© Copyright JASSS

  JASSS logo ----

G. Deffuant, G. Weisbuch, F. Amblard and T. Faure (2003)

Simple is beautiful … and necessary

Journal of Artificial Societies and Social Simulation vol. 6, no. 1
<http://jasss.soc.surrey.ac.uk/6/1/6.html>

To cite articles published in the Journal of Artificial Societies and Social Simulation, please reference the above information and include paragraph numbers if necessary

Received: 9- Jan-2003      Published: 31-Jan-2003


* Abstract

A contribution to the JASSS forum, in reaction to the paper in FASZ about our model of extremism

Keywords:
Extremist Attitudes; Opinion Dynamics; Sociophysics

1.1
First of all, we would like to thank the journalists, Gero von Randow, from FASZ and Isabelle Cuchet from Courrier International for their interest in our work, as well as Klaus Troitzsch for the translation. The precise description of our model in a large audience newspaper, including the reproduction of several graphs, was a very good surprise for us. Moreover, the journalist used the model as a metaphoric source of reflection and questioning about the present social situation of Germany. We were very pleased by this rich and interesting use of our model, although we point out some caveats below.

1.2
Moreover, Gero von Randow surveyed other scientists about our paper, which is fair game, especially about such sensitive topics. Most of these views are critical, and we feel that it is also fair game to give some answers to these criticisms.

1.3
Rainer Hegselmann's remark is easily answered: we don't say that the centre is always insecure. We say that WHEN it is, extremism prevails. The paper gives examples of a secure centre leading to a global low influence of the extremists.

1.4
G. von Randow's criticism about the extremists dynamics, which according to him always leads to a clustering, is also easily answered: this is a wrong interpretation of the model. On the contrary, when the uncertainty decreases (i.e. when extremists get more and more convinced), the dynamics leads to a multiplicity of smaller and smaller clusters. This could be related to the observation that extreme movements tend to produce many small groups. Interestingly enough, our model predicts that this tendency to produce small groups decreases when the extremists have an audience in a larger population of more uncertain people, because uncertain people make a kind of bridge between isolated convinced small groups.

1.5
Petra Ahrweiler and Klaus Troitzsch raise a deeper debate about methodological options in social simulations.

1.6
We recognise that the incorporation of well grounded results from the social sciences in the models is a legitimate goal, as well as confronting the models with experimental data.

1.7
However, complicated models are not necessarily more realistic than simple ones. For example, popular models such as "Belief Desire Intention" architectures of agents manipulate common sense cognitive concepts, without much scientific ground. The status and the reality of an intention is not clearly established in neurosciences, psychology or philosophy. Other agent models include symbol processing or pseudo-language processing. This is not a guarantee of more realism either. These models are in general very far from the state of the art in linguistics. Moreover, little is known about the connection between language and emotions or desires.

1.8
A close link to the social sciences is not a guarantee of realism either. The current state of the social sciences shows a lot of different schools in competition, without any clear recognised paradigm providing generic rules of individual or collective behaviour. Moreover, well-recognised academic trends such as methodological individualism or structuralism seem to us quite strong approximations.

1.9
To summarise, making complicated models is very easy. Establishing strong results about their dynamical properties and relating them to evidence from psychology or sociology is the difficult part. We argue that the study of simple approximations is a good strategy for progress in this direction. We would like to illustrate this on the example of our model.

1.10
Our model has a clear link to common sense psychological observations:
  • beyond a given limit on disagreement, others' opinion are considered as foolish and are thus ignored,
  • very certain people tend to be more convincing than uncertain ones.

1.11
We simply defined mathematical functions which translate these rules, incorporating the hypothesis of continuity of the influence function. These options can of course be criticised, and refined, but they have the merit of relating to clear common sense hypotheses.

1.12
The social interpretation of the collective dynamics exhibited by the model is more delicate, and we left it open in the paper. Our result, summarised by the journalist as, "When the uncertainty of the centre is high, extremism prevails'', suggests the application of the model to political extremism. Common observations of political history advocate such an interpretation: extremist political movements tend to increase their audience in periods of high uncertainty. The Bolsheviks took power during the crisis of World War I in Russia, the Nazi party after the defeat of Germany and the huge financial crisis, one could probably find strong internal doubts in the Muslim world which are contemporary to a larger audience of Muslim extremism, and the rise of extremism in Eastern Europe and Russia after the collapse of the Soviet Union can also be an illustration.

1.13
However, the statement "when the uncertainty is high, extremism prevails" is not our main result. In some sense, it was built into the rules of interaction: the influence of the extremists increases when the uncertainty of the moderates increases. But one unexpected result is the prediction of quasi-unanimity on one extreme opinion, even if there was initially a balanced influence from both extremes. Quasi-unanimity was not observed with the previously mentioned political extremist movements, hence the social interpretation in terms of political extremism could not be the most relevant, although it is very tempting. It could be more relevant to relate the model to the dynamics of opinion in fashion, cultural traits or mob movements where such a quasi-unanimity is frequently observed. Consider for instance opinion about abortion or divorce in France: 50 years ago, to accept them as normal behaviour was an extreme position. This once extreme position is now quasi unanimously adopted by the whole population. We find exactly the same type of switches in the model. Moreover, it seems reasonable to consider that the evolution of these opinions is made by repeated small changes due to social influences, as they take place in the model.

1.14
In fact, the initial goal of this model was to represent the evolution of the opinion about the environment. This model was one part of a more complicated model of the diffusion of environmental practices within farmer communities. The complete model incorporates other features such as: economic rational anticipation, decision function over several criteria, dynamics of information transmission, etc. (see Deffuant 2001, Deffuant et al. 2002). This more complicated model was partly derived from interviews and experimental data. This model would probably be more appropriate for describing political extremism, because it includes economic interests, expectations, past history, which can be determinant in political choices. In this respect, we recognise with P. Ahrweiler and K. Troitzsch that the model presented in the JASSS paper is probably too simple to be applied to concrete political situations without much caution.

1.15
However, in our view, this does not lower the interest of this model. First, because a serious study of the more complete model requires us to understand how this simple model works. Even if a systematic study of this complete model by simulation was tractable (which it is not because of the number of parameters), the interpretation of the results of millions of simulations is almost impossible without a minimum knowledge about the more elementary model.

1.16
Therefore, even though we acknowledge the legitimacy of exploring more elaborated psychological or social dynamics by simulation, we do argue that studying simplifications of these elaborated models is a necessary stage. It allows the researchers to establish robust results and stylised facts which constitute references for the study of more complicated dynamics.

1.17
Moreover, the considered simplification could have an interesting social interpretation on its own, as an idealisation of the evolution of cultural opinions, which is part of the story in political opinion evolution. Therefore, we claim that our model establishes a clear stylised fact, which constitutes a useful reference in the study of extremism influence in general, even though it corresponds to a very idealised situation.

* References

DEFFUANT, G. 2001. Final report of project FAIR 3 CT 2092. Improving Agri-environmental Policies: A Simulation Approach to the Cognitive Properties of Farmers and Institutions. <http://wwwlisc.clermont.cemagref.fr/imagesproject/default.asp>

DEFFUANT, G., Huet, S., Bousset, J.P., Henriot, J., Amon, G., Weisbuch, G. 2002. "Agent based simulation of organic farming conversion in Allier département". in Complexity and Ecosystem Management. M.A. Janssen editor. Edward Elgar Publishers.

----

ButtonReturn to Contents of this issue

© Copyright Journal of Artificial Societies and Social Simulation, [2003]