Stage Nine. Building and Improving Models

Perhaps the most amazing thing in trying to understand how the world works is that taking into account only the finite sets of relations in an infinite world, we often succeed in achieving our goals. Whether the world is arranged “simply”, either ourselves are very “limited”, or our interaction with the world is “superficial” - these are philosophical questions, but the fact is that finite, simplified models allow us to successfully learn and transform (!) the infinite world. However, though not many models are suitable for this, some meet a number of requirements that we generalized in the concept of adequacy (see Chapter 3, Part I).

As already noted in Chapter 3, without modeling, no activity is possible at all. In the systems analysis, the problem situation model is needed to test possible interventions on it, to cut off not only those that are unimproved but also to choose the ones that improve the most (according to our criteria) among the ones that improve.

It should be emphasized that the contribution to the construction of a situation model is made at each previous and at all subsequent stages (both with one’s own contribution and the decision to return to some early stage to replenish the model with information). Therefore, in reality, there is no separate, special “stage of model building”. Nevertheless, it is worthwhile to focus on the features of building models, or rather “completing them” (i.e., joining new elements or removing unnecessary elements). Therefore, let’s do this in the form of separation of these operations as if in a separate stage of analysis.

Much important for this work we have already discussed earlier.

First, in Chapter 2, we outlined the fact that there are only three types of models of a system: the black box (a list of essential connections with the environment), the composition (a list of essential compound parts), and the structure (a list of essential ties between parts), along with their necessary combinations. The difficulties faced by those who build these models were also discussed and will not be repeated.

Second, in Chapter 3, we discussed how to build models through analysis and synthesis.

Third, in Chapter 4, we examined the method of trials and errors, in which the completion, building, modification, and correction of models are carried out by including in it new information obtained during every next experiment with the system.

Fourth, in the section on language models, we emphasized that as the degree of knowledge of the system increases, the system model goes from its “soft”, “loose” design in a verbal, qualitative form, through filling with new information (expressed in “professional” languages), up to (if necessary, i.e., if the problem has not been solved at the current stage) its increasingly “hard”, more and more formalized description, in the end leading to a mathematical one.

Perhaps the most amazing thing in trying to understand how the world works is that taking into account only the finite sets of relations in an infinite world, we often succeed in achieving our goals. Whether the world is arranged “simply”, either ourselves are very “limited”, or our interaction with the world is “superficial” — these are philosophical questions, but the fact is that finite, simplified models allow us to successfully learn and transform (!) infinite world. However, though many models are suitable for this, some meet a number of requirements that we generalized in the concept of adequacy (see Chapter 3).

About qualitative models. Building “soft”, “loose” qualitative models is more an art than a science. But the following are some useful tips: ^{[1] [2] [3]}

variable. (Let us recall the example of a bus company from Chapter 1 when it turned out that it was productive to bring conductors out of buses to bus stops at peak hours.) Of course, it’s promising to study a factor that is not controlled due to lack of knowledge about it.

• 4. It is useful to keep in mind that the desire to bring all ties to cause-and- effect often leads to inadequacy of the model. The acorn is not the only factor leading to the growth of the oak, but many other conditions are needed, without which the oak will not grow out of an acorn: soil, moisture, temperature, light, etc. It is useful to use the concepts of directional correlation, producer-product, environment and conditions, etc.
• 5. Of the scientific constructions that significantly promote the construction of models in a “soft” situation, some examples include the theory of the situational management of the Moscow school, headed by D.A. Pospelov [6], and the theory of detection and amplification patterns laws of the Novosibirsk school of N.G. Zagoruiko and G.S. Lbov [7].

About quantitative models. In practice, increasing importance is attached to quantitative modeling. In this case, the “transparent box” model (a combination of black box models, the composition, and structure of the system) is embodied in the form of some formula or algorithm linking the input variables X with the output Y Y = f(X). Quantitative models can be descriptive and phenomenological when a formula is constructed heuristically and its coefficients are selected for best agreement with experimental data.

An important aspect of such a construction of quantitative models was highlighted by R. Ackoff in his 21st “f-law of management” [1]:

“The less managers understand their business, the more variables they need to explain it. E = me (special theory of relativity) contains one independent variable, m, and explains, perhaps, the most complex phenomenon understood by scientists. Then why does it take 35 variables to explain why people prefer a particular store and buy a certain cereal? The answer is obvious: these phenomena are not understood. The less clear something is the more variables are needed to create an alleged explanation for this.

That is why when managers do not understand what is happening, they collect all the information they can. Not knowing what information is relevant, they are afraid to miss something significant. As a result, they suffer much more from an excess of irrelevant information than from a lack of essential information”.

Another form of quantitative model is preferred when the formula is derived from certain theoretical assumptions. In any case, the task is to identify the model, that is, determine the model parameters in which theoretical predictions and practical observations best fit.

Questions and Tasks

• 1. What is the difference between qualitative and quantitative models?
• 2. What does it mean to turn a black box into a transparent one?
• 3. What is “model identification”?

• [1] It is necessary to divide all inputs of the task into controllable and unmanaged ones. Controlled variables are subject to us, and unmanaged variablescharacterize conditions and constraints of the problem situation.
• [2] When allocating controlled variables, it must be borne in mind that thecorrelative relationship between variables may be mistaken for causativeones. Let’s consider an example. In one U.S. city, it was found that in areaswhere soot air pollution is greater, there is a higher incidence of tuberculosis. Effective measures were taken to combat soot emissions. After a fewyears, air pollution decreased significantly, but the incidence of tuberculosisdid not. It turned out that malnutrition was the main cause of the disease.The connection with air pollution was indirect: in areas with a poor environment, the rent was lower and there were mostly poor families living thatwere poorly fed. So, the trap in this case is that we consider only knownfactors familiar to us from experience to be controlled. You can get aroundthis danger by creating interdisciplinary teams of developers who look atthe problem from different angles.
• [3] When considering uncontrollable factors, a very promising solution to aproblem is the transformation of an unmanaged variable into a controlled