p.1 Modeling is a principal tool -- perhaps the primary tool -- for studying the
behavior of large complex systems.... Modeling, then, calls for some basic principles to manage this complexity.
We must separate what is essential from what is dispensable in order to capture in our models a simplified picture
of reality which, nevertheless, will allow us to make the inferences that are important to our goals.
p.1 When we model systems, we are usually... interested in their dynamic behavior.
p.3 if we could forecast [the future], perhaps there are some actions we could take to alleviate its ill
effects and enhance its favorable ones.
p.6 we should redesign our modeling efforts, as far as possible, to focus them on the questions that we
can answer more or less definitively.
p.6-7 the feedback loops in social systems are not passive but predictive. Each of the participants
may be trying to forecast the behavior of the other actors and of the system in order to adapt his or her own behavior advantageously.
p.7 Generally, modeling serves policy. We construct and run models because we want to understand
the consequences of taking one decision or another. Predictive models are only a special case
p.8 Intelligent approximation, not brute force computation, is still the key to effective modeling.
p.8-9 Good attention to feedback and correction will get us to our goal even without very specific temporal
plans.
p.9 Our practical concern in planning for the future is what we must do now to bring that future about.
We use our future goals to detect what may be irreversible present actions that we must avoid, and to disclose
gaps in our knowledge... that must be closed soon so that choices may be made later. Our decisions today require us
to know our goals, but not the exact path along which we will reach them.
p.10-11 the systems, both natural and man-made, in which we are interested are almost always hierarchical...
the behavior of the units at any specific level [of the hierarchy] can be described and explained without the need for a detailed
picture of the structures and behavior at the levels below... Only gross, aggregate, features of the layer below
show up in the next layer above... the dynamic behavior of the system can be factored into different frequency ranges,
each corresponding to a particular level of interaction rates... We can then aggregate, replacing the detail of each
cluster be a few aggregate properties... Using these aggregates, we can then examine the behavior of the clusters
at the next higher level, over a somewhat longer span of time. The process can be repeated indefinitely, as long
as the property of near decomposability is present at the appropriate level.
p.12-13 we will do a better job if, before we begin, we ask what our goals are -- what questions we are
trying to answer... We need to ask whether we can simplify the systems we are modeling by making use of their hierarchical
properties to aggregate, or in other ways. And we need to ask whether there are aspects of the situation of interest
that are better modeled symbolically, in words or pictures, rather than numerically.
We need to ask these questions because, at best, the situations we wish to model are orders of magnitude
more complex than the most elaborate models that supercomputers of the present and future will sustain. We need to apply keen
intelligence, whether of people or computers, to make sure that we capture in our models the aspects of the world's systems
that are important to us.
