Complex Decision-Making in Economy and Finance. Pierre Massotte
at the global level, each agent is good at autonomy; each agent carries out his or her own tasks and has his or her own operating modes in a predefined evolution space (field of eligible solutions, still called “prototypes”). Each agent is a complex system that can be subjected, under given conditions, to deterministic chaos or unpredictable behavior. In such systems, the difficulty results from the implementation of agents and interactions between several agents whose functions are different but complementary. In terms of a model and always referring to cellular approaches, the system’s activity is based on the exchange of messages from an agent. Messages are associated with an address (to allow for their dissemination and controlled distribution) and intercepted by other agents and then interpreted and executed.
It should be noted, which is not yet a generality in our industrial systems and is not taken into account in the models used, that the functions in question here are capable of controlled modifications: for example, multiple assignments of equipment or agents to the same task, moving an activity from one agent to another (migration), task differentiation or specialization, adapting an agent to changing contexts, task inhibition or agent removal. Moreover, the order in which these activities or inactivities are programmed reflects an organization’s maturity level. In terms of functionalities and organization, we find the characteristics encountered in increasingly sophisticated and complex societies: indeed, the difficulty increases when it comes to managing, depending on the context, problems of assignment and differentiation (variety of form), migration, adaptation (modification of function) and, finally, “suicide”.
1.2.1.1. Importance of interactions in social behavior
By always comparing itself to supposed models of living systems, the current industrial approach is based on the molecular approach that attempts to explain the behavior of a system by the direct action of a function. Models are then developed to study the propagation of effects, their synchronization, etc., so the effort focuses on the diversity of functional responses in a system. However, it is not the functions performed by the various elements or agents and their scheduling that generate, produce or determine a good or service, but the interactions existing between these agents. Indeed, social control of an activity (or inactivity) condemns the agents involved in a system to interdependence and coexistence. This implies the implementation of an alternative using specialized agents (distributed functions) performing complementary production and control functions, following stacked structures (nested cell architecture) corresponding, themselves, to hierarchical sequences of operations (functional organization specific to a given level), with control and regulation steps. Their behavior in such a communicating system is dictated by the situation and condition of neighboring agents.
1.2.1.2. Interconnections
As we have just discussed, it is necessary here to highlight the fractal nature of the organization or its effects. In such groups, there is a stacking structure with multiple levels (operation, equipment, cell, workshop, factory, etc. in the sense of computer-integrated manufacturing). This fractal structure is equipped with self-similarity or invariance of scale. The same basic mechanisms can be used at these different levels, and the operating point (system configuration) is only a variety (an attractor) of the system, itself based on a small number of initial conditions or simplifying assumptions. The problem here is how such a property is reflected at the upper structural level.
From the above remarks, a number of favorable conditions are nevertheless met for the presence of chaos (diversity factor), studied on the basis of modeling on simple basic cells and associated properties, to exist on more elaborate assemblies. This also leads us to talk about the increasing complexity of systems, the subject of the following section, which will allow us to deduce new approaches to control and management.
1.3. The complexity of systems
It is necessary to understand how increasingly sophisticated systems can be made available as they evolve and how they are structured and organized. These facts can be pinpointed down to several reasons.
1.3.1. The basic principles of complexification
In conventional autonomous systems, the dynamics of evolution are regulated by the eight archetypal changes, grouped here two by two:
– at the interaction level: the relationship between the network elements will be able to be expressed in a more or less strong way and will identify either aggregations of elements or unbundling. Hence, the notion of structure that will appear through the operations of “division” and “combination”;
– at the control level of the element: whether it is a centralized system or not, a network or not, the control can be supervised or controlled by a coordination element. It follows that we will have elements characterized by “autonomy” or “dependence”;
– at the activity level: the functions or programs provided by each element will be expressed or inhibited depending on the environment. Like active sites located on a genome in the field of proteomics, we will have two states: “life” and “death”;
– at the nature of the program level: the element or cell will have to adapt to its environment or may remain generic. It is said that there is “specification” (specialization) or “generalization”.
The first two criteria play a role in the structure, architecture or even the configuration of the upper assembly thus achieved. The last two criteria correspond to the notion of function and make it possible to organize and ensure the functional role of aggregation at a higher level. For example, and in a simplified way, the synthesis of macromolecules is subject to the process of complexification of a category, according to a given strategy (external elements “to be absorbed”, set of objects and links “to be deleted”, etc.). This explains the appearance of more complex objects during the development of a system, during its “growth” and later during its evolution. However, this complexification is carried out according to a set of constraints: economic means with the lowest material, temporal, computer and energy costs.
1.3.2. The complexification process
The complexification process can be repeated, and successive complexifications lead to the appearance of a hierarchy or heterarchy forming a most complex set. The construction, or evolutionary growth of the complex assembly, is done in stages and can be applied to the evolution of the universe as well as to the development of biological or social systems. In the case of a neural system subjected to a simple stimulus, learning neurons will be activated. The resulting recognition (or convergence) activity will allow for classification.
More generally, the development of a complex body is done with cells – or differentiated elements – belonging to very specific categories of the environment. These can change over time (problem of acquisition and intuition). As we have seen in biology, the successive levels we encounter will be able to ensure a set of increasingly elaborate global functions until the constitution of higher order mental objects and cognitive processes.
1.3.3. The smoothing property of chaotic characteristics
In this section, we propose to make the link between Laplacian determinism, according to which “nothing new can happen unless it is already contained in the initial conditions” and deterministic chaos which stipulates that very simple systems give rise to very complicated trajectories and unpredictable evolutions. In the case at hand, the problem is whether a chaotic or turbulent phenomenon at a microscopic level generates a coherent and stable state at a higher level.
To explain this fact, we recall the work of E. Lorenz who highlighted the phenomenon of exponential sensitivity compared to initial conditions: the size of the disturbances doubles each time the time spent increases by a given unit [ROB 01]. Thus, two initially neighboring