Social Coordination: Principles, Artefacts and Theories (SOCIAL.PATH)

Coordination is a basic feature of human interaction. Indeed, failure to achieve coordination might expose you ridicule, censure, or concerns about your mental capacity. On the other hand, human coordination is fragile, difficult to achieve, and often short-lived. At first sight, it may seem an easy task to engineer coordination in artificial societies to match that found in human ones. But, as I hope to show in this talk, understanding and modelling coordination is much more difficult than it seems. There are some fundamental problems in identifying coordination that computational systems tend at the moment to sidestep. I start from a definition of coordination that it is ‘the state of working together’ (definition derived from http://en.wiktionary.org/wiki/ coordination) The first problem is to identify instances of coordination. Simply observing the behaviours of two actors that take place simultaneously is not sufficient: we don’t know from observation of physical events that there is coordination involved. The behaviours may be the same, but performed independently and even without knowledge of the other. Or the behaviours may be different, yet coordinated. The second problem is that even when we are told or come to understand what is going on, we still don’t know whether the actions are really coordinated, especially if the intended coordination fails. This is because humans, unlike some computational agents, are not able to look inside the cognitions of their colleagues. The third problem is that coordination may be on the basis of cooperation (the case we usually think of), or competition. The Prisoner’s Dilemma is a good case of where it is not so clear whether coordination exists, in a context of cooperation and competition. The fourth problem is that it may seem that coordination is best identified by asking the actors involved, but they face much the same kinds of difficulty as observers do in attributing intentionality to their allegedly coordinating partners or even to their own behaviours. The fifth problem is that ‘working together’ implies some rules (or norms) about what to do, yet there are difficulties in applying rules to action that make the idea of ‘rule-governed action’ deeply problematic. In this talk, I shall illustrate these problems with a real example, and begin to suggest what might be involved in designing a computational model that respects the limitations faced by human actors in attempting to achieve coordination.