p.6 An area of AI research which the current work resembles is the speech understanding system, Hearsay-II (Erman, 1975), specifically with respect to the issues of knowledge integration and focus of attention (Hayes-Roth 1976). In Hearsay-II the central task is to build a sentence hypothesis which is a best explanation of the given speech input data. An "iterative guess-building" process takes place, in which a number of different knowledge sources (facts, algorithms, heuristics), operating on various descriptions of the hypothesis, must cooperate. In order to use the knowledge sources efficiently a global data base - the "blackboard" - is constructed which contains the currently active hypothesis elements, at all levels of description. The decision to activate a particular knowledge source is not fixed, but depends at any point on what has thus far been established and what available knowledge source is most likely to make further progress.

p.24-25 There are several choices of control structure faced by the designer of a knowledge-based system. Basically the choices are among points on a spectrum, at the extremes of which are goal-driven and event-driven systems. In a goal-driven system... the rule interpreter selects a rule which concludes with the goal being sought... An alternate way to focus attention is to employ an event-driven control structure. In this scheme the current state of the hypothesis space determines what to do next. The monitor continually refers to a list of current events - the event-lists mentioned in the rules discussed above - which is used to trigger those knowledge sources most likely to make further headway. As a knowledge source makes a change in the current hypothesis, it also places a symbol on the event-list to signify the type of change made. Thus as events are drawn from the event-list for processing, new events are added, so that under normal conditions the monitor always has a means for choosing its next move.

 The system we are currently developing operates in both goal-driven and event-driven modes, with an emphasis on the latter. The normal iterative cycle of problem solving uses the event-list to trigger knowledge sources, which create or change hypothesis elements and place new events on the event-lists. Thus the system's behavior is "opportunistic" in that it is guided primarily by what was most recently discovered, rather than by a necessity to satisfy sub-goals.