An Attempt to Mechanize Sociology by Artificial Intelligence Methodologies Jacques Calmet Karlsruhe Institute of Technology (KIT), Institute IKS, Am Fasanengarten 5 Karlsruhe, D-76131, Germany [email protected]

Extended abstract Keywords: Sociology, Information Technology, Specification, Trust, Mathematics, Virtual Communities. The goal of this document is to design a model enabling a formal specification of sociology in the framework of information and communication technology (ICT). We do not aim to investigate how sociological theories are used in ICT. We assume that although IT is an efficient tool for sociological studies, the two fields are based upon different axioms or paradigms. This goal leads to the mechanization through AI techniques of sociology. Mechanization of reasoning is achieved in Mathematics for instance because there is a well-defined theoretical contents of the field. This is however not true in Sociology and leads to a different approach. The link between sociology and social networks looks pretty obvious these days. However, the functionalities of a social network (SN) are set in a component that is usually called the community engine. Even when users are given access to this module, they cannot modify the functionalities (because of business reasons although technically simple) of the society represented by a SN since functionalities usually cover the action of individuals within a society. There are also attempts in social informatics to investigate applications of sociological theories. We wish to investigate whether Sociology may have an impact on ICT. As in several IT applications, we design a formal model with relevant specifications. We have to make a few decisions before formalizing our approach. A first decision is to concentrate our goals to the actions performed by the actors or agents of a community. This shapes the society to which agents belong. This means that we go back to the main basic works of Sociology due to Max Weber [Weber] and Emile Durkheim [Durkheim] and do omit the numerous pieces of sociological works generated by these founding fathers' works. As usual, we assume that the actions of the agents of a community define the society for these agents not their political system. To define the latter depending on the degree of freedom and autonomy allocated to the choice of the governing body would be possible to model within this abstract approach. But, we distinguish between Sociology and Political Science. We have selected the founding work of Weber stating that a society is the result of the actions of individuals. To select the Weber's approach is due to a variety of purposes and goals. A first one is to master the evolution of systems. This evolution is the result of the actions performed by the agents. This is easier according to Weber than when relying on Durkheim. A second remark is that Weber's is suitable to perform a mathematical description of economy as shown by Ekeland [Ekeland]. Another comment is that the results of actions can be assessed through utilities functions as already demonstrated by Von Neumann and Morgestern [Neumann]. A purely mathematical specification of sociology is however not possible since the latter works show that parts of the theory of functions are still missing for such a goal. A different comment is that some famous attempts, such as the one of Castells [Castells], to intro1

duce methodologies of ICT into sociology have been proposed. We do not select them because they usually dress up existing concepts of sociology by ad-hoc features of IT. We want to investigate a more general and formal methodology. An important factor is that we agree with the opinion of the mathematician Morris Kline [Kline] who wrote that Mathematics is at the origin of the Western culture. Several other cultures have expressed a similar point of view as pointed out by Pierre Cartier, another famous mathematician, in his invited talk at the CICM conference in Paris in 2010 "Can we make Mathematics universal as well as fully reliable". A quotation from Kline's book is as follows: "Mathematics has determined the direction and content of much philosophic thought, has destroyed and rebuilt religious doctrines, has supplied substance to economic and political theories, has fashioned major painting, musical, architectural, and literary styles, has fathered our logic, and has furnished the best answers we have to fundamental questions about the nature of man and his universe". The consequences of the preceding remarks are that we will have to design methods to mechanize sociological reasoning that are as close as possible to the mechanization of mathematics. We know that Artificial Intelligence is a framework where such a mechanization procedure can be designed. For feasibility reasons, we investigate a bottom-up approach to this challenge. It is indeed safe to assume that in most applications based upon IT we have highly complex systems that are set up to mechanize some sort of decision making. A top-down approach would lead to system impossible to manage efficiently in ICT. We now provide a first sketch of what a specification such as those outlined above in very general terms may look like. In [Racam] a concept for agent-oriented abstraction was proposed. It requires a concept of society of agents defined as follows: A society of agents is the societal organization arising from the actions performed by individual agents in the agent world assigned to a problem. The main motivation for this abstraction mechanism was to have a continuous translation from a single agent system to a society of agents. Usual approaches for the concept of agent societies are based either on a too large number of agents or on a fixed description of the relationship between agents and their environment. But, they are usually disconnected from what sociologists call "a society". Agent-oriented abstraction was inspired by the object-oriented programming paradigm but with additional cognitive goals. A next step is to notice that this agent abstraction can be expressed through the design of ABIT, an abstraction-based information technology [Calmet 2009]. ABIT is directly inspired by the mechanization of reasoning as used in the domains of automatic theorem proving or computer algebra. It is inspired but not generalized since the methodologies designed in theorem proving or algebra cannot be extended to other domain of mathematics. A track is thus to model cognitive knowledge as we would do of Mathematics. To achieve such a goal we need to define a concept of type for knowledge (different from ontology) and then the properties associated to a specific type of knowledge. [Calmet 2009] introduces the basic concepts but they can be refined for our purposes here. The link to Sociology ought to be now easier to identify. Our framework is set within Weber's approach to what a society is. Then, we can specify as types and properties the relevant facts and knowledge related to a given model within this framework. This provides the necessary tools to simulate several possible sociological models of societies. For instance, we can assume that we have then suitable tools to investigate a multi-cultural society for which the types are defined as different cultures (involving various languages or religions for instance) with specific properties such as the relative importance or the link to today accepted common denominators such as "broken English". The type and properties can be closer than expected. For instance, in a plant part of the control is performed either by machines or by software. The society is then the plant while the types of knowledge are hardware or software control and the properties are the characteristics of the respective hardware 2

and software equipment. Indeed, such a plant is a single society with different culture and knowledge and obvious communication difficulties. The proposed chapter will detail these models. It must be noted in addition that two features not always incorporated into sociology are not fully required but at least useful. A first one is that since we deal with knowledge and we do not wish to design knowledge warehouses leading to complex and inefficient systems, it is very convenient to rely on virtual knowledge communities allowing a bottom-up and thus manageable design of models. A second feature that is becoming a requirement in nowadays applications is to enforce trust in what systems are doing. The former concept of virtual components will be described. The latter is being investigated at present. It is indeed mandatory to assess the trust one can have in actions performed by agents of any society. The result is definitively a mechanization of selected aspects of sociology as seen by computer scientists and mathematicians. This works focuses on technical features not on foundation concepts of Sociology. References [Weber] Max Weber, "L’éthique protestante et l’esprit du capitalisme", original in German 1904, French translation, Plon, 1964. [Durkheim] Emile Durkheim, "Le suicide", 1897. New edition by Plon, 1964. [Neumann] John von Neumann, Oskar Morgenstern, Theory of games and economic behavior, Princeton Univ. Press, 1944 [Ekeland] Ivar Ekeland, "La modélisation mathématique en économie", Gazette vol. 78, pp. 51-62, SMF, 1998 [Castells] M. Castells, "The Information Age trilogy", Blackwell Pub. 1996, 1997 and 1998. [Kline] Morris Kline, “Mathematics in Western Culture” Oxford Univ. Press, 1953 [Racam] J. Calmet, P. Maret, R. Endsuleit, "Agent-Oriented Abstraction", Rev. R. Acad. Cien. Serie A. Mat. Vol. 98 (1), pp. 1–7, 2004 [Calmet 2009] J. Calmet, "A Framework for Open Mechanized Reasoning". Invited talk at Calculemus 2009. In “Conference on Intelligent Computer Mathematics” (CICM). Eds. J. Carette et al., Springer LNCS 5625.

3