Towards a Coordination Cookbook: Recipes for Multi-Agent Action

Keywords:

Abstract:

This thesis presents the first steps towards a theory of coordination in the form of what I call a coordination cookbook. My goal in this research is hypothesis generation rather than hypothesis testing: I attempt to develop a theory of coordination grounded in detailed empirical observation. I am especially interested in using this theory to identify ways of coordinating that may become more desirable when information technology is used to perform some of the coordination.
I address the following question: how can we represent what people do to coordinate their actions when they work together on common goals, in a way that reveals alternative approaches to achieving those goals? To answer this question, I study groups of people making engineering changes to complex products as an example of a coordination-intensive task. I perform detailed case studies of the change process in three organizations: an automobile manufacturer, a commercial aircraft manufacturer and a computer system software developer.
To analyze these cases, I develop a technique for describing the behaviour of the members of an organization, based on research in distributed artificial intelligence (DAI). I first develop a data-flow model of the change process to identify what information was used and how it was processed by the different members of the organization. Then, using ideas from DAI, I model what each individual must have known about the task and the rest of the organization to act as observed.
To develop a theory of coordination, I generalize from these specific individuals to the kinds of tasks they performed. I develop a typology of interdependencies between organizational tasks and objects in the world (including resources and products). This typology includes four categories of coordination needs, due to interdependencies between: (1) different tasks, (2) tasks and subtasks, (3) tasks and objects in the world and (4) different objects.
I then re-examine the cases to identify the coordination methods used to address these needs. (These coordination methods are similar in spirit to the weak problem solving methods of cognitive science.) I represent each method by a set of what I call coordination recipes that identify the goals, capabilities and knowledge of the individuals involved. In some cases, consideration of the possible distributions of these elements suggests approaches other than those actually observed. This framework allows an analyst to abstract from a description of how a particular organization performs a task to a description of the coordination needs of that task and a set of alternative coordination methods that could be used to address those needs.
The results of my thesis should be useful in several ways. A better understanding of how individuals work together may provide a more principled approach for designing new computer applications, for analyzing the way organizations are currently coordinated and for explaining perceived problems with existing approaches to coordination. By systematically exploring the space of possible coordination strategies, we may be able to discover new kinds of organizations—organizations in which humans and computers work together in as yet unimagined ways.