I think
that it is generally agreed that planning is a good exercise for any continuing activity. The only question
is how to do this in a computer chess program, or in any intelligent activity, in a way that improves performance:
p.xi"To
be able to plan, one needs intelligence. To act intelligently, however, one needs to plan. The questions that this paradox
raises express the goal of this book. The abilities to anticipate and plan are essential features of intelligent system, whether
they are human or machine. We might go further and contemplate that a better planning results in higher achievements. As a
consequence, understanding and improving planning is important."
Planning is something we humans seem to do with ease. How do we put down in words, or even better
in an algorithm, how we perform the essential task of planning?
p.1"No living thing seems to be conscious
of the future, and none seems concerned to design for that future, except Man. But every man looks ahead and attempts to organize
for tomorrow, the future of the next day or of the next generation... By his very nature every man plans constantly."
Predictive models and anticipation are important aspects to planning.
p.2"As a starting point, we presume that all intelligent systems use anticipation to plan (van Wezel
and Jorna, 2001). An anticipatory system is 'a system containing a predictive model of itself and/or its environment, which
allows it to change state at an instant in accordance with the model's predictions pertaining to a later instant' (Rosen,
1985)."
We now look at the basic elements of planning. There appear to be 3 main elements. Note that
here we bring up a model of the future. This involves insight, anticipation and information. Note that our plan cannot be
specified in more detail than the model of our future allows. If we have a general model of the future, then our plans must
by nature be general, or at least flexible.
p.3"three main elements of planning.
First, it is important to acknowledge
that some entity must make the plan... The planning entity needs some kind of model of the future... Planning and anticipation
presume that such a predictive model is available... Information must be collected..
Second, someone or something must execute
the plan
The third element of planning is the
plan itself... It can never be a full specification of the future itself because it can never be specified more precisely
than the model of the future allows."
There is a performance tradeoff between being manageable and being able to succeed at accomplishing
goals. Plans that are too complex or detailed might fail to execute, but simplistic plans might not accomplish much.
p.4"Somehow, intelligent systems know
how to make planning simple enough to be manageable but complex enough to attain advantageous goals."
Planning involves making decisions about the future. This involves modeling the future and making
(possibly) uncertain predictions in our model. We might wish to avoid this side effect to planning since it involves uncertainty,
and this might make us feel uncomfortable. However, we simply have to do the best we can about predicting the future so that
we can plan for it. Perhaps this task should be entrusted to the most senior member of an organization or to the person who
has worked the most number of jobs or talked with the most diverse number of people.
p.7"First, planning is a synthetic rather
than an analytic (diagnosis) or modification (repair) task. Second, planning involves decisions about the future and not the
execution of these decisions. Third, an important feature of planning is that it is about choosing one alternative out of
a huge number of of alternatives that are structurally similar."
If we cannot precisely predict the future, it seems that we must use probability to generate
a forecast for the future, including our uncertainty. We might not know what our opponent will do in a game, but
we can focus on likely moves and the consequences of those likely moves. Moves deemed unlikely we will not spend too much
time looking at.
p.17"Our starting point in this book
is that planning is always in essence about the same thing: anticipating on the future [sic] and determining courses of action."
It is not "to plan", or even the plan itself that is our ultimate objective. It is the desire
to reach the goal that is the foundation for the plan. Planning, in simple terms, is the behavior of people with a strong
desire to win an objective. The short term goal of planning is to allocate resources so that one is properly positioned
for the problems that will inevitably arrive from the tomorrow that one day will be a today.
p.62"In fact, planning is
no longer an end in itself, but rather a means to reach supervision and control goals... a plan does not include
all the details needed for its implementation... a plan always integrates a projection in the future"
A plan involves a forecast that has a reasonable chance of being accurate. If you cannot forecast
the future, or cannot forecast your uncertainty about the future, then you should not expect any degree of success for plans.
p.65"At the symbolic level, anticipation
is elaborating a representation of the future (forcasting)... expert process operators have more planning ability than beginners
because they anticipate more"
Certain events cannot be predicted and must be handled on an event-driven basis. The limits of planning
are best seen when they begin to take resources away from managing problems in the here and now.
p.96"Anticipation is costly and limited
by time constraints. It can be a handicap when it goes beyond actual competency, so that reactive strategies are necessary.
The focus on planning strategies should not let us forget the adaptive power of reactive strategies, especially when time
constraints are high or process knowledge is weak. The main problem to solve is finding an efficient compromise between
anticipative and reactive strategies to reach a satisfying performance level."
We are trying to transform a given situation into a desired
situation. We begin by recognizing the discrepancy that exists between the two situations, identifying our goals and our resources,
identifying the capabilities of our resources, locate our forecast of the future, and then go to work setting up a set of
sub-goals that will move us closer to our desired goal.
p.109"In
the Encyclopedia of Cognitive Science a problem is defined as 'transforming a given situation into a desired situation
or goal' (Wilson and Keil, 1999, p.674). Problem solving is the recognition of a discrepancy between an existing and desired
situation for which no automated series of actions exist to resolve that discrepancy. Otherwise, it is an algorithm... decision
making is the process of choosing a preferred option or course of action from among a set of alternatives"
Planning is a kind of problem solving where the focus is the future.
p.110"The common sense idea is that planning,
problem solving, and decision making are overlapping. They are not the same, but they have various activities, starting points,
and aspects in common."
Planners use simple techniques, often back-of-the-envelope counting, to set the sub goals and estimate
best use for resource allocation.
p.118-119"Mietus (1994) showed that planning
as an integrated task consists of various subtasks such as counting, negotiating, evaluating, and problem solving itself.
She found that in some cases simple counting takes more than 15% of the time to solve the planning problem... Our expectation...
is that planners will use simple schemes without much quantification. This is in line with many remarks from practice saying
that planners normally make their plans and schedules on the back of an envelope."
Planning as a cognitive ability must somehow be produced in a machine if the machine is
to exhibit intelligent behavior. We recognize intelligent behavior by the struggle to achieve goals and the allocation of
resources for that meaningful end.
p.177"Planning is a basic cognitive
ability that humans use to organize their life and to achieve objectives."
Intelligent activity can be simplified and described as a procedure involving grouping, focusing
attention, and searching.
p.336"The abbreviation GFACS is deciphered
as 'Grouping, Focusing Attention, and Combinatorial Search'.. Most of the elementary procedures that are being applied for...
other intelligent activities are, in turn, based on GFACS set of procedures."
Intelligence transforms information into successful system functioning in a (possibly) competitive
environment. It is hard to be successful without the ability to 1. observe the present situation 2. predict the future 3.
set meaningful goals and 4. construct plans for achieving goals.
p.340"Intelligence is a property of the
system that emerges when the procedures of direct and inverse generalization (including focusing attention, combinatorial
search, and grouping) transform the available information in order to produce the process of successful system functioning...
the main advantage of intelligence is the ability to deal with unexpected predicaments."
Planning according to Newell and Simon:
p.503"Newell and Simon assume the following
steps in planning: (1) abstracting by omitting certain details of the original objects and operators, (2) forming the corresponding
problem in the abstract problem space, (3) when the abstract problem has been solved, using its solution to provide a plan
for solving the original problem, (4) translating the plan back into the original problem space and executing it"
Most goals that are difficult to reach directly can ultimately be reached by first accomplishing
sub-goals.
p.504"In his [Sacerdoti's] view, planning
is performed by recursively decomposing goals into subgoals, until a subgoal can be reached by elementary actions.
If you don't have a goal, it is hard to form a plan to reach that goal.
p.505"In this model, planning is about
how to find the actions that solve a problem, or more generally, reach a goal... The process itself is about formulating goals,
finding similar solved goals, finding existing plans, adapting plans, learning, and storing plans in such a way that they
can easily be found for future use."
Simulation of the future can help generate plans.
p.505"To make a plan, an actor somehow
anticipates the future by simulating the actions he will make. The plan is virtually executed in the model of the future...
This requires the existence of (internal) representations.. In this paradigm, planning is searching for a sequence of actions
that will bring the agent from its current state to the goal state."
Anticipation and modeling are critical components to planning.
p.517"Meystel... states that intelligent
systems... are capable of planning because they are able to simulate the effects of decisions before the decisions are actually
implemented. Here, 'to simulate' means to anticipate and to model... Planning is understood as searching for appropriate future
actions leading to the goal."
What is the main task of an intelligent system?
p.529"The main task of an intelligent
system is to intelligently determine what options will be looked at... the building blocks for this are grouping, focusing
attention, and [combinatorial] searching (GFACS)"
When we plan we search for a path to our goal that has the best chance to succeed, based on what
we know, and even based on what we don't know.
p.529"Planning for yourself is considered
to be problem solving... planning is defined as search for a trajectory through space... looking for a solution in a chess
game may also be conceptualized as searching through a problem or conceptual space."
