This paper introduces a multi-agent-based simulation toolkit (called SOTCAC) that uses autonomous, intelligent agents to represent the components of coevolving terrorist network (TNet) and counterterrorist network (CTNet).
The model (currently in development) is predicated on the proposition that adaptive agents can be used to describe the self-organised, emergent behaviour of TNets – conceived as complex adaptive systems – on three interrelated dynamical levels: (1) dynamics on networks, in which notional terrorist agents process and interpret information, search and acquire resources and adapt to other agents' actions; (2) dynamics of networks, in which the TNet itself is a fully dynamic, adaptive entity and whose agents build, maintain and modify the network's local (and therefore, collectively, its global) topology and (3) dynamics between networks, in which the TNet and CTNet mutually coevolve.
The TNet's ‘goal’ is to achieve the critical infrastructure (of manpower, weapons, financial resources and logistics) required to strike, while the CTNet's mission is to prevent the TNet from doing so.
The ultimate goal for developing SOTCAC is to provide INTEL analysts with a larger context of ‘plausibly possible’ TNet ↔ CTNet coevolutions – as they could be – so that TNets can be better understood as they are and as they are likely to adapt. And since the way in which the CTNet collects, assimilates, fuses and derives inferences about what the terrorist ‘looks like’ is prescribed (via local, context-dependent properties and behaviours) – not scripted – SOTCAC may also help analysts discover novel data-fusion and ‘ground truth’ inference strategies.
[…]
# Fundamental Problem
Figure 1 illustrates, schematically, the mathematically distilled form of the fundamental problem that SOTCAC is being designed to address.
We first introduce some basic terminology. A Graph, G, is a set of abstract objects (nodes) with connections between them (links). Nodes can be simple – for example, they can be used to merely label the parts of a system; or complex, possessing dynamic internal states that change as a function of the states of other nodes in their local neighbourhood (and possibly even consisting of nested networks that evolve inside of them).
Likewise, links can be simple abstract lines that represent a basic connection between two nodes; or they can be more complex, and represent different forms of information flow, have directionality (with arrows to represent flow into and out of agiven node) and/or consist of multiple simultaneous edges.
G itself may, in general, be used to represent any of the wide class of complex dynamical systems that perform some function F. In the context of this paper, of course, we are using G as a mathematical model of the agents and communication links that make up a hypothetical TNet. Furthermore, we assume that this network is not isolated, either in space or in time, and that it exists within a more general environment ( ＝ EG), which surrounds, interpenetrates and interacts with G. (EG may consist, in part, of other, interrelated, networks). We assume that G performs some function (or, more generally, a set of functions), labelled, generically, by the symbol F. In SOTCAC, F includes the various recruiting, training, weapon acquisition, resource gathering and allocation and cell-coordination tasks needed for a TNet to attain a ‘mission ready’ state.
Now, consider the set of all unweighted, undirected labelled graphs, VG; that is, graphs containing only symmetric unit-valued links between originating and terminating nodes, where each node is assigned a unique label. Implicit in G’s membership in VG is that some subset of VG best performs F, where ‘best’ is understood to be modulo some efficiency measure – labelled eG ＝ eG(EG,F) in Figure 1 – of how well a given graph performs F in the context of environment EG. For example, we expect a minimally efficient G to be one with very few (individually efficient) nodes and few links (or none at all). A highly efficient G might contain a moderate number of ‘just the right kinds of’ nodes with ‘just the right kind and numbers of’ links among them necessary to perform a given function. The fundamental problem, of course, is to make precise the otherwise vague phrase, ‘just the right kind of ...’
As a start, we might ask: ‘What set of primitive local and/or global features of G (labelled as {fi(G;E,F)} in Figure 1) are most relevant to describing G behaviour?’ followed by ‘What set of values of {fi} maximise G’s efficiency at performing function F?’ It is not a priori obvious which (if any) of the conventional suite of social network metrics – Degree, Centrality, Betweenness, Closeness, etc. [36] – is best suited for a given context, as the network evolves and adapts to whatever actions are carried out against selected targets. Conceptually, at least, part of SOTCAC’s charter is to allow analysts to explore the veracity of alternative ‘feature sets’ and metrics that describe dynamic networks.
The more general problem, shown schematically on the right side of Figure 1, addresses how G’s behaviour changes, and in what ways, if its ambient environment is not entirely innate, but consists in part of a separate system, S, that deliberately intervenes in G’s activity (in SOTCAC, S , counterterrorist force) in hopes of preventing G from performing its function: EG ! E 0 ¼ EG þ S. The presence and impact of S are modelled as a set of interventions (or actions), {ai}, that S implements on G. The fundamental problem is to find a policy of actions, ^ pS ; ð ~ a1; ~ a2; ...; ~ at; ...Þ, where ~ at ; ðat 1; at 2; ...; at nÞ,suchthatwhen ^ pS is applied to G, the value of eGðEG þ S; FÞ is minimised. The major objective – from S’s point of view (and from the analyst’s point of view when using SOTCAC) – is to find a policy for which that eGðEG þ S; FÞ ! eGðEG; FÞ.
[…]
~
ILACHINSKI, Andrew, 2012. Modelling insurgent and terrorist networks as self-organised complex adaptive systems. International Journal of Parallel, Emergent and Distributed Systems. 2012. Vol. 27, no. 1, p. 45–77.