79Published by the IEEE Computer Society Rather, flexecution entails changing the goals themselves based on discoveries made during execution. In pursuing illdefined goals, we must expect to revise and even replace the goals we initially stated during the planning phase.One of my objectives in writing these essays is to describe the importance of goal discovery to planners and project managers who might not appreciate the difficulties posed by ill-defined goals. A second objective is to offer suggestions to computer scientists who design mixedinitiative systems about how to better support planning and execution with ill-defined goals.We'll begin with the notion of planning to set the stage for comparing planning using clear goals versus planning and execution using emergent goals.
PlanningPlans are prescriptions or roadmaps for procedures that can be followed to reach some goal, with perhaps some modification based on monitoring outcomes. Planning is generally considered the process of choosing and organizing courses of action on the basis of assumptions about what will happen in the future. 1 Plans describe contingencies and interdependencies such as actions that must occur first as a precondition for later actions.The more clear and detailed your goals, the easier it is to construct a plan, prepare a timeline, and gauge progress. Clear goals can be sufficiently specified to enable planners to identify tasks that, if followed, will reach the intended goal. A plan specifies a set of actions with which you intend to transform a current situation into a goal state. This type of planning is essentially problem solving-finding a way to take some actions that will result in the transformation, as represented in figure 1a.Planning-as-problem-solving generally relies on a set of well-understood representations and mechanisms, such as planning graphs, task networks, search in a state space, chaining, type hierarchies, and constraints on variables. [2][3][4] These mechanisms and approaches seem to run into difficulty when faced with the intractability of large, messy, real-world problems. For such planning problems, the state spaces aren't completely predefined, and planning requires human decision making based on knowledge and sensitivity to context. [5][6][7] The field of intelligent planning systems has reached into new areas in recent decades, such as distributed planning and continuous planning 8 and mixed-initiative planning or "planning assistants." 9 My proposal is consistent with these developments, especially the notion of planning as a dialogue among human and machine agents. 10
Ill-defined goals and wicked problemsThe challenge of ill-defined goals arises when many or most of the goals' features can't be specified in advance, as illustrated in figure 1b. Herbert Simon and other researchers have written about the differences between wellstructured and ill-structured problems. 11 Here, I focus on the clarity of goals, which is just one dimension of the problem structure.We can represent well-defined goals using ...