Inhibiting Emergence in and Destabilizing Multi-Agent Networks
Kathleen M. Carley
kathleen.carley@cmu.edu

     Standard social network measures are insufficient to address the issue of network change as they focus only on the social network ˆ who knows/talks to who. Here a meta-network approach is proposed that links agents, knowledge and tasks.  Processes that alter one network have a cascade effect changing the entire meta-network.  Within this meta-network some agents play more critical roles than others ˆ such as the agent with high degree centrality (who is most connected) and the agent with high cognitive load (who has the most cognitive processing due to talking to others, handling information, working on tasks.
                              
                                    People                    Knowledge                      Tasks
People                         social network        knowledge network          assignment network
Knowledge                                                information network         needs network
Tasks                                                                                                  precedence network

     Using multi-agent technology it is possible to model and examine the network dynamics and the factors affecting emergence and stability.  In fact, multi-agent network models are generally useful for examining complex systems with policy ramifications (Carley, 2001). Most research that use multi-agent techniques have focused on the emergence of new behaviors or group level phenomena.  In this paper, the question is reversed.  Instead, it is asked, what can be done to inhibit emergence and to destabilize multi-agent networks.  This study illustrates some of the difficulties in destabilizing multi-agent networks. Such an illustration is particularly salient in lieu of the tragic events of September 11, 2001 and the consequent issues relating to potential weaponized biological attacks such as anthrax.CONSTRUCT-O (Carley & Hill, 2001) is a multi-agent network model in which the social and knowledge networks co-evolve over time and effect the performance of individuals and the group.  Issues of information diffusion, network
change, organizational design, impact of new information-technology, and proximity can be examined using this model.  The predecessor of this model, CONSTRUCT, was used to examine the factors enabling group stability (Carley, 1991) and the evolution of networks (Carley, 1999).  Within CONSTRUCT-O the agents differ in terms of their socio-demographic characteristics (such as age, gender, education) and by their knowledge and beliefs.  Individuals forget and interact if they are available for interaction and are motivated to do so.  There are two basic motivations ˆ relative similarity and relative expertise ˆ both of which are basic to human nature.  Relative similarity is the tendency of people to choose to interact with those who are more similar. Relative expertise is the tendency of people to seek out new information from those whom they perceive to be more expert.  When people interact they learn and when they learn that knowledge changes who they view as most relatively similar or expert; thus, causing the social network to change
     Using CONSTRUCT-O a series of virtual experiments are conducted.  The type of network, the number of agents that are isolated, the order of isolation, and the rational for isolation are explored. Impacts are examined on organizational performance and the diffusion of new information. Note, we focus on the isolation of agents, i.e., node extraction, as earlier studies indicate that the performance impact on a network is highest for personnel changes.  We isolate agents based on three rationals that require varying level of information about the network: random, degree centrality, and the emergent leader (agent with highest cognitive load).  Results indicate that the nature of the network and the goal vis-a-vie inhibiting emergence impacts the choice of destabilization strategy. Moreover, results indicate that destabilization
may have unintended consequence; e.g., isolating agents may actually in a cellular network improve the rate of information exchange initially.  

References

Carley, Kathleen M.  2001.  "Computational Organization Science:  A New Frontier."  Paper prepared for the NAS sponsored Sackler Colloquium, Irvine CA, October 2001.

Carley, Kathleen M. 1999. "On the Evolution of Social and Organizational Networks." In Steven B.  Andrews and David Knoke (Eds.) Vol. 16 special issue of Research in the Sociology of Organizations. on "Networks In and Around Organizations."  JAI Press, Inc. Stamford, CT, pp. 3-30.

Carley, Kathleen, 1991. "A Theory of Group Stability." American Sociological Review , 56(3): 331-354.

Carley, Kathleen M.  & Vanessa Hill, 2001, "Structural Change and Learning Within Organizations".  In Dynamics of Organizations:  Computational Modeling and Organizational Theories. Edited by Alessandro Lomi and Erik R. Larsen, MIT Press/AAAI Press/Live Oak, Ch. 2. pp 63-92.

Carley, Kathleen M., Lee, Ju-Sung and David Krackhardt, 2001. Destabilizing Networks. Connections 24(3): 79-92.