In their everyday lives people flexibly handle many tasks. They work
on one at a time, make quick decisions about which one to work on, and
put aside tasks whenever attending to them is not required to achieve
the task's goal. This last capability is critical because rather than
fixate on a blocked task, a person can work on some other task. For
example, a person making bean soup wouldn't watch the beans as they
soak overnight. Instead she would take the inability to affect the
progress of making bean soup as an opportunity to work on some other,
possibly less important task where progress is possible. When taking
advantage of these opportunities, people don't completely forget what
they were doing. Instead, the put-aside tasks guide the selection of
the new tasks. After the beans have soaked and the soup is simmering,
a person might go out to get other chores done, but she probably would
not forget the soup and eat out.  By putting aside a blocked task and
remembering it at the appropriate time, a person can complete many
other tasks and still accomplish the blocked one in a timely manner.

My thesis work describes how an artificial agent, called Laureli, can
maintain goal-directed behavior while suspending her blocked tasks
similarly to that described above. Laureli serves as an exemplar
agent, grounding in detailed and understandable examples the problems
that occur when she suspends her tasks for later reactivation.
Laureli's suspension-reactivation mechanisms provide for interleaving
more available tasks during the task's {\it slack} time (while the
time the task is blocked).  A task's {\it availability} is defined as
how likely the agent expects that working on the task will make progress
toward the task's goal.  A task's availability changes over time, and
depends on both the agent's actions and input from the environment.
Laureli's suspension-reactivation mechanisms are her method for
representing large changes in a task's availability over time.

Representing a task's availability to the agent is important, because
the agent can then better schedule its tasks as it executes them. If
the task's availability over time is known in advance, then the agent
can use that knowledge and generate a schedule that can be simply
followed at execution time. However, in many cases the agent either
doesn't know or it is difficult to know how the task's availability
will change over time. The second step in making bean soup, "boiling
the beans for an hour", is such an example. Perhaps Laureli could
measure the water, look up the specific heat of the beans and then
using the effective heat transfer from the stove, calculate how long
until boiling. However, she could also just put the pan of beans and
water on the stove, stay in the area, and occasionally check to see if
the water was boiling. My thesis advocates this second approach, so
the agent can monitor a task's execution rather than attempt to
schedule the task, when the agent doesn't have all the availability
knowledge. 

Laureli's suspending and reactivating of her tasks is similar to
deliberately excluding the suspended tasks from her decisions about
what action to take.  However, some decisions based on the smaller
number of tasks will be functionally different from similar decisions
that consider all the tasks.  Since these decisions affect the agent's
external behavior, suspending tasks can affect the agent's apparent
{\it rationality} in achieving its goals.  In many cases, Laureli can
act similarly to an agent with access to all its tasks, because she
can access all the tasks that affect the decision. Conflicting tasks
are accessable because when Laureli suspends a task she builds special
conflict-detecting reactivation rules. These rules detect potentially
selectable tasks that will conflict if selected with the suspended
task. These rules monitor Laureli's choices, pointing out
conflicts. At some level of detail these conflicts would be the same
conflicts that would be considered, if she were scheduling these tasks
before execution. Similar rules are expected to make synergistic tasks
accessable.  In the cases where Laureli doesn't have access to all her
tasks, I am looking at mechanisms that might enable her behavior more
closely to approximate her behavior with access to all her tasks.  In
addition, I am also investigating when suspension is worth the extra
effort of the special mechanisms.