1 RESILIENCE AND NETWORKS Jan

The purpose of this paper is to apply network science to the field of resilience engineering. Starting with the trade-off between prepared versus deliberated knowledge, I argue that socio-technical systems are first and foremost networked systems that need to connect modules of prepared knowledge or instances of deliberated knowledge by means of protocols. I hypothesize that particular frameworks for organizing protocols, or ‘architectures’, are more resilient than others. In network science terms, protocols are interaction patterns that may lead to sustained adaptability in the face of unexpected events. Research in a variety of domains has shown that scale-free network structures, with a power-law degree distribution, have the highest resilience. The relevance of this finding for social networks and for the concept of resilience as sustained adaptability remains to be demonstrated. It is clear, however, that social network analysis in particular, as a novel research methodology in this field, offers a more quantitative base to establish resilience engineering research upon. 1 THE PREPARATION VERSUS DELIBERATION TRADE-OFF One of the main issues within the field of resilience engineering is the question how systems deal with surprise events. More generally, systems are able to perform under a variety of conditions by drawing upon a mix of prepared knowledge or deliberated knowledge (Newell, 1990). Prepared knowledge facilitates the recognition of familiar events and results in efficient, robust, and rule-based performance. Deliberated knowledge is required when systems are confronted with surprise events. In this case, the emphasis is on thoroughness and analysis of multiple options. The variety of situations leads to a specific mix of deliberation and preparation, and the architecture of a specific system determines the extent to which a response stems from deliberation or preparation. Humans, for instance, generally rely on prepared knowledge, as demonstrated by the finding that at least 80% of their decisions are recognition-primed rather than analytical. The human cognitive architecture, as well as the externally fixed time to respond, simply do not allow for much deliberation. Intelligent computer systems have different architectures and are able to engage in extensive search, comparing millions of situations per task, but generally have little prepared knowledge to bring to bear. There is thus a preparation versus deliberation trade-off that has been known for some time for intelligent systems (see Newell, 1990) and is depicted below: Figure 1 Preparation versus deliberation trade-off (after Newell, 1990)