Bayesian Rationality, Social Epistemology, And The Choreographer

This paper suggests that a social norm is better explained as a choreographer—a correlating device with causal effectivity for a correlated equilibrium of an underlying stage game—rather than a Nash equilibrium of the stage game. Whereas the
of 26
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  Bayesian Rationality, Social Epistemology, and theChoreographer Herbert Gintis  September 26, 2009 Abstract This paper suggests that a social norm is better explained as a  choreog-rapher  —a correlating device with causal effectivity for a correlated equilib-rium of an underlying stage game—rather than a Nash equilibrium of thestage game. Whereas the epistemological requirements for rational agentsplaying Nash equilibria are very stringent and usually implausible, the re-quirements for a correlated equilibrium amount to the existence of   common priors , which we interpret as induced by the cultural system of the society inquestion. In thisview, human beings may be modeled as rational agents withspecial neural circuitry dedicated to reacting to, evaluating, and sustainingsocial norms.When the choreographer has at least as much informationas the players,we need in additiononly to posit that individualsobey the social norm whenit is costless to do so. When players have some information that is not avail-able to the choreographer (i.e., not all social roles can be fully incentivized),obediencetothesocial normrequiresthatindividualshave apredispositiontofollowthe norm even when it is costly to do so. The latter case explains whysocial norms are associated with other-regarding preferences and provides abasis for a general analysis of corruption in business and government.Social normsarethusnotexplainedintermsofgame theoryandBayesianrationality, but rather are an  emergent property  of human society, which is acomplex adaptive system guided by natural selection. Social norms providea dimension of causal efficacy to social theory, whereas game theory alonerecognizes no causal efficacy above the level of individualchoice behavior.Because of the independent causal effectivity of social norms, the stan-dard methodological individualismof classical game theory is untenable. Inparticular, social norms are predicated upon certain mental predispositions,  Santa Fe Institute and Central European University. Some of this material is adapted frommy book   The Bounds of Reason: Game Theory and the Unification of the Behavioral Sciences(Princeton, 2009). 1  a  social epistemology , that is also a product of natural selection. This so-cial epistemology fosters the interpersonal sharing of mental concepts, and justifies the assumption of common priors upon which the identification of Bayesian rationalitywith correlated equilibriumrests.Keywords: Nash equilibrium, correlated equilibrium,social norm, socialepistemology, correlating device, honesty, corruption, Bayesian rationality. 1 Introduction The coordination of social life is effected by  social norms  that indicate the appro-priate behavior of the individual in particular social roles. There is a long tradi-tion in social theory of treating social norms as Nash equilibriaof non-cooperativegames played by rational agents (Lewis 1969, Taylor 1976, 1982, 1987, Sugden1986, 1989, Bicchieri 1992, 1999, 2006, Binmore 1993, 1998, 2005). The in-sight underlying these contributions is that if agents play a game  G   with severalNash equilibria, a social norm can serve to choose among these equilibria, thuscoordinating expectations and behavior, as well as possibly avoiding socially in-efficient equilibria. While this insight applies to some social conventions, suchas driving on the right side of the road, it does not apply to most important casesof human cooperation (e.g., norms concerning proper behavior in particular socialroles, such as physician or police officer), and it cannot account for the salience of other-regarding preferences as part of thehuman behavioral repertoire, because theNash equilibriumcriterion presupposesthat agents choosebest responsesgiven thegame’s payoffs. More generally, a social norm takes the form of adding a correlat-ing device, so that the resulting behavior is a correlated equilibriumof the srcinalstage name.For instance, in the automobile traffic case, correlating devices such as lightsand stop signs augment the traffic conventions and effect a correlated equilibrium.Similarly, in the hawk-dove game, adding the so-called “bourgeois” social norm,which says that the player who first occupies a desired location behaves like ahawk and the player who confronts such an incumbent acts like a dove leads tosocial efficiency. In these cases, we may assume players are self-regarding butshare a common prior as to the nature of the correlating device. A third example isthe social norm “a police officer must not accept a bribe from a driver in decidingwhether to issue a traffic ticket.” Assuming the driver’s behavior is known onlyto the driver the police officer in question, the social norm is effective only if theofficer is willing to sacrifice personal gain in favor of complying with the norm.The correlating device that maintains the reputation of the police force in this caseis the social norm of refusing bribes, compliance with which may be high wheneach officer considers the norm legitimate, perhaps because the norm is socially2  beneficial and others officers appear to comply with the norm.In this paper I will suggested that a social norm is a  choreographer   of a su-pergame  G  C of   G  . By the term ‘choreographer’ I mean an correlating device thatis common knowledge to the players, and that is causally effective in implement-ing a correlated equilibrium of   G   in which all agents play best responses to thechoreographer’s signals, given their personal (perhaps partially other-regarding)preferences and a common prior.The social norm as choreographer has three attractive properties lacking in thesocial norm as Nash equilibrium. First, the conditions under which rational agentsplay Nash equilibria are generally complex and implausible (e.g., that all playersshareacommon conjectureconcerningthestrategychoiceofeach player), whereasrational agents with a common prior canonically play a correlated equilibrium.Second, thesocial norms as Nash equilibriaapproach cannot explain whycom-pliance with social norms is often based on other-regarding moral preferences inwhich agents choose to sacrifice some personal gain to comply with a social norm.We can explain this association between norms and morality in terms of the in-complete information possessed by the choreographer. Morality, in this view, isdoing the right thing even if no one is looking—as long as the cost of so doingis not excessive. Because the motivation to behave morally depends on cost andthe general level of compliance, moral behavior may characterize some but not allsocial equilibria.Finally, there are many more correlated equilibriathan Nash equilibriato mostgames. Some of these equilibria are Pareto-superior to all Nash equilibria and/orallow a distribution of payoffs among players that is unavailable with Nash equi-libria. 2 Epistemic Games Bayesian rational players have beliefs concerning the behavior of the other play-ers, and they maximize their expected utility by choosing best responses giventhese beliefs. Thus, to investigate the relationship between Bayesian rationalityand strategicbehaviorwe must incorporatebeliefs intothedescriptionof thegame.An epistemicgame G  consistsofa normalform gamewithplayers i  D 1;:::;n and a finite pure-strategy set  S  i  for each player  i , so  S   D Q ni D 1  S  i  is the set of pure-strategy profiles for  G  , with payoffs   i W S  ! R . In addition,  G   includes a setof possible states    of the game, a  knowledge partition P  i  of     for each player  i ,and a  subjective prior  p i . I !/  over    that is a function of the current state  !  2   .A state  !  specifies, possibly among other aspects of the game, the strategy profile s  used in the game. We write this  s  D  s .!/ . Similarly, we write  s i  D  s i .!/  and3  s  i  D s  i .!/ .The subjective prior  p i . I !/  represents  i ’s beliefs concerning the state of thegame, including the choices of the other players, when the actual state is  ! . Thus, p i .! 0 I !/  is the probability  i  places on the current state being  ! 0 when the actualstate is  ! . We write the cell of the partition  P  i  containing state  !  as  P i ! , andwe interpret  P i !  2  P  i  as the set of states that  i  considers possible (i.e., that havestrictly positive probability) when the actual state is  ! . Therefore, we require that P i !  D f ! 0 2   j p i .! 0 I !/ > 0 g . Because  i  cannot condition behavior on aparticular state in the cell  P i !  of the knowledge partition  P  i ,  i ’s subjective priormust satisfy  p i .! 00 I !/ D p i .! 00 I ! 0 /  for all  ! 00 2   and all  ! 0 2  P i ! . Moreover,we assume a player believes the actual state is possible, so  p i .! I !/ > 0  for all !  2  .If   .!/  is a proposition that is true or false at  !  for each  !  2   , we write Œ  Df !  2  j .!/ D true g ; i.e.,  Œ  is the set of states for which    is true.We call a set  E      an  event  , and we say that player  i  knows  the event  E  atstate !  if  P i !   E ; i.e.,  ! 0 2  E  for all states  ! 0 that  i  considers possibleat  ! . Wewrite K i E  for the event that  i  knows  E .Given a possibilityoperator P i , we define the  knowledge operator  K i  by K i E  Df ! j P i !   E g : The most important property of the knowledge operator is  K i E    E ; i.e., if anagent knows an event  E  in state  !  (i.e.,  !  2  K i E/ , then  E  is true in state  !  (i.e., !  2  E/ . This follows directly from  !  2  P i !  for all  !  2   . We can recoverthe possibility operator  P i !  for an individual from his knowledge operator  K i ,because P i !  D \ f E j !  2 K i E g : Since each state  !  in epistemic game  G   specifies the players’ pure strategychoices s .!/ D . s 1 .!/;:::; s n .!// 2 S  , the players’ subjectivepriors must specifytheir beliefs   !1  ;:::; !n  concerning the choices of the other players. We have  !i  2  S   i , where  S   is the set of probability distributions over set  S  , whichallows i  to assume other players’ choices are correlated. This is because, while theother players choose independently, they may have communalities in beliefs thatlead them independentlyto choose objectively correlated strategies.We call   !i  player  i ’s  conjecture  concerning the behavior of the other play-ers at  ! . Player  i ’s conjecture is derived from  i ’s  subjective prior   by noting that Œs  i   D def   Œ s  i .!/ D s  i   is an event, so we define   !i  .s  i / D p i .Œs  i  I !/ , where Œs  i       is the event that the other players choose strategy profile  s  i . Thus,at state  ! , each player  i  takes the action  s i .!/  2  S  i  and has the subjective prior4  probability distribution  !i  over  S   i . A player  i  is deemed  Bayesian rational  at  ! if  s i .!/  maximizes   i .s i ; !i  / , where  i .s i ; !i  / D def  X s  i 2 S  i  !i  .s  i / i .s i ;s  i /:  (1)In other words, player  i  is Bayesian rational in epistemic game  G   if his pure-strategy choice s i .!/ 2 S  i  for every state  !  2   satisfies  i . s i .!/; !i  /   i .s i ; !i  /  for  s i  2 S  i :  (2) 3 The Epistemic Conditions for Nash Equilibrium Suppose that rational agents know one another’s conjectures in state  ! , so that forall  i  and  j  ¤ i , if   !i  .s  i / > 0  and  s j   2  S  j   is player j ’s pure strategy in  s  i , then s j   is a best response to his conjecture   !j   . We then have a genuine “equilibriuminconjectures,” as now no agent has an incentive to change his pure strategy choice s i , given his conjectures.We say a Nash equilibriumin conjectures  . !1  ;:::; !n /  occurs at  !  if for eachplayer  i ,  s i .!/  is a best response to   !i  . We then have the following theorem(Aumann and Brandenburger 1995):T HEOREM  1.  Let   G   be an epistemic game with  n > 2  players, and let    ! D  !1  ;:::; !n  be a set of conjectures. Suppose the players have a common prior  p  that assignspositiveprobabilityto it being mutuallyknown that the game is G   , it is mutuallyknown that all players are rational at   ! 2   , and it is commonlyknownat  !  that   ! is the set of conjectures for thegame. Then, for each  j  D 1;:::;n  , all i ¤ j  induce the same conjecture    j  .!/  about   j ’s action, and   .  1 .!/;:::;  n .!//  form a Nash equilibrium of   G  . These conditionsare not necessary, but they are strict; i.e, relaxing any one rendersthe conclusion untrue. The condition that the conjectures be commonly knownis especially stringent, and is highly implausible except possibly for games thatare extremely simple, have a unique Nash equilibrium, this equilibrium uses purestrategies, and it can be found by the elimination of strictly dominated strategies(Gintis 2009, Ch. 4). Moreover, unless all conjectures are themselves pure strate-gies, there is no incentive for any player actually to conform to the conjecturedplay. This is because all of the mixed strategies that occur in a player’s best re-sponsehave equal payoffs against theconjectures of theother players, so all mixedstrategies in the support of the best response are equally good candidates for play.Moreover, because this is common knowledge, no player has grounds for believ-ing the other players will play their conjectures either. It follows that no player5
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!