ABSTRACT
Since its origins, the field of Multi-Agent Systems (MAS) has been entrenched with the notion of direct communication [Hewitt, 1977; Smith, 1980; Bond and Gasser, 1988; Durfee, 1988].
In particular, the shift from blackboard systems [Engelmore and Morgan, 1988], as protoMAS, to proper MAS was characterized by abandoning a form of indirect coordination in
favor of a more direct strategy. The ‘agents’ in a blackboard system used to leave their partial solutions to a problem in a data structure shared with other ‘agents’ and each reacted to the product of the work of other agents by producing new work (i.e. new partial solutions). At the time, a single thread of control in writing in the blackboard created a bottleneck in problem solving. This constraint was overcome, at the beginning, by adopting direct communication between multiple blackboard systems (via message passing) and then, eventually, by dropping the indirect approach altogether. Though a new seminal approach was born, the baby, or at least a useful toy of his, was thrown out with the bath water.
