Krieg & Frieden

Da die Prädiktion von Krieg und Frieden einen wesentlichen Schwerpunkt unserer Forschung darstellt, werden im folgenden die neuesten Resultate unserer Arbeit vorgestellt (Olbrich & Hergovich, 2001).

 

A Neural Net for Predicting

War and Peace.

A.    OLBRICH, & A. HERGOVICH,

Department of Psychology, University of Vienna, Austria

 

Abstract

Background: Social Identity Theory (Turner, 1986), Theory of Integrative Complexity (Tetlock, 1985) and the Theory of Groupthink (Janis & Mann, 1977) provide powerful tools for predicting international conflicts and wars.

Aims: The aim of this study is to develop an application of artificial intelligence for predicting war and peace.

Methods: 192 documents of US-American history (articles from magazines and political speeches), which were published after WWII (1946-1990), were analysed with a questionnaire containing items from Social Identity Theory, Theory of Integrative Complexity and the Theory of Groupthink. The time-span 1946-1990 was divided in half-year-terms. Each term was evaluated as peace or war.  A 3-Layer feedforward neural net (1 Inputlayer, 1 Hidden Layer and 1 Outputlayer) was trained with 110 documents.  20 documents formed the validation set.

Results: 65% of the articles were classified correctly. The neural net predicted 90% of the wars. Though there was a bias to classify peaceful terms as war-like, the neural net seems valid.

Conclusions: The application of neural nets and the three theories of Social Psychology yield a good basis for political psychological consulting in international relations and conflict prediction.   

 Introduction

Since 1945 the international system  has changed in a significant manner. After WWII a bipolar world system emerged with two antagonistic super powers, which enlarged their hegemonic influence in the 1950's and 1960's. The decolonialization brought a third power to the system. The states of the Third World and the non-aligned countries resist the superpowers' hegemonic attraction. The rise of this third power strengthens the international and regional organisations (e.g.. U.N.; Rosenau, 1990). After the end of the Cold War and/or beginning of the  U.S.S.R's decay these organisations have grown to the most important dispute resolution entities. Today dispute resolution competence is distributed on a wide range of states and institutions. The complex system needs higher analytical abilities. Changes in subsystems have to be analysed and their implications for politics predicted. “Perhaps the most glaring paradox in the current era is that the need  for co-ordinated problem-solving on a global scale - in matters of security, economics, and ecology - is arguably greater than ever before, at the same time that global institutional capabilities are diminishing and 'central guidance' mechanisms seem  less feasible than in previous historical periods” (Rochester, 1990, p. 141).  The prediction of conflicts and also the resolution of disputes are difficult in a complex world.

Three psychological Theories

1.     Integrative Complexity

The measurement of integrative complexity assesses the degree to which a person is found to differentiate and integrate argumentation in information processing. According to the cognitive manager model it is preferable to gear one’s efforts to perform the tasks of differentiation and integration so as to adjust to situational requirements. Complex strategies are often more costly than simpler ones in both time and effort, and may divert attention and resources from crucial to trivial information.

Tetlock(1985) measured integrative complexity on a seven-point scale. It depends on the differentiation and the integration of facts into the problem-solving process.  High integrative problem solvers consider many facts for their decisions and try to satisfy all conflict parties. On the other hand low integrative decision makers know the facts but do not have the ability of making conclusions.

Suedfeld, Tetlock and Ramirez (1977) analysed speeches of Near-East diplomats at the General Assembly of the UN. They also found that the integrative complexity decreases before militarised disputes. Allies change in their complexity of arguments, as well. The U.S. complexity correlates with the Israeli complexity. Suedfeld, Tetlock and Ramirez suppose that the fate of  Israel influences the hegemonic structure of the super power. In contrast no correlation was found between the Arab states and the former USSR. A plausible interpretation is that the USSR has no concerns or interests in this region.

 

2.     Social Identity Theory

Tajfel and Turner (1986) proposed the social identity theory (SIT). All individuals strive to achieve a positive social identity, which is based on favorable comparisons that can be made between the ingroup and relevant outgroups. For a positive social identity the in-group must be perceived as positively differentiated or distinct from the relevant outgroups. When the social identity is unsatisfactory for the individuals, they will strive to leave the ingroup and join a positively distinct outgroup. Another possible way is to make the existing ingroup more positively distinct.

Wagner (1994) found, that opposite argumentation and opposition in general are not perceived in the beginning of war and crises. The groups (in-group and out-group) tend to be homogenous. The evaluation of ingroup and its leader get more positive. The outgroup and its leader get negatively evaluated.

 

3.     Groupthink

Constituent pressure may lead to Groupthink (Janis, 1982). Groupthink refers to a deterioration of mental efficiency, reality testing and moral judgement that results from in-group pressures. Three conditions must be satisfied for defective group behaviour: (1) a high homogeneity of group members, (2) a provocative situation (value threat to group), and (3) low self-esteem of group members. The cohesive group members desire unanimous agreement but other antecedents relate to basic structural faults in the group (lack of impartial leadership and to the immediate decision-making context, external threats and temporarily decreased self-esteem strengthen the tendency for biased discussion and consideration of objectives and alternative solutions. The advice of experts outside the group not to threat the weak group opinion remains unheard.

 

Methods

A questionnaire (22 items) was constructed on the basis of these three psychological theories (Integrative Complexity-Scale, Social Identity-Scale, Groupthink-items).

The time-span 1946-1990 was divided in half-year-terms. Each term was evaluated as peace or war. 130 articles (Newsweek, Time), which were published in this time period and 62 speeches (Inaugural addresses, State of the Union) were analysed by trained subjects.

First, two discriminant analysis were conducted.

In a second step the data were fed into a three-layered Feedforward Neural Net. The input layer consists of 22 neurons (every neuron represents one item of the questionaire). 10 neurons form the hidden layer and in the output layer 2 neurons represent the warlike and peaceful states. Training is based on the Backpropagation algorithm and is performed 1.000 times to maximize the discriminant validity, i. E. to minimize the predictive error. Figure 1 illustrates the design of the neural net.

Figure 1. The design of the Neural net

                                                                                                       

Results

In a first step two discriminant analysis revealed encouraging results in discriminating data from war and peace. 79.2% of the articles were correctly classified (cross-validation: 61.5%). 82.3% of the speeches were correctly classified (cross-validation: 59.7%). The social identity theory and groupthink-phenomenon seem to be valid theories for predicting war. The integrative complexity score show a tendentious decrease before war in the articles (This result was not confirmed in the speeches-data).

In a second step the neural net was trained with data (110 articles). 20 articles, which formed the validation-set, were classified by the neural net. 65% of the articles were correctly classified. 90% of the war-data and 20% of the peace-data were correctly classified. This means, that there is a strong bias toward war-classification.

References

Janis, I. L., Mann, L. (1977). Decision making: A psychological analysis of conflict, choice, and commitment. New York: Free Press.

Janis, I. L. (1982). Groupthink (2nd ed.). Boston: Houghton Mifflin.

Rochester, J. M. (1990). Global policy and the future of the United Nations. Journal of Peace Research, 27, 141-154.

Rosenau, J. N. (1990). Turbulence in World Politics. New York: Harvester-Wheatsheaf.

Suedfeld, P., Tetlock, P. E., & Ramirez, C. (1977). War, peace and integrative complexity: UN speeches on the Middle East problem, 1947-1976. Journal of Conflict Resolution, 21, 427-442.

Tajfel, H., & Turner, J. C. (1986). The social identity theory of intergroup behavior. In S. Worchel and W. G, Austin (eds.), Psychology of intergroup relations (pp. 7-24). Chicago: Nelson-Hall.

Tetlock, P. E. (1985). Integrative complexity of American and Soviet foreign policy rhetoric: A time-series analysis. Journal of Personality and Social Psychology, 49, 1565-1585.

Wagner, U. (1994). Eine sozialpsychologische Analyse von Intergruppenbeziehungen. Göttingen: Hogrefe.