Synopsis of key lessons of cooperative economics
This article is a synopsis of a series of articles titled "Key Lessons of Cooperative Economics", based on the book "A Cooperative Species, Human Reciprocity and Its Evolution" By Samuel Bowles & Herbert Gintis, 2011. The objective is to provide a steelmanning of the book's arguments.
This book is about many things, but this series of articles focuses on (a) the lessons it has regarding cooperative games, and (b) lessons it has regarding how people behave in small teamwork settings.
How did I come to this?
This amitious book covers many areas of thought including policy, society, economics, game theory, psychology, just to name a few. I discovered this book after studying algorithmic game theory (AGT). AGT introduced me to the notion of correlated equilibrium -- a key concept in cooperative economics. I started out mostly interested in what the book had to say about algorithmic game theory (AGT), but ended up becoming equally fascinated by what it has to say about the human aspects of teamplay.
In other words, I came for the AGT but stayed for the sociology.
Cooperative game theory is a key idea of the book. Cooperative games allow the use of binding agreement. What makes it possible to bind the agreements is some mechanism outside the game that enforces the agreement and makes them have consequence, such as being enforced by a third party.
Alternatively the group may abide by a code of conduct that is respected by each participant. This could be achieved by each group member following their own "conscience." Where that conscience comes from is a key theme of the book, but is not a focus of my review. Another way to instill rules and best practices is by rigorous practice, such as the drills used by sports teams and armed forces. This can be viewed as a type of evolutionary game, which is another key idea of the book.
Enforcement may be an external centralized entity, such as a police force. Enforcement can also be implemented in a decentralized manner. This is a key theme of the book. Group members take it on themselves to enforce the group norm. How this is done in an economical way is another key theme of the book.
To be effective, players in a cooperative game must either acknowledge their “conscience” or respect the enforcer. The mechanism used to coordinate the players is known as the choreographer, also called a correlating device, or correlating signal.
Though "choreographer" sounds better and "correlating signal" is also more natural in many of the practical settings, I'll stick with "correlating device," because it is more common in the literature.
Equilibrium concepts are used in game theory for understanding group behavior. For example, we may want players to converge to a set of states that are deemed "good", and tend to remain in that set of good states.
- Walking down a path. In this case two people are walking towards each other down the middle of a narrow path. In the typical scenario, one walker will move to one side, and then the other walker will move to the other side.
In the walking path example, a good equilibrium would be that one walker moves to their left, and the other walker moves to their left, and thereafter they remain on those respective sides until they pass each other. A less good, albeit possibly comical progression of states, might be where both walkers simultaneously move to their left, then both walkers simultaneously move to their right, repeating this until by facial expression, hand gesture, body language, or verbal outcry (say) the impasse is resolved.
There are many types of economic equilibrium. The key equilibrium concept in cooperative games is the correlated equilibrium.
It is possible for players can learn from each other's historical plays and end up converging to a correlated equilibrium. In the walking path example, each walker can see what the other walker is doing, as well as possibly their facial expression, hand gestures or other body language, the direction of the gaze, and response to their surroundings. In a short number of plays most players in this game will converge to a mutually compatible equilibrium for the remainder of the encounter.
Mechanisms for coordinating behavior
The correlating device is a key component to cooperative games. It allows groups to achieve an equilibrium state that is coordinate their behavior in a way that is to the mutual benefit of everyone playing the game.
Examples of correlating device in cooperative games :
- Contract law. A contract is an enforceable agreement to cooperate.
- Public signal. A hiking group agrees to meet unless it rains. Subsequently members may not be able to communicate with any other member, but can decide for themselves whether the notion of "to rain" has been satisfied. The group is able to coordinate even though they are unable to communicate after the initial agreement was achieved.
- Traffic signal. This is also an external public signal that all players can see, and is also another example of a contractual agreement to behave in a preordained manner as stipulated by the registration process required to obtain a driver license.
- Metronome. A music teacher uses this to get musicians to synchronize their playing.
The signal need not be completely public, it can be partially public.
In cooperative games it is possible to make a binding agreement. In Noncooperative games it is not possible to cooperate through agreements or enforcement. Cooperation is possible but is achieved only by players playing rational strategies based on self-interest, only. Much of the book is devoted to comparing and contrasting these two models. In particular it shows how in real life, cooperative game theory does a better job at predicting how human players will behave in many of these games where there are problematic scenarios, such as supposed paradoxes or types of information unraveling where assuming that everyone is playing according to a self-regarding strategy leads to a worse (and sometimes even the worst) outcome for all players, known as game theory dilemma.
Before presenting these dilemma, let's return to the walking path scenario.
There are two actions: walk on the left, or walk on the right. So far, so good -- this is easy to capture with non-cooperative game theory with the standard payoff matrix.
If a walker A continued to walk down the middle even after walker B has moved to one side, in the real world this could be construed by B as A being not fully aware of the situation. It turns out that classical economics commonly assumes that every player is fully aware of what is going on in the game. We already have a simple real world scenario that breaks a key assumption of the standard classical economic model.
We could fix this by having B temporarily leave the game, say, by stepping off the path until A passes. This is equivalent to adding an additional action -- step aside. This changes the game from one having two actions to one having three actions. Setting that aside for now, let's suppose that B's strategy might include some type of signaling to get the attention of A. Of course, that is an additional action, not represented in our original model. Setting that aside for the moment, suppose that A does not respond. This could be excused as A just not playing the game well, i.e., not being a strategic player. For example, A might be a small child or a (nonhuman) animal, or a person who is not feeling well and is not behaving normally.
Many of the key results in classical economics rely upon every player being a rational and strategic player. This makes it unusable for this simple scenario.
The seemingly simple walking path example is becoming a tad subtle. How can cooperative game theory can come to the rescue? Is there a means of providing for a separate enforcer or hinting signal? This path is out in the middle of a secluded area where there is no police force. There are only two people on the path. There is no hinting signal to suggest to A to move to one side of the path. So, where is our correlating device?
As the book explains in many other scenarios, cooperative game theory accounts for this by allowing B to have a model of what comprises "normal" behavior. B's model is private information. However, B could convey some of this information to A by acting indignant towards A. Indignance is a social emotion, but the expression of indignance using facial expression, speach, or body language is a social signal that A expects is understood by other members of civic-minded norm-abiding members of society. In such a society, A might take the hint, and move to the other side of the path.
But what if A did the something else instead? Say, if A switched over to the same side of the path as B, and continued to match the side of B even after B switches to the other side, A's behavior in the previous paragraph could be construed by B as an aggressive move by A. In the real world this would trigger a whole host of other possible responses by B, such as to fight or to flee.
The walking path scenario is one of the simplest scenario involving coordination and cooperation between two players. Handling this seemingly simple scenario becomes increasingly difficult using a classical normal form game payoff matrix. This scenario is more readily handled using an extensive form game, using prior knowledge, knowledge learned from experience, or a combination of both.
Yes, the extensive form game is a simpler model than the classical normal form game. However the idea of Occam's razor in science is to use a model that is a simple as possible but no simpler.
Social norm and social preference as correlating device
Handling players that are either irrational or nonstrategic would seem to require some understanding of what comprises typical behavior. Returning to the walking path scenario, if from what B can gather in this situation, B believes that A is capable of doing the right thing and still is not doing so, it could be construed by B as "rude" behavior. Such "rude" behavior triggers a a social emotion called "indignance". This indignance felt by B is a social emotion triggered by A's failure to comply with a social norm held by B. B's expectation is a social norm. The indignance response by B would trigger one or more strategies. B could assess the situation and decide to convey indignance to A. Or, B could keep that information private.
A key theme of the book is that a social norm is readily modeled as a correlating device, a key component of cooperative game theory. One of the book's key messages that social norms can be modeled using cooperative game theory.
Much of the action in applying game theory in the real-world is in using non-cooperative game theory in scenarios involving large numbers of competing players. There are many examples of real-life scenarios involving a small number of players that map onto games used in game theory, however it is fairly easy to imagine scenarios akin to the ones we imagined for the walking path scenario where non-cooperative game theory becomes unwieldy. Not only that but many paradoxes arise where classical game theory is not only unwieldy but gives a worse outcome than what we would expect to be the case when played in real life by human players.
- The volunteer's dilemma is resolved by altruism,
- the coordination game is resolved by agreement,
- in the unscrupulous diner's dilemma everyone is worse off by playing the strategy that is dictated by non-cooperative game theory than they would have been if they collectively pursued another strategy,
- enforceable agreements are a typical solution to the prisoner's dilemma in real world scenarios
- the Nash equilibrium strategy solution is a bad predictor of people’s behavior in a traveler's dilemma, and most people do not use strategies based on pure self-regard, and moreover the strategies that they actually use are optimal
- in many repeated games Nash equilibrium strategies lead to outcomes known as "information unraveling", that lead to various dilemma.
Non-cooperative game theory is relatively less flexible for modeling smaller groups of human players, such as in social loafing scenarios, and in many other games. The book presents many other examples than the ones given above.
Why this matters
Correlated equilibria are computationally less expensive to find than the more well-known and much celebrated Nash equilibrium that dominate non-cooperative game theory.
Finding Nash equilibria in general involves solving systems of nonlinear inequalities, which is hard (both informally and from a computational complexity standpoint). Finding correlated equilibria, on the other hand, requires solving a linear programming problem, which is exceptionally easy for computers.
Calculating a correlated equilibrium only requires only a linear program (cf. Papadimitriou and Roughgardenm, 2008) which is easy for computers to solve. Calculating a Nash equilibrium involves solving systems of nonlinear inequalities, which is in general much more computationally demandng.
A typical example of a real-world problem that maps to calculating a Nash equilibrium is that of estimating the equilibrium prices in an exchange economy.
Another reason why the concept of correlated equilibrium is more useful is because real life scenarios are typically correlated equilibrium and not Nash equilibrium. There is typically some level of communication available among the players in most games encountered in real life, whether we are aware of it or not.
Decentralized consensus and coordination
The book demonstrates many ways that cooperative games in human groups are decentralized using social norms and social emotions.
This is not to say that people do not also use non-cooperative strategies or evolutionary strategies, the book admits that they do.
Along with cultural institutions such as codes of ethics and rule of law these social norms and social emotions play a primary role in guiding human groups to self-organize in ways that are mutually beneficial within the group, while also making them more effective at competing with other groups.
For the purpose of brevity and in order to provide an unbiased steelmanning of the author's arguments, most explanations will kept terse with a minimum of elaboration.
I focus on conclusions instead of methodology. I leave it to others to validate the science.
The authors do seem to set up a strawman -- sturdy as it is -- in the form of classical economics. Not all classical economists disagree with criticisms of homo economicus, though to be sure, many still do.
The book also often equates classical economics with non-cooperative game theory, whereas many economists commonly use evolutionary game theory, bounded rationality, behavioral economics, and econometrics, among numerous other tools, some of which are flexible enough to to model the players' subjective beliefs about what actions their opponents might take, possibly including correlation among the actions of subsets of opponents.
The book's authors are avowed marxists.
For the record I am not a marxist.
In this series of articles that follow, I am trying my best to steelman the arguments. However it would be naive for me to just say that and hope that that will inoculate me from some type of complicity in some kind of political agenda.
The authors' politics may put off many readers. However, I believe that key lessons of this book are useful without resorting to even marxism's milder cousin, socialism. I explain why below.
I feel it necessary to state these qualifications up front. I read this book seeking actionable ideas that can be converted to working code to solve real-world problems out in the wild. However, some of the book's themes are risky to write about because the book wends through a veritable minefield of value-laden terms, ideologies, and dogma.
There are some common memes that I believe to be mistaken:
- Cooperation is anti-competition, therefore cooperation is anti-competitive.
- This is a fallacy.
- Cooperation improves competitiveness.
- This is addressed in the book, and in my review.
- Cooperative economics implies socialism.
- This is addressed in the book, and in my review.
- Many counterexamples are provided in the book showing how cooperative economics was used by people to self-organize well before socialism emerged.
There is one meme that does hold some credence:
- The book is anti-capitalism.
The book criticizes capitalism openly and often. After all, it is proposing an alternative. Nonetheless, the core ideas are useful regardless of politics. It is my hope and intention that my take on the book is entirely apolitical.
But first, to remind just how charged these topics are, here is a sample of some entertaining quotes on the matter.
The book's authors do tend to criticize "classical economics", and from what is readily available online it is also apparently true that the book's authors sympathize with non-capitalist ideas. But they are not the only ones:
Capitalism is the extraordinary belief that the nastiest of men for the nastiest of motives will somehow work for the benefit of all. -- John Maynard Keynes
Although some claim that Keynes did not utter these words, apparently he did, in the book “Christianity and Human Relations in Industry”, 1951.
One of the most damning criticisms of all is hypocrisy:
“they used to say a rising tide lifted all boats. Now the rising tide just seems to lift the yachts." -- British Labour MP Ed Miliband
Many may consider "economic freedom" to be a necessary part of this discussion, for example as measured by the Index of Economic Freedom by The Heritage Foundation and The Wall Street Journal, in which as of 2018 USA ranked 18th.
"Economic freedom" is a term used in policy debates and in the philosophy of economics. I'd consider freedom to be an objective in its own right. The notion of economic freedom is another one of the trigger issues that either side of a polarized debate may use to demonstrate that their idealogy is the better one. It is also one of the themes tha I found to be absent in the book, and that I hope and intend to be completely absent from my take on it.
Clearly, marxism and communism are loathed by capitalists:
"Under capitalism, man exploits man. Under communism, it's just the opposite" -- John Kenneth Galbraith, quoting what he describes as an old Polish joke. "A life in our times"
"How do you tell a communist? Well, it's someone who reads Marx and Lenin. And how do you tell an anti-Communist? It's someone who understands Marx and Lenin" - Ronald Reagan Source: BBC , 1987
Communism doesn't work because people like to own stuff. -- Frank Zappa
Communism and marxism don't get any airtime in my take on this book.
However, many equate socialism to communism, or believe that socialism inevitably leads to communism:
Socialism is the same as Communism, only better English. George Bernard Shaw
This is where the slippery slope fallacy can be wielded. By condoning cooperation, one can be viewed as supporting socialism. Socialism is sometimes used interchangeably with communism but the two philosophies have some stark differences. Most notably, communism is a political system, whereas socialism is primarily an economic system.
There is very little connection between economic performance and welfare expenditure. On some scales of "most socialist countries" you may find six scandinavian countries in the top ten, which also rate highly in economic terms and general levels of happiness. However, this is beside the point because the themes I am exploring are applicable outside of a socialist economic system.
Socialism is an economic system where the ways of making a living are owned by a society as a whole. These ideas can be leveraged in a completely apolitical system, such as an ecosystem of algorithmic agents. The ideas from this book are already utilized in market-based economies where the signalling is done by inanimate agents that are owned privately.
The key tool of cooperative game theory and cooperative economics is the correlating device -- a mechanism used widely in free markets. Using it does not imply some measure of socialism.
A correlating device provides hints to people, which if followed will help people self-organize more effectively and avoid costly setbacks. Hints are suggestions that can be ignored to more or less degree depending on the situation. An example is a stop sign at a road intersection. Another is the markings on the roadway. That there is a correlating device does not mean that it will be enforced with violent means. In reality, these hints are often ignored without any ramifications to the "offender".
If you put a sign on your private property warning intruders to "beware of dog" then you are using a correlating device.
A contract is another correlating device. The contract is a key tool of capitalism.
On the other hand, a spike barrieris not just a suggestion, being a mechanism that can enforce its "recommendation" without due process of law.
Consider the scenario when you board an airplane. There are very strict rules about when and how you can enter and exit the airplane. These are a contract that can and will be enforced forcibly. If you do not comply you can be forcibly removed from the aircraft and detained. Even though this may be a private enterprise there are strict rules that airlines have agreed to enforce.
How do the arguments hold up?
There are numerous models that compete and overlap with this book's model, such as evolutionary game theory (and evolutionary economics), behavioral economics, and myriad other flavors of economics. However, the authors seem especially critical of classical economics.
This book argues fairly persuasively that cooperative economics is the better model in certain scenarios. Could it displace classical economics more generally? So far as I can tell there is no grand unified model of economics. Even this book admits that classical economics has proven extremely successful in many practical situations. Moreover, this book admits that non-cooperative game theory is arguably a simpler model in many (though not all) practical scenarios. Likewise, evolutionary economics seems to be better suited in many practical scenarios.
Cooperative, non-cooperative, and evolutionary theories each have strengths in certain regimes. The book presents many such scenarios. It even presents scenarios in which the player encounters a scenario where it needs to switch from one type of model to a different one mid-game -- initially, the other players may seem like cooperating participants, only to suddenly snap into focus as self-regarding opponents.
Measured in dollars one could argue that classical economics dominates economics in the real world. Measured in number of interactions over the history of life as we know it then evolutionary economics could be argued as the most prevalent form. When measured in terms of what is most predictiveof what is going to happen in a high-stakes game involving nuanced gameplay by savvy human players, then the cooperative model has to be a strong contender for best model.
Non-cooperative models try to avoid the need for contracts, for example, by relying on reputation. At the end of the day though we still rely heavily on contracts in the real-world. Contracts are not going to disappear for the foreseeable future. Contracts are a correlating device, and so would seem to fall squarely within cooperative economics. Many of the apparent paradoxes in non-cooperative game theory evaporate when a correlating device is introduced.
The solution concepts presented in this book are computationally less expensive and more natural to apply in real life scenarios. For these reasons alone I considered the book worth my sustained attention. That it seemed to have lessons to teach about how people actually behave in group settings was a a bonus.
The end of the beginning
The remainder of the articles in this series (mostly) refrain from my editorializing and opinionating.
Here are links to the remaining parts:
Part i : Introduction.
Part 0 : The 14 lessons.
Part 1 : Overview of competing alternatives.
Part 2 : Failures of non-cooperative theory.
Part 3 : Evolutionary economics, rise of institutions, and the co-evolution of genes and culture.
There are comments.