Andart

Norms and Hortators in Small Worlds

How do we achieve cooperation without coercive central organisations? The classical example is Axelrod's analysis of strategies in the evolutionary iterated prisoner's dilemma, where he could show how reciprocal strategies could eventually dominate and form largely cooperative societies. Based on the idea of hortators, individuals acting as coordinators of social information and reciprocal rejection of defectors, I discuss some simulations and ideas around the formation of networks for voluntaristic enforcement of cooperation in this somewhat rambling account of unfinished hobby experiments in simulated sociology.

In the novel The Golden Age (and sequels) by John C. Wright a future libertarian society is described. In this near-utopia there are very few restrictions on individual freedom and nearly infinite possibilities of abusing the advanced technology available. In order to manage this problem the institution of the Hortators has developed. Essentially the Hortators are busybodies maintaining the social order by pronouncing different things desirable or undesirable, and if somebody breaks the social order or behaves unethically they ostracise him or her. Since the vast majority supports the Hortators they will also follow their lead, making an ostracism unpleasant, a serious economic hurt or even near-fatal if it is total and indefinite. Of course, the Hortators are not perfect (and most of the novel deals with one individual's struggle against them when he thinks they are wrong), but Wright makes an interesting point: an entirely voluntary organisation can likely maintain social norms with a high degree of compliance if it is broad enough. But what about smaller organisations, or societies with several different groups?

Axelrod's Norm Game (Axelrod,R. (1986). An evolutionary approach to norms. American Political Science Review 80, 1095-1111) is a variant of this issue. Axelrod is maybe most well known for his studies of the evolution of strategies in the iterated prisoner's dilemma. The Norm Game is a somewhat simpler abstraction but still with nontrivial effects. Individuals can choose to behave well or badly, and will gain a certain reward for acting badly while incurring a global penalty on everybody else (it could be overgrazing of a common, polluting or maybe spamming). If strategies are adopted depending on how successful they have been in the near past we get an evolutionary game. Without any other assumptions the only stable strategy here is of course to defect all the time. What if other individuals who sees the defection can strike back, incurring a penalty for the wrongdoer (but also losing out a bit themselves)? It turns out again that defection reigns. By introducing meta-punishments where individuals punish others who see wrongdoings but do nothing Axelrod found a stable cooperative state.

Others have explored changing his assumptions, introducing bounties for norm-enforcers, exploring different kinds of networks etc. In general the effect of the social network seems to be complex. If everybody is connected to everybody else many will react to a norm breaker and punish him, producing a strong normative effect. Real social networks are "small world networks" where most of my friends know each other (they are local, like a network based on being close to each other in space) but there still are enough long-range links to make it possible to find a short path from one person to another: the classic "six degrees of separation" (I discovered to my amusement that I am just three social steps from Saddam Hussein myself). In a local network there are fewer people who directly know me, and hence norm management is going to be local. While local links such as in a family are likey stronger than long-range links, the effect weakens the normative effect. When the network is a small world information about a transgression can spread widely, but it is unlikely to exist a direct social link from the vast angry majority to the transgressor.

I did some simulations myself of the Norm game, assuming different kinds of social networks, sizes of rewards and punishments and how information about a norm breaker spread. Very interesting but hard to say anything general about, since there are so many interacting parameters. Dense social networks promote cooperation, since many are likely to notice norm breaking. One thing I noticed was that an efficient way of achieving norms was to have punishments to be global: regardless of social distance others incurred costs for the norm breaker. While meta-normativity may still be a stabilizing factor, often the combined effect of many people punishing an offender individually weakly had enough a strong deterring effect to make everybody behave themselves stably. Fortunately this makes sense given the assumption of global costs incurred by the offender; he is behaving badly in the public sphere, and while information might travel according to social networks, the punishment also takes place in the public sphere.

In a small world network certain widely linked individuals can help spread information (or a rumor or epidemic; from a modelling perspective it does not matter if it is memes or genes). Hence if these act as Hortators they would enable concerted action. Also, by being strongly linked they may be efficient at convincing people about the rightness of their actions: some other interesting simulations of social networks with 'leaders' who seek to convince others suggest that the virtue of being well connected is more powerful than persuasive 'strength'. In simulations with deliberately introduced widely connected individuals these tend to act as mediators of information and make sure the offender is broadly punished, even when the cost of punishment for the punisher was noticeable.

So it seems that having Hortators might make norm enforcement more powerful. But how do we get Hortators? They have to pay enforcement costs, and since they can be expected to react more strongly than the average population (which contains individuals that may not be fully cooperative) they might have a disadvantage. In my simulations I often noticed the spontaneous appearance of Hortators that persisted for a few generations but then vanished, simply because they scored less than the less conscientous population. If hortatorship was something agents could decide on, it usually seemed to decline in the population over each generation.

In Wright's world people subscribed to the Hortators, apparently paying a small fee for their maintenance. They also had support from several major monopolies that gave them not just economic support but significant social credibility. One could imagine subscribing to a hortator service being an useful social signal. Members of the service would display membership, showing that they are less likely to defect and hence gain an advantage over non-members when another individual seeks a trading partner. It would be similar to how ISO9000 is sometimes used: in order to ensure quality, certified companies often seek out other certified companies as suppliers, making it an advantage to join. If the advantage of joining is greater than the cost of occasionally punishing defectors and maintaining the hortator(s) at the core, then the network will grow. As it grows, the cost of defecting increases and defection tends to decrease. Eventually an equlibrium likely develops, where the cost of being part of the network exactly equals the benefit. Occasional fluctuations occur due to random defections. This seems to be a very fruitful area to model. Also, this kind of scheme would allow multiple competing (or cooperating) hortator networks.

Preliminary simulations suggest that there are strong threshold effects in when hortator networks manage to enforce norms; they need to be sufficiently large initially. When individuals can change their social links depending on past experience with others many interesting patterns can develop (c.f. partner selection in the iterated prisoner's dilemma), and it seems likely that these are necessary to fully model how a working hortator structure might work.

Hortator networks are an interesting case in the middle between individual distributed norm enforcement and having a government or other dedicated organisation perform it. Under the right conditions hortators can emerge as spontaneous orders. However, we still need to understand their dynamics in order to find out under what conditions they can be used as effective tools, and how they in that case should be organised, paid for and what their contracts would include.

Note that the values transmitted by Hortators do not have to be good values. It is quite possible to imagine Hortators sustaining stifling values in a society. This is why it is important to model interactions between hortator groups and how they affect multiple norms - just as we want to reduce defection from social values we want to reduce overconformism. Maybe there is a meta benefit in the relative instability of most self-organised schemes to deal with defection.

Or more simply, the frailty of human nature saves us from overly powerful institutions.

Posted by Anders at January 5, 2004 03:47 PM