely based on flawed assumptions.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: