© Copyright JASSS
Frédéric Gannon (2001)
Commentary on: Juliette Rouchier, Martin O'Connor, François Bousquet (2001)
Creation of reputation in an artificial society organized by a gift system
Journal of Artificial Societies and Social Simulation
vol. 4, no. 2,
To cite articles published in the Journal of Artificial Societies and Social Simulation, please reference the above information and include paragraph numbers if necessary
Paper's aims and assumptions
- The goal of the authors is to modelize the time dynamics of interindividual gifts in a little -50 members- artificial society. They aim to mimic well-documented ethnologic observations such as Potlatch, for which a periodic and ostentatious giving away of one's wealth is the cornerstone of social behaviour. . In this artificial society, giving away at each period is compulsory only under condition of minimum earnings, that is: only agents "wealthy" enough at the beginning of the period can afford to participate to this procedure. Two types of gifts are contemplated: sharing gift (denoted hereafter S) and prestige gift (denoted hereafter P).
- Let agent i having rank 20 at the beginning of period t. We'll assume he can afford any type of gift (hence, its wealth is more than three times the minimum consumption level). The sets of agents to which its sharing gift and its prestige gift would be given are, respectively: S(20)=(19,21,...,50), P(20)= (1,2,...19,21). This agent in turn belongs both to the sets P(j) of agents j ranking from 19 to 50, hence S(20), and to the sets S(k) of agents k ranking from 1 to 21, hence P(20).
- In the model, the agent does not calculate an optimal pure strategy (prestige or sharing gift) by means of reasoning but instead chooses it at random, as is shown in figure 2. However, two key parameters are used to modelize the choice of the type of gift the agent eventually makes. The first one is its degree of self-confidence, that the authors call esteem. It is assumed to be initially randomly distributed among agents (between values 2 and 8). When it is allowed to vary with time, it increases whenever the agent makes a gift and decreases twice as much when the latter receives no gift at all. Since current esteem level explains the kind of gift the agent is eventually going to make -the higher the esteem the more likely it will opt for a prestige gift, ceteris paribus- a succession of periods with no gift received will induce this agent not to make gifts at all. The second one is its motivation for prestige (a parameter lying between 0 and 10) and integration (idem), respectively. This notion reflects the idiosyncratic taste of the agent for the two fundamental "quality indicators" that this society makes up. Essentially, the agents are, like in real life, not necessarily equal as to their natural tastes.
Paper's main result : a typology of social order
- The authors melt the two idiosyncratic attributes -motivation and esteem- into the concept of "personality", which is an a priori reasonable though strong artefact of reality. They identify it with that of "preferences" used by economics. This assimilation is correct since in a basic utilitarist setting, initial structures of preferences are given, even if, in more elaborated models (time evolving utility functions, as in adicted-based utility functions for example), current preferences are explained by past individual actions. In that sense, the rationality of any player can be qualified of perfect. Here, agents are interrelated because of their mutual belonging to any other's P() or S() sets. This "game" has a common knowledge structure as for the reputation of the agent, since it is given by its current rank. It is not quite clear in the model, however, whether its motivation and esteem current levels are common knoledge. It seems these are personal data, which can only be approximatively deduced by the trajectory of the individual. There would then be an information asymmetry in the model, but it plays no strategic role.
- Basically, the central result of the paper lies in the "social order" typology conditioned by the assumptions made about the time evolution of the parameters. Explicitly, this social order consists in the long-run number of differentiated "classes" of agents in this artificial small society: elite, intermediate and dominated.
Questions and comments
- Hereafter, a series of questions and remarks is proposed, which have no other aim but helping a profane reader (as the present referee) to make sure he (she) really understands the underlying structure of the model.
The role of randomness
- Randomness plays an important part in the model, either in the initial distribution of ranks and individual characteristics or in the making of "strategy". If the number of simulations made by the authors allows for a comprehensive analysis of the first type of randomness, the way the second type (strategic-related one) is contemplated seems to lack precision. What is the exact motivation for the test procedure presented in figure 3? It appears that the drawing of a random (real) number between 0 and 1 (??) is used three times at least and four times at most. This has an important consequence, in terms of the agents' "rationality": since randomness seems to play such an important part, one cannot even evoke a game-theoretical-approach concept of rationality. The agents in fact does not have strategies, since they obey simultaneously to their intrinsic "personality" (to use the term the authors use themselves) and to randomness. Clearly then, this model could not be analysed in a game theory approach, which crucially relies on a minimum "real" rationality, hence on the possibility to make errors. This in turn leads to another question: what about the "mutants", which characterize usually any evolutionary system? Here they are simply out of scope, they are not even contemplated. Since that in evolutionary game theory, the key concept is evolutionary stability, i.e., grossly, the degree to which the system is mutant-resistant, this present model offers no "bridge" connecting the two types of approaches, Multi-agent system and Evolutionary Game.
- Another form of uncertainty lies in the model. Agents' and community's memory is limited to 25 periods, so that the collective ranking is updated. So, it means the agent knows the consequences of the current actions (its and the others') only with a time lag of 25 periods. He can then only make assumptions about the strategies of others, which seriously reduce the interaction between agents. Clearly, the overwhelming use of randomness and uncertainty in the model forbids a parallel analysis in a game-theoretical approach. In effect, even for a Bayesian equilibrium, the beliefs' revision needs a regular feeding of just past actions and payoffs of actions for the players to behave rationally.
- In the paper, long-run evolution (1000 time-steps) is analysed. "Social order" is then implicitly defined by the state of the system at the end of the simulation's time length. But no strictly defined concept of equilibrium is given. Is there possibility to do it, for example a almost stationary equilibrium such that two successive iterations give two almost similar state? In the same line of reasoning, is it possible to deduce the system's trajectory backward? That is, given a state at time t, characterize past from the last 25 periods (corresponding to the memory length of the institution) evolution?
Equivalence between inter-individual gift exchange and impersonal offer.
- This paper deals with a personal -though artificial- gift exchange device intermediated by the community of all members. No personal stable relationship can then build up, and every agent has to rely on the rules of random distribution of its gift, leaving it only the random-based choice of the kind of gift it is more efficient for him to do. But, other forms of giving away (partially) one's wealth have existed and still exist in real societies. Let us think for example of the ex voto tradition in the catholic churches. There, a saint is chosen by members of the community, to whom gifts are made to thank him (her) for his (her) recent intervention in the life of the thanksgiving parishionner. In particular, these gifts can take the form of marble (expensive) plates riveted on the ground or on the walls of the church, clearly and publicly indicating the name of the donator as well as the reason why the ex voto has been made, rendering them common knowledge to the community. In this tradition, the saint serves just as in the present model as an intermediation between people, who show off their ability to make expensive offers, and who signal themselves as well-off members, creating an incentive for other well-off agents to imitate them and to form a group of respected people. So intensively followed was this practice in the Renaissance Florence that documents attest of churches from the ceilings of which hung hundreds of saint-looking-mannequins very richly dressed, carrying the name of the family that had made the offer, so that everybody could easily evaluate and follow the reputation of the participating families. One could also, in the same order of ideas, evoke the habit of building always higher towers on the dwellings of rich merchants' dynasties, etc. In a word, is the assumption of interindividual gifts a necessary condition for a comparable society to evolve the way analysed in the paper?
The role of imitation
As Harrington (1999) puts it, "social systems are mainly hierarchical, and high-rank agents are imitated. The motivation to "move up" though not equally shared among all individuals is one of the essential feature of the aspiration-level approach of society". In this model, nothing is said explicitly about imitation, even if it works implicitly through the intermediated choice of agents to whom each type of gift -prestige or sharing- is eventually made. Agents act independently, without even trying to take into account the actions of members they know more prestigious than they are. Would explicit imitation dramatically modify the dynamics of the system?
HARRINGTON J.E, (1999) Rigidity of social Systems. Journal of Political Economy. Vol. 1, N° 107, pp 40-64.
Return to Contents of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2001