Artificial Intelligence for the Avoidance of Crises and Wars

Enormous resources have been invested in the use of AI methods and AI techniques in military defence. They are mainly aimed at replacing humans in battlefields and at improving military decisions. Can AI methods be used only "to make warefare efficient", or can they also contribute to avoiding wars, e.g. by supporting decision makers who want to avoid war in a situation of crisis, or by putting solid arguments into the hands of political groups who are justly opposing a government's policy heading at war? Can AI methods furthermore serve as scientific tools for international relations studies?

Summary of the Project "The Use of Artificial Intelligence Methods in Aiding Decision Makers or their Advisors to Prevent the Outbreak of Hostilities/Wars or to End Them": Following the motto of this research program "Promoting Peace and Preventing Violence", in the research project the question was studied how artificial intelligence methods, especially neural computation and machine learning methods, applied to conflict management databses, could aid decision-makers, their advisors, and also concerned groups of persons in their efforts to prevent the outbreak of hostilities or wars or to end them. A project team, consisting of Jacob Bercovitch, Georg Dorffner, Johannes Fürnkranz, Erik Hörtnagl, Christian Holzbaur, Jürgen Rattenberger, and Robert Trappl (project leader) undertook this task.

As a result of prior research by some members of the team the Confman database of international conflict management, developed under the guidance of Jacob Bercovitch of the University of Canterbury in New Zealand, had turned out to be the most comprehensive database, worldwide. However, the entrances in the database ended in 1995. Therefore, as a first step of the project, in a big effort the database had to be updated to the year 2000, a period with many conflicts. The updated database now contains 333 conflicts with 5066 conflict management attempts, each one chracterized by 218 independent variables.

As a next step, appropriate AI methods had to be applied to obtain the desired results. As regards neural computation, two methods were used: Multilayer perceptrons (MLP) as non-linear classifiers to predict the outcome of conflict management, and self-organizing maps (SOM) as a flexible clustering and visualization method to explore the structure of the data.

The first two explorations were performed using the previously available version of Confman, containing data from 1945-1995. A thorough analysis of non-linear classification revealed only minor differences as compared to linear classifiers. However, classification performance significantly above chance could be reached. Self-organizing maps proved to be a viable technique for revealing interesting clusters and substructure in the data.

The third exploration was performed with the extended Confman, containing the data from 1945-2000. Here, focus was put on evaluating whether there are significant differences in the two subsets 1945-1989 and 1990-2000, assuming that 1990 marks the end of the cold war. Results show that this is indeed the case. Conflict management outcome is more predictable after 1989 and the SOM analysis leads to interesting interpretations.

Decision trees have the additional advantage of giving the amount of influence of each variable on the successful outcome of a conflict management attempt. They therefore enable to guide decision-makers to more succcessful actions. The decision trees computed for the subsets 1945- 1989 and 1990-2000 are not only significantly different, but the one for the time period 1990-2000 has a higher predictive validity than the other one. This result is in concordance with the one obtained by the MLP analysis, however, with the added advantage of the decision trees. Furthermore, decision trees for different world regions were computed and the differences between the regions analyzed, thus giving decision-makers hints in which region which values of variables are more favourable for a successful conflict management outcome.

A database as such, with millions of data, does not make life easy for someone who wants to obtain more complex informations. Therefore, the development of an interface was begun to make the database useful for a political scientist or a student who wants to study the relationship between different variables and conflict management outcome.

A special concern was given to the decision-maker who is faced with a specific crisis situation and who wants to know which crisis situations are most similar to the current one in order to study which conflict management attempts were successful in these similar situations, also seeing differences in these other situations in order to adapt her/his measures to the current circumstances. This is done in this interface by another method of machine learning, case-based reasoning (CBR), by finding the five most similar cases to the given case.

By using screen-shots, the functionality of the interface is explained. This interface has yet to be improved regarding several aspects and then has to be tested by domain experts with respect to its usefulness before it can be made available to a larger community. But, judging from first experiences, at least political scientists seem to like it.

In order to obtain expert feedback on our project methods and results and to give a forum for the exchange of other ideas on how to handle these problems, a two-day workshop was organized in Vienna, at which 14 scientists participated, namely three from Austria, one from England, two from Germany, one from Switzerland, and seven from the USA.

The area of presentations ranged from the improvement of better information capturing, methods for the analyses of newswires, information-theoretic measures, attempts for conflict resolution in multi-agent systems, to scenario building. Not only did the members of the project team get useful comments, they were also asked for advice on other tasks, and in one instance already a cooperation with a US-American research group materialized.

The participants agreed that this was a very useful meeting and most of them will contribute to a book which will present the results of this project, jointly with the efforts of other research groups which all aim at "Promoting Peace and Preventing Violence".

Publications

  • Trappl R. (ed.), 2006: Programming for Peace: Computer-Aided Methods for International Conflict Resolution and Prevention. Springer Academic Publishers, Dordrecht, NL., 2006 Synopsis of the Book: Sadly enough, war, conflicts and terrorism appear to stay with us in the 21st century. But what is our outlook on new methods for preventing and ending them? Present-day hard- and software enables the development of large crisis, conflict, and conflict management databases with many variables, sometimes with automated updates, statistical analyses of a high complexity, elaborate simulation models, and even interactive uses of these databases. In this book, these methods are presented, further developed, and applied in relation to the main issue: the resolution and prevention of intra- and international conflicts. Conflicts are a worldwide phenomenon. Therefore, internationally leading researchers from the USA, Austria, Canada, Germany, New Zealand and Switzerland have contributed. The Introduction of the book is available online.
  • Kovar K., Fürnkranz J., Petrak J., Pfahringer B., Trappl R., Widmer G., 2000 Searching for Patterns in Political Event Sequences: Experiments with the KEDS Database, Cybernetics and Systems, 31(6)649-668
  • Robert Trappl, Johannes Fürnkranz, Johann Petrak, and Jacob Bercovitch, 1997 Machine Learning and Case-based Reasoning: Their Potential Role in Preventing the Outbreak of War or in Ending Them. to appear in: G.Della Riccia, R.Kruse and H.-J.Lenz (eds.), "Learning, Networks and Statistics". CISM Courses and Lectures no.382, Springer-Verlag, Wien New York, pp.209-225
  • Johannes Fürnkranz, Johann Petrak, and Robert Trappl, 1997 Knowledge Discovery in International Conflict Databases. Applied Artificial Intelligence, 11(2), pp. 91-118
  • Robert Trappl, Johannes Fürnkranz, and Johann Petrak, 1996 Digging for Peace: Using Machine Learning Methods for Assessing International Conflict Databases. In Proceedings of the 12th European Conference on Artificial Intelligence (ECAI-96), pages 453-457, Budapest, Hungary, 1996. John Wiley & Sons.
  • Johannes Fürnkranz, Johann Petrak, Robert Trappl, and Jacob Bercovitch, 1994 Machine Learning Methods for International Conflict Databases: A Case Study in Predicting Mediation Outcome. Technical Report TR-94-33. Available as gzipped postscript file
  • Robert Kopecny, 1994, Die Untersuchung der Konfliktentwicklung in der Alker-Datenbank mittles C4.5. Technical Report TR-94-30.
  • Johann Petrak, Robert Trappl, and Johannes Fürnkranz, 1994 The Possible Contribution of AI to the Avoidance of Crises and Wars: Using CBR Methods with the KOSIMO Database of Conflicts. Technical Report TR-94-32. Available as gzipped postscript file
  • Johann Petrak, 1994, VIE-CBR - Vienna Case-Based Reasoning Tool, Version 1.0: Programmer's and Installation Manual. Technical Report TR-94-34.
  • Johann Petrak, 1995, VIE-CBR2 - An Object-Oriented Case-Based Learning System. Diploma Thesis.
  • Robert Trappl, 1986, Reducing International Tension through Artificial Intelligence: A Proposal for 3 Projects, pp.97-103 of: R. Trappl (Ed.), Power, Autonomy, Utopia: New Approaches Towards Complex Systems. Plenum: New York.
  • Robert Trappl and Silvia Miksch, 1991, Can Artificial Intelligence Contribute to Peacefare?, pp.21-30 of: V. Marik (Ed.), Aplikace Umele Inteligence AI'91, Elektrotechnicka fakulta CVUT, Prague.
  • Robert Trappl, 1992, The Role of Artificial Intelligence in the Avoidance of War, pp.1665-1680 of: R. Trappl (Ed.), Cybernetics and Systems '92. World Scientific Publishing: Singapore.
  • Robert Trappl, Sigrid Unseld, and Silvia Miksch, 1993, Artificial Intelligence und die mögliche Vermeidung von Krisen und Kriegen in: G. Hanke (Ed.), Informations- und Kommunikationstechnologie für das neue Europa. Vienna: ADV.

Research staff

Partners

Key facts