Article Preview
TopIntroduction
In previously published papers (Kopanov, Tchalakov, Keskinova, 2010) we formulated the main principles in modelling the behavior a specific type of actor-networks we called ‘stacked actor-networks’ (SAN). There one and the same actors are simultaneously present in several actor-networks under different ‘identities’. Studying given economy, for example, we can find the same companies and organizations interacting in a set of corporate interlocking directorates, and then a being shared between the same set owners, and also located in close geographical proximity, etc., thus “…each company serving as a node that allow you to make sense of how all these networks are stacked together” (Muniesa & Tchalakov, 2012, p. 15).
The stacked network model aims at describing the processes and interactions in this specific type of actor-networks and possibly at predicting (or rather ‘understanding’, at least qualitatively) their behavior. It also allows accounting for the heterogeneity of actor-networks, using specific stochastic automata to model not only human individual or corporate agents, but also technologies, artefacts, money, and so forth.
This paper continues the efforts presented in the two papers cited above, focusing on the key problem of how to account for different identities of one and the same actor in the different SANs, providing an initial outline of the necessary steps towards its formalization. We will begin with the brief discussion about the very possibility to build a formal model that mimics sufficiently enough the behavior of a real actor-network. We mean formal (mathematical) modelling of SAN which, with given inputs, generates output data that correspond to those generated by real-world agents in an actual actor-network. There are two options in this modelling: first, if actual outputs are determined, the models must also be determined and match the real ones; and second - if the actual outputs are stochastic, those in the model must also be stochastic, so that the initial distributions of the real agents and the models should be the same in the framework of the statistical test.
It is important to note that we are aiming at a model that rather mimic SANs functionally, and not so much as structure – in fact we mean specific realization of Turing's famous test proving the presence of intelligence.1 In it, the examiner carries out a natural language conversation with another person and with the machine being tested (both man and the machine are presented as people), asking them questions and receiving answers without having direct contact with them. If during the conversation tester cannot determine with certainty which of his interlocutors is a machine and who is a man, then it must be admitted that the machine imparts intelligence to a person (whatever that may be).
From what has been said, it follows that the answer to the question posed at the beginning about the possibility of building a formal model of agent behavior in the network is positive not only in principle. These specific examples show that creating an adequate model of a network of human and non-human actors is a difficult but realistic goal accessible to modern standard computer systems.
These considerations allow us to modify the above-mentioned goal of the study in the following way: modeling network operators in such a way that an outside observer cannot distinguish their “playing” behavior from that of any real “players”.