1.3.1.1. The systems movement from the 1940s to the 1970s

A revived interest in the concept of systems emerged in the 1940s, in the wake of the first-order cybernetics whose seminal figure is Norbert Wiener (1894–1964). His well-known book Cybernetics: Or the Study of Control and Communication in the Animal and the Machine (Wiener 1948) was published in 1948 and is considered as a landmark in the field of controlled mechanisms (servo-mechanisms). The word “animal” in the title of Wiener’s book reflects his collaboration with the Mexican physiologist Arturo Rosenblueth (1900–1970) of the Harvard medical School. He worked on transmission processes in nervous systems and favored teleological, non-mechanistic models of living organisms.

Cybernetics is the science that studies the abstract principles to control and regulate complex organizational structures. It is concerned not so much with what systems consist of but with their function capabilities and their articulations. Cybernetics is applied to design and manufacture purpose-focused systems of nonself-reorganizable components. By design, these mechanistic systems can sustain a certain range of constraints from the environment – never forget that the surroundings in which a system is embedded are part of the system – through feedback and/or feed-forward loops and from the failure of some of its components. In general, this last situation is dealt with at the design stage for securing a “graceful” degradation of operations. This first-order cybernetics clearly refers to Descartes’ school of thought: courses of action controlled by memorized instructions.

Shortcomings were revealed when the cybernetics corpus of concepts was applied to non-technical fields and especially social and management sciences. Second-order cybernetics was worked out under the impellent of Heinz von Förster (1978). The core idea was to distinguish the object (the system) and the subject (the system designer and controller). This delineation focuses on the role of the subject that decides on the rules and means a given set of interacting entities (the object) has to operate with. The subject can be a complex system. It may constitute in different parts, a human being or group and technical proxies thereof encapsulating the rules chosen by the human entity. These rules are subject to dynamical changes due to evolving constraints and set targets that are fulfilled or not. The presence of human cognitive capabilities in the control loop allows for securing the sustainability of systems in evolving ecosystems.

Another important stakeholder – not to say the founding father – in the conception and dissemination of the system paradigm is the biologist Ludwig von Bertalanffy (1901–1972). It is relevant to describe his scientific contribution to systems thinking, contribution that goes far beyond his well-known book General Systems Theory published in 1968 and often referred to as GST (von Bertalanffy 1968). With the economist Kenneth Boulding, R.W. Gerard and the biomathematician A. Rapoport in 1954 founded a think-tank (Society for General Systems Theory) whose objectives were to define a corpus of concepts and rules relevant to system design, analysis and control. GTS was imagined by Ludwig von Bertanlaffy as a tool to design models in all domains where a “scientific” approach can be secured. Opposite to the mathematical approach of Norbert Wiener, Ludwig von Bertalanffy describes models in a non-formal language striving to translate relations between objects and phenomena by sets of interacting components, the environment being a full part of the system. These interacting components match an organizational structure with an inner dynamical assembling device like living organisms. Contrary to Norbert Wiener’s cybernetics feedback and feed-forward mechanisms, actions according to Ludwig von Bertalanffy’s view are not only applied to objectively given things but can result in self-(re)organization of systems structures to reach and/or maintain a certain state, as happens in living organisms. In a world where data travels at the speed of light, the response time to adjust a human organization to an evolving environment is critical. The environment of any system is a source of uncertainty, because it is generally out of the control of the system designer and operator.

1.3.1.2. The systems movement in the 1980s: complexity and chaos

The systems movement in the 1980s became aware of two extents that were up to this time unconsciously not considered, i.e. that complexity cutting through all the scientific disciplines from physics to biology and economics is a subject matter in itself and that some researchers in a limited number of fields pioneered investigation in chaotic behaviors of systems independently. The Santa Fe Institute (SFI) played and still plays a major leading role in the interdisciplinary approach to complexity.

The SFI initiative brings a telling insight into the consciousness felt in the 1980s by scholars of different disciplines that they share the same issue, complexity, and that interdisciplinary discussions could help tackle this common stumbling block to achieve progress in knowledge.

SFI was founded in Santa Fe (California) in 1984 by scientists (including several Nobel laureates) mainly coming from the Los Alamos Laboratory. It is a non-profit organization and was created to be a visiting institution with no permanent positions. It consists of a small number of resident faculty, post-doctoral researchers and a large group of external faculty. Funding comes from private donors, grant-making foundations, government science agencies and companies affiliated with its business network. Its budget in 2014 was about 14 million US dollars. The primary focus is theoretical research in wide-ranging models and theories of complexity. Educational programs are also run from undergraduate level to professional department.

As viewed by SFI, “complexity science is the mathematical and computational study of evolving physical, biological, social, cultural and technological systems” (SFI website). Research themes and initiatives “emerge from the multidisciplinary collaboration of (their) research community” (SFI website).

SFI’s current fields of research which are described on their website demonstrate the wide spectrum of subject matters considered relevant today:

– complex intelligence: natural, artificial and collective (measuring and comparing unique and species-spanning forms of intelligence);

– complex time (can a theory of complex time explain aging across physical and biological systems?);

– invention and innovation (how does novelty – both advantageous and unsuccessful – define evolutionary processes in technological, biological and social systems?).

M. Mitchell Waldrop (1992) chronicles the events that happened in SFI from its foundation to the early 1990s. It is outside the scope of this context to survey all the interdisciplinary workshops run during this period of time. I will elaborate on the contributions of John H. Holland and W. Brian Arthur.

The lecture “The Global Economy as an Adaptive Process” delivered by John H. Holland, Professor of Psychology and Professor of Computer Science and Engineering at the University of Michigan, at a workshop held on September 8, 1987 contains the following main points that are of general application:

• – Economy is the model “par excellence” of what is called “complex adaptive systems” (CAS), term coined by the SFI. They share crucial properties and refer to the natural world (brains, immune systems, cells, developing embryos, etc.) and to the human world (political parties, business organizations, etc.). Each of these systems is a network of agents acting in parallel and interacting. This view implies that the environment of any agent is produced by other acting and reacting agents. The control of this type of system is highly distributed as long as no agent turns to be a controlling supervisor.

CAS has to be contrasted with “complex physical systems” (CPS). CPS follow fixed physical laws usually expressed by differential equations – Newton’s laws of gravity and Maxwell’s laws of electromagnetism are cases in point. In CPS, neither the laws nor the nature of elements change over time; only the states of the elements change according to the rules of the relevant laws. The variables of differential equations describe element states. CPS will be investigated in section 1.3.1.3.

• – Complex adaptive systems have a layered architecture. Agents of lower layers deliver services to agents of higher levels. Furthermore, all agents engineer changes to cope with the environmental requirements perceived through incoming signals.
• – Complex adaptive systems have capabilities for anticipation and prediction encoded in their genes.

Irish-born W. Brian Arthur who shifted from operation research to economics when joining the SFI has kept working in the field of complexity and economics. He produced an SFI working paper in 2013 (Arthur 2013) summarizing his ideas about complexity, economics and complexity economics. This term was first used by W. Brian Arthur in 1999.

Here are the main features of W. Brian Arthur’s positions:

• – “Complexity is not a theory but a movement in the sciences that studies how the interacting elements in a system create overall patterns and how those overall patterns in turn cause the interacting elements to change or adapt … Complexity is about formation – the formation of structures – and how this formation affects the objects causing it” (p 4). This means that most systems experience feedback loops of some sort, which entails nonlinear behaviors “genetically”.

• – An economic system cannot be in equilibrium. “Complexity economics sees the economy as in motion, perpetually ‘computing’ itself – perpetually constructing itself anew”. (p 1). Equilibrium, studied by neoclassical theory, is an idealized case of non-equilibrium which does not reflect the behavior of the real economy. This simplification makes mathematical equations tractable.

This point of view is in line with the dissipative systems studied by Ilya Prigogine. An economic system like a living organism operates far from equilibrium.

A lot of distinguished scholars brought significant contributions to the SFI activities. It is outside the scope of this framework to describe all of them.

1.3.1.3. Ilya Prigogine: “the future is not included in the present”

Ilya Prigogine and his research team played a major role in the study of irreversible processes with the approach of thermodynamics. His reputation outside the scientific realm comes from his co-working with Isabelle Stengers with whom he co-authored some seminal books about the epistemology of science (Prigogine 1979; Prigogine and Stengers 1979 and 1984).

He received the 1977 Nobel Prize for Chemistry for his contributions to non-equilibrium thermodynamics, particularly the theory of dissipative structures that are the very core of all phenomena in nature. He started his research at a time when non-equilibrium thermodynamics was not considered worthwhile since all thermodynamic states were supposed to reach equilibrium and stability sooner or later as time passes by. Only equilibrium and near-equilibrium thermodynamic systems were subject to academic research.

The pioneer work of Ilya Prigogine proved that non-equilibrium systems are widespread in nature: many of them are natural organic processes, namely evolving with time without any possible time reverse. That holds true clearly for biological processes. Science cannot yet reverse aging!

According to Ilya Prigogine, systems are not complex themselves, but their behaviors are. He contrasted classical and quantum dynamics models and thermodynamics models. In classical dynamics (Newton’s laws) and quantum mechanics, equations of motion and wave functions conserve information as time elapses: initial conditions can be restored by reversing the variable time t into –t. Thermodynamics models refer to a paradigm of processes destroying and creating information without letting initial conditions be restored. Time symmetry is broken: this is expressed by comparing time to an arrow which never comes back to its starting point contrary to boomerangs.

A key component of any system is its environment. When this part of the system is not taken into account, the system is qualified as a “closed system”. It is in fact a simplifying abstraction – a falsification in Popper’s terms – which facilitates reasoning and computations. In reality, all thermodynamics systems are “open”, i.e. exchanging matter and energy with their environments. Prigogine called them “dissipative systems”. Why this qualifier? For physicists, irreversible processes, such as the heat transfer from a hot source to a cold source, are associated with energy dissipation, degradation of a sort. There is exhaustion of the available energy in the system. If we want to transfer energy from a cold source to a hot source, energy must be delivered.

A dissipative system is a thermodynamically open system operating out of, often far from, thermodynamics equilibrium in an environment with which it exchanges matter and energy. A dissipative structure is a dissipative system with a reproducible steady state, naturally or under constraints. They constitute a large class of systems found in nature.

All living organisms exchange matter and energy with their environment and are, as such, dissipative systems. That explains why biologists, L. von Bertalanffy among others, were the first to develop interest in this field of research a long time ago. Biological processes are considered by many people as “complex”.

Basically, a chemical system is symbolized by a reaction in which a molecule of species A can combine with a molecule of species B to produce one molecule of species C and one molecule of species D. This process is represented by

[1.1]

A and B are the reactants, whereas C and D are the products. A and B combine at a certain rate and disappear. C and D are formed and appear as the reaction proceeds. In an isolated system, A and B do not disappear completely. After a certain lapse of time, the concentrations [A], [B], [C] and [D] reach a fixed value of the ratio [C] [D]/[A] [B]. How can we come to terms with equation [1.1]? In fact, a reverse reaction takes place

[1.2]

When both reactions occur with the same velocity, a balanced state, equilibrium, is reached. Equations [1.1] and [1.2] are represented as a reversible reaction [1.3] complying with what is called the “law of mass action”.

[1.3]

When a set-up allows the outflow of C and D, the system becomes open. Conditions can be created for the system to attain a state in which concentrations [A], [B], [C] and [D] remain constant, namely d[A]/dt = d[B]/dt = d[C]/dt = d[D]/dt = 0. This state is called the stationary non-equilibrium state. Reaction [1.2] does not logically operate.

Reaction [1.3] may not correspond to an actual reaction but to the final outcome of a number of intermediate reactions. The law of mass action must be only applied to actual reactions. When the situation is described by the two reactions [1.4] and [1.5],

[1.4]

[1.5]

G is described as a catalyst.

The situation described by reactions (4) and (5) is modeled by a system shown in Figure 1.1.

D is drained away as the product targeted is C.

Inside a fluid system, diffusion of reagents and products and chemical reactions are the two main phenomena that take place. A variety of irregularities (inhomogeneous distribution of chemical agents and physical parameters such as diffusibilities of chemicals, temperature and pressure) and random disturbances such as Brownian motion and statistical fluctuations occur in the system and may give rise to unstable equilibria leading to new physico-chemical configurations in terms of space and time. Unstable equilibrium is not a condition which occurs naturally. It usually requires some other artificial interference.

Three salient features characterize the complex behaviors of chemical systems according to Gregoire Nicolis and Ilya Prigogine (Nicolis and Prigogine 1989).

1.3.2.1. Darwin and Keynes on emotions

Perspectives on emotions from the evolutionary theory were initiated by Charles Darwin’s 1872 book The Expressions of the Emotions in Man and Animals. Darwin argued that emotions serve a purpose for humans in communication and aiding their survival. According to Darwin, emotions evolved via natural selection and cross-cultural counterparts. He also detailed the virtues of experiencing emotions and the parallel experiences that take place in animals.

John Maynard Keynes used the term “animal spirits” in his 1936 book The General Theory of Employment, Interest and Money (Keynes 1936) to describe the instincts and emotions that influence and guide human behavior. This human behavior is assessed in terms of “consumer confidence” and “trust”, which are both assumed to be produced by “animal spirits”. This concept of “animal spirits” is still a subject matter of interest as George Akerlof and Robert Schiller’s book Animal Spirits: How Human Psychology Drives the Economy and Why it Matters for Global Capitalism (Akerlof and Schiller 2009) proves.

1.3.2.2. Neurosciences on emotions

The central issue is the distinction between mind and body. In spite of the fact that a lot of philosophers (Thomas Aquinas, Nicolas Malebranche, Bendict Spinoza, John Stuart Mill, Franz Brentano, among others) studied this subject matter, the Cartesian dualism introduced by René Descartes in the 17th Century (Meditationes de Prima Philosophia – 1641) continued to dominate the philosophy of mind for most of the 18th and 19th Centuries. Even nowadays, in some contexts such as data processing, mathematicians are inclined to model the brain function capabilities as software programs made of set instructions without imagining that some disruptive physiological (re)actions can branch out their linear execution.

Descartes proposed that whatever is physical is spatial and whatever is mental is non-spatial and unexpended. Mind for Descartes is a “thinking thing” and a thought is a mental state.

Increased potential in neuro-imaging has allowed investigations into the various parts of the brain and their interactions. The delineation between mind on one side and brain and body on the other side is strongly contended by Antonio Damasio, a neuroscientist, in his book Descartes’ Error (Damasio 2006). He thinks that the famous sentence “cogito, ergo sum” published in Principae Philosophiae (1644) has pushed biologists up to the present time to model biological processes as time-based mechanisms decontrolled from our conscience. In Descartes’ wording, res cognitans, our spirit and res extensa, our functional organs are two disconnected subsystems. Antonio Damasio is convinced that emotion, reason and brain are closely interrelated, which makes the analysis and understanding of human behaviors a conundrum.

In the introduction to his book, Antonio Damasio writes (page XXIII): “the strategies of human reason probably did not develop, in either evolution or any single individual, without the guiding force of the mechanisms of biological regulation, of which emotion and feeling are notable expressions. Moreover, even after reasoning strategies become established in the formative years, their effective deployment probably depends, to a considerable extent, on a continued ability to experience feelings”. This is one of the main tenets that Damasio unfolds in his book: the body and the mind are not independent, but are closely correlated through the brain. The brain and the body are indissolubly integrated to generate mental states.

Let us elaborate on two key words mentioned above, i.e. emotion and feeling. Antonio Damasio (p. 134) distinguishes between primary and secondary emotions. Primary emotions are wired in at birth and innate. Secondary emotions “occur once we begin experiencing feelings and forming systematic connections between categories of objects and situations, on the one hand, and primary emotions, on the other”. Secondary emotions begin with the conscious consideration entertained about a person or situation, and emotions develop as mental images in a thought process. They cause changes in the body state resulting in an “emotional body state”. Feeling is those changes being experienced. Some feelings have a major impact on cognitive processes. Feelings based on universal emotions are happiness, sadness, anger, fear and disgust.

The relationships between emotion and feeling are summarized by Antonio Damasio in this quotation: “emotion and feeling rely on two basic processes:

• 1) the view of a certain body state juxtaposed to the collection of triggering and evaluative images which cause the body state;
• 2) a particular style and level of efficiency of cognitive process which accompanies the events described in (1), but is operated in parallel” (p. 162).

Feelings may change beliefs, and as a result, may also change attitudes. On the basis of the theories mentioned above, events from emotions to attitudes follow this orderly sequence: emotions causing specific body states that generate feelings by cognitive processes and have an impact on beliefs and attitudes. Among the feelings based on universal emotions, mutual fear can be considered as one of the major psychological factors affecting relationships in collaborative networked environments. Fostering trust appears to be a relevant countermeasure to mutual fear. A general argumentative review of the biology of emotional states is given in Kandel’s panoramic autobiography (Kandel 2007).

At this point, it seems relevant to recall that the brain contains a hundred billion nerve cells interconnected by a hundred trillion links and that it is not an independent actor. It is part of an extended system reaching out to permeate, influence and be influenced by every entity of our body system. All our physical and intellectual activities are directly or indirectly controlled by the action of the nervous system of which the brain is the central part. The brain receives a constant flow of information from our body and the outside world via sensory nerves and blood vessels feeding it with real-time data.

When discussing the brain we are faced with a self-referencing paradox: we think about our brain with our brain! We are caught in a conundrum that is difficult to escape and sheds light on the partiality on our own current and future in-depth knowledge of brain processes.

Other relevant contributors to the study of emotion are worth mentioning such as Joseph Le Doux (The Emotional Brain) (Le Doux 1996), Derek Denton (The Primordial Emotions: The Dawning of Consciousness) (Denton 2006) and Elaine Fox (Emotion Science: An Integration of Cognitive and Neuroscientific Approaches) (Fox 2008). The two last references show that the understanding of emotion that is the very fabric of our human lives still motivates scholarship research without the hope of getting a full explanation in the future, considering the intricate pattern of the structure of the brain and the influences of the basic physiology and stress system of the human body.

1.4.3.1. “Traditional” practice

Modeling is the answer humans have found to cope with understanding the ecosystems we live in. Galileo and Newton are considered as the founding fathers of modern physics: their achievements were to construct models, i.e. representations of a part of the world described with a mathematical language allowing for computations to be carried out.

Today, the modeling and system approaches go hand in hand for challenging the increasing complexity of the societal ecosystem in which we are embedded. We are exposed to a deluge of new faceless digital technologies. We are not yet fully aware of the impacts of what is called big data on our behaviors and private lives. For humans and businesses, the lake of data received from their environments is to be filtered and processed accordingly to help them survive.

A model is a constructed representation of a part of the world. Representations are derived from our sensory perceptions of our environments. However, all parts of our environments are not accessible through our five senses (sight, hearing, smell, touch and taste). Some parts of the world are hidden to our sensory perception and their perceived aspects are fuzzy and intangible. Microphysics as well as social behaviors come into this category. When interacting with other people, we have no access to the intentions and ideas behind their mind.

A model is necessarily an abstraction. It captures what we estimate to be the relevant characteristics of the real world from a particular perspective and ignores others. Therefore, the same object and the same entity can be represented by different models.

Models are intended to be manipulated and the results can be fed back to the real world through interpretation. They are supposed to help find solutions to real-world problems.

Many a model has been worked out to meet the requirements of its purpose. A variety of models with respect to their purposes can be listed:

• – heuristic model: this model is intended to mimic the behaviors of a real-world process, as it is perceived by outside observers. The actual courses of action at work may be very different from those imagined when working out the model. The real world is a black box with controlled inputs and received outputs.
• – functional model: this model is developed on the basis of the laws of nature, formal and behavioral logics encapsulated in a theory. It results in deriving equations, allowing the dynamical study of real-world processes.
• – cybernetics model: cybernetics is in fact a modeling approach to the delineation in the world between the human motivation and purpose, and the construction and control of an artifact by human actors to fulfill tasks on their behalf. There is a transfer of a sort of human responsibility for operational courses of action to an artifact agent whose instructions for action are remotely controlled by humans.

1.4.3.2. Object-oriented modeling for information processing

Today, ubiquitous information technology has been relying on software development for the last 50 years. One of the fundamental problems in software development is: how does someone model the real world so that computation of a sort can be carried out? We think that all the efforts made to tackle the issues of software modeling can benefit other fields of knowledge where explicit or implicit modeling is the cornerstone of what can become knowledgeable.

In the 1950s and 1960s, the focus of software system developers was on the algorithmic. The main concerns at that time were solving computation problems, designing efficient algorithms and controlling the complexity of computation. The models used were computation-oriented and the decomposition of complex systems was primarily based on control flow.

In the 1970s and 1980s, different types of software systems emerged to address the complexity of data to process. These systems were centered on data items and data flows with computation becoming less requiring. The models used were data-oriented and the decomposition of complex systems was primarily based on data flows.

What is called object-oriented modeling is a balanced view of the computation and data features of software systems. The decomposition of complex systems is based on the structure of objects and their relationships. This approach dates back to the late 1960s. It has a narrow spine but as a matter of fact offers a wide embrace. It can be used as a blueprint for a modeling mindset in different domains where the real world has to be analyzed and modeled to yield design patterns and frameworks.

We discuss here the fundamental concepts and the principles of object-oriented modeling (Booch 1994). The interpretation of the real world is translated into its representation by objects and classes (sets of objects). A class serves as a template for creating objects (instances of a class). The state of an object is characterized by its attributes, and their values and its behavior are defined by a set of methods (operations) which manipulate the states. Objects communicate with one another by passing messages. A message consists of the receiver’s method to be invoked. Complexity is dealt with by decomposing a system into a set of highly cohesive (functional relatedness) but loosely coupled (interdependency) modules. With the mindset familiar in the realm of information technology (IT), the interactions between objects can be understood as client–server relationships, i.e. services are exchanged between objects alternatively acting as emitters and receivers of services.

In order to maintain a stable overall architectural structure and at the same time to allow for changing the inner organizations of an object, a succinct and precise description of the object’s functional capabilities known as its “contractual” interface should be made available to other objects. That technique, called “encapsulation”, separates the “public” description of the object’s inputs and outputs from how the object’s functional capabilities are implemented.

We will examine whether the concepts used for describing the real world in object-oriented programming can be used to produce a reference framework to assess models developed by different knowledge disciplines. The BDI agent modeling human organizations will be vetted accordingly.

1.4.3.3. Artificial intelligence and system modeling

1.4.3.3.2. Brief historical overview of AI development

Development from the 1960s to the 1980s

Artificial intelligence (AI) research can be traced back to the 1960s. From inception, the idea was to use predicate calculus. Therefore, the first papers on AI principally dealt with automatic theorem-proving, a learning draught-playing system and GPS (General Problem Solving) (Feigenbaum and Feldman 1963). In 1968, M. Minsky edited a book (Minsky 1968) in which research on semantic networks, natural language processing and recognition of geometrical patterns and so on was discussed in detail. At the same time, a specific programming language LISP was developed and a book produced by its creators (McCarthy et al. 1962).

LISP is a general-purpose language, enabling symbol manipulation procedures to be expressed in a simple way. Other programming languages based on LISP but with added and improved facilities were developed in the 1970s which include INTER-LISP, PLANNER, CONNIVER and POP-2.

Applications of AI were first qualified as “expert systems” because they incorporated the expertise or competencies of one or many experts. Their architecture was structured in a way that separates the knowledge of the experts from the detailed traits of a case under study. The stored knowledge is applied to the case and, from deductive reasoning, advice can be delivered to support decisions for data analyses. An important field of application was in health care to help diagnose diseases.

The main components of an expert system are the knowledge base, the inference engine, the explanation subsystem and the knowledge acquisition subsystem.

• – The knowledge base contains the domain-specific knowledge acquired from experts and the particular data captured about the case under study.
• – The inference engine is in charge of dynamically applying the domain-specific knowledge to the case being processed.
• – The explanation subsystem is the interface allowing users to enter requests about how a conclusion has been reached or why a question is being asked. This facility gives users access to the various steps of the reasoning deployment.
• – The knowledge acquisition subsystem enables the capture of knowledge by the expert or generates general rules from data derived from past cases by automated induction.

A key issue is the representation of knowledge. Different forms have different domains of suitability. Rule-based representations are common in many expert systems, particularly in small ones. Much knowledge is declarative rather than procedural. It can be cast in the mould of propositions linked together by logical relationships. Logic-supported rules are known as if-then rules (if antecedent condition, then consequent) associated with the logical connectives AND, OR, NOT.

This structure may be conveniently combined with the representation of attributes and values of objects (A-V pairs) within the very rules. For large systems, an Object-Attribute-Value (O-A-V) representation appears to be a better strategy. Semantic networks portray the connections between objects and concepts. They are very flexible. They have a formal composition of nodes connected by directed links describing their relationships. Objects are nodes. They may be characterized by attribute and value nodes. Has-a, is-a and is-a-value-of are standard names of links associated with object, attribute and value nodes. Some examples in Figure 1.3 show that how this concept is flexible and makes it possible to represent inheritance hierarchies. If a class of objects has an attribute, any object which belongs to this class is characterized by the same attribute.

It is supposed that experts rationally represent their knowledge as various interconnected concepts (frames). Frames consist of slots in which entries are made for a particular occurrence of the type frame. The ability of frames for default values to fill the slots, and reference procedures and their possible linking by pointers, provides a powerful tool for describing knowledge. Frames made it possible to change to a different frame if an expected event does not materialize.

The objective of Minsky’s frame theory was to explain human thought patterns, and its application scope is more far-reaching than its focus on AI. A language called FRL (Frame Representation Language) has been developed and is simply an extension of LISP.

In order to materialize the concept of frame, Figure 1.4 captures the main features of a frame for a long-term loan.

Different principles of inference can be used whether reasoning is concerned with certainty or some degree of uncertainty is involved when complex thinking processes are associated with expertise. In this case, the techniques of reasoning with certainty and fuzzy set theory interact. Fuzzy set theory interprets predicates as denoting fuzzy sets. A fuzzy set is a set of objects of which a part is definitely in the set. Another part is definitely not and some are only probably or possibly within the set. The issue of modeling uncertain knowledge is addressed by the development of what is called neural networks.

Inference control strategies ensure that the inference engine carries out reasoning in an exhaustive and efficient manner. Systems which attempt to prove selected goals by establishing facts for proving these goals, and facts which themselves become secondary goals are backward-chaining systems or goal-driven systems. This approach contrasts with forward-chaining systems which attempt to derive a conclusion from existing established facts. All major expert systems use backward chaining.

Backward-chaining systems postulate a conclusion and then determine whether or not that conclusion is true. For instance, if the name of an illness is postulated, then a rule which provides that illness as a conclusion is sought. If a rule is found, it is checked whether all the preconditions of that rule are fulfilled, i.e. are compatible with the symptoms. If not, then another hypothesis is adopted and the procedure is repeated till the hypothesis chosen holds.

A state of the art in AI in terms of hardware and software by the mid-1980s is compiled in a book edited by Pierre Vandeginste (Vandeginste 1987).

Development of neural networks

Neural networks (also called connectionist networks) gained attractiveness because of their supposed similarities with human information processing in the brain. Why supposed? Because we still have little understanding of how the brain of living animals actually works.

The best approximation to highlight the neural network research that has been most influential would be to combine the papers edited by Anderson and Rosenfeld (1988), with historical remarks by Rumelhart and McClelland (1986), Hinton and Anderson (1981), Nilsson (1965), Duda and Hart (1973) and Hecht-Nielsen (1990).

A neural network can serve as the knowledge base for an expert system that does classification tasks. The major advantage of this approach is that the learning algorithms can rely on training examples and generate expert systems automatically. Any decision is a candidate to a neural network expert system if training data is available. The availability of data lakes and computing power makes it possible to envision the day when short-lived expert systems for temporary tasks become constructed as customized simulation models of human information processing for individual use.

Briefly, a neural network model consists of a set of computational units (called cells) and a set of one-way data connection joining units. At certain times, a unit analyzes its input signals and delivers a computed signed number called activation as its output. The activation signal is passed along those connections to other units. Each connection has a signed number called weight. This weight determines whether an activation traveling along a connection influences the receiving cell to produce a similar or a different activation output signal according to the sign (+ or -) of the incoming signal weight. The aggregated level of all incoming signals received by a cell determines the output or activation delivered by this cell to other cells. An example is shown in Figure 1.5.

This type of cell is called semi-linear because of the dependence on the linear weighted sum S_i of the weights of all the incoming signals from the cells u_j connected to the cell u_i. The activation function f( ) is usually nonlinear, bounded and piecewise differentiable such as:

From a computational point of view, neural networks offer two salient features, namely learning and representation.

Learning

Machine learning refers to computer models that improve their performance significantly on the basis of their input data. Two techniques are used and are divided into supervised and unsupervised models.

In supervised learning, a “teacher” supplies additional data that gives an assessment of how well a program is performing during a training phase, and delivers recommendations to improve the performance of the learning algorithm. It is an iterative learning algorithm. The most common form of supervised learning is trying to duplicate human behavior specified by a set of training examples consisting of input data and the corresponding correct output.

Unsupervised learning is a one-shot algorithm. No performance evaluation is carried out so that, without any knowledge of what constitutes a correct answer or an incorrect one, these systems are used to construct groups of similar input patterns. This is known as clustering.

Practically speaking, machine learning in neural networks consists of adjusting connection weights through a learning algorithm which is the critical mechanism of the whole system.

Knowledge representation

The issue is to define how to record whatever has been learned. In connectionist models, knowledge representation includes the network description of cells (nodes), the connection weights between the cells, and the semantic interpretations attached to the cells and their activations.

Once the learning process has been completed, it appears to be impossible for outside observers to trace back the various steps of reasoning in the form of a sequence of instructions, which is our “natural” way of analyzing our decision processes. This lack of interpretability or “auditability” of machine learning output when neural network techniques are implemented may result in a lack of trust in churned-out results. When important issues arise in risk management such as data privacy, adherence to relevant protocols and cyber security, decision-makers have to be ensured that applications do what they are intended to do. This uncertainty of a sort may appear in the future as a serious hurdle to the mass dissemination of neural network applications in critical contexts.

1.5.1. Ludwig von Bertalanffy (1901–1972)

Ludwig von Bertalanffy was born near Vienna in 1901. He studied the history of art and philosophy first at the University of Innsbruck and later at the University of Vienna where he qualified for a PhD in 1924. Then, he started studying biology and published his first book on theoretical biology in 1928 (Kritishe Theorie der Formbildung). In the 1930s, von Bertalanffy formulated the organismic system theory that later became the kernel of the GST. His starting objective was to derive the phenomena of life from a spontaneous grouping of system impulses. He assumed that a dynamical process exists inside the organic system and modeled it by a heuristic procedure as an open system striving towards a steady state. That results in assigning a selforganizational dynamics to biological systems. In 1934, he published the first volume of his Theoretische Biologie. From 1934 till 1948, he held a teaching position at the University of Vienna. In 1939, he was appointed to an extraordinary “lehrstuhl”.

As a Rockefeller Fellow at the Chicago University (1937–1938), he delivered his first lecture about the GTS as a methodology for all fields of science. Its content was published in 1949 in Biologia Generalis (vol. 195, pp. 114–129) under the title “Zu einer algemeine Systemlehre”.

In the 1940s, he evolved his “theory of open systems” from a thermodynamic point of view. A stable open system is characterized by dynamically irreversible processes. The components are synchronized to one another and to the “Eigeneschwindigheit” of the complex whole. The general system experiences a kind of self-regulation comparable to the behavior of an organic system. Self-regulation should not to be confused with self-organization.

A whole made of interacting elements maintains its structure by an assemblage process of a sort and tends to restore itself after disturbances analogous to the features of living organisms. Isomorphic patterns exist between living organisms, social systems and cybernetic machines. Cybernetics feedback mechanisms (regulation) are controlled by constraints, whereas dynamical systems reflect the interplay of binding powers.

The minimum entropy production stabilizes the system structure and the dynamics of flows. The system will achieve the dissipative state that configures a structure by maintaining a state far from equilibrium.

Ludwig von Bertalanffy always made efforts to disseminate his ideas outside the world of biology long before his well-known book (General Systems Theory) was published in English and became the landmark many people appeal to for using the corpus of von Bertalanffy’s ideas in various disciplines. In particular, he published a paper about GST and the philosophy of science (von Bertalanffy 1950) and another paper about GST and the behavioural sciences (von Bertalanffy 1960).

1.5.2. Heinz von Förster (1911–2002)

Heinz von Förster was born in 1911 in Vienna. He studied physics at the Technical University of Vienna and at the University of Breslau where in 1944 he received a PhD in physics. He worked in a radar laboratory during the war in spite of the fact that he was of Jewish descent. He moved to the USA in 1949 and worked at the University of Illinois where he held a professorship of electrical engineering until his retirement (1951–1975). In addition, he was also a professor of biophysics (1962–1975) and director of the Biological Computer Laboratory (1958–1975). He was an eclectic figure having developed an interest in anthropological research (1963–1965) within the framework of the Wenner-Gren Foundation. He had the opportunity to meet well-known scientists such as John von Neumann, Norbert Wiener, Humberto Maturana and so on.

The BCL (Biological Computer Laboratory) was created one year before the Self-organizing Systems and Bionics conferences were launched in 1959. Analog machines, made of electrical circuits mirroring the behaviors of differential equations like those studied by Turing in morphogenesis, were used in an attempt to reify “biological” computers for simulating the materialization of abstract theoretical concepts like cybernetics, the functional capabilities of experimental artifacts and the social coordination of observers.

He dubbed the order-from-noise principle. Noise in a complex system might well lead to further organization. Noise is a universal phenomenon, considered as a nuisance which may actually play a significant role in large classes of both natural systems and artifacts. It provides undesirable disturbances or fluctuations and as a consequence randomness in repeated measurements. These phenomena are well known in electronic and communication technology, subject matters close to the electrical engineering studied and taught by Heinz von Förster. This principle has been included for some time in what is now called “stochastic resonance” (SR). It has been observed, quantified and described in a plethora of physical and biological systems including neurons, climate models, electronic circuits, lasers and chemical reactions. SR was first used by Roberto Benzi at the NATO International School of Climatology (1980).

SR is a mechanism by which a system embedded in a noisy environment acquires an enhanced sensitivity to small external time-dependent forcing when the noise signal reaches a certain threshold in terms of power and frequency range. It plays a significant role in the emergence of randomness in natural and technical systems as well as in social systems (Barttholomew 1982; Wallace et al. 1997). Interactions between noise-induced randomness and nonlinearity of systems can bring about a self-reorganization of systems exposed to such conditions.

He is the real architect of what is called “second-order cybernetics”. Relying on Humberto Maturana’s theorem “Anything said is said by an observer”, Heinz von Förster derived a corollary “Anything said is said to an observer”. He contended that observed and observing systems cannot be considered independently, each having its own purpose and its own complexity. Information circulates between these two systems, between the mechanistic model and the people who have designed and made these models (artifacts). The paradigmatic concept of first-order cybernetics is the homeostatic loophole. When disturbances of a sort change the state of the mechanistic system, variables that are internal or external to the system are engineered to bring it to the set operational conditions (Ashby 1956).

He devoted his efforts to the study of observers. They are connected through language and form the very nucleus of a society. The triad, observers, language and society, makes up the architecture of observing systems. This view led him to focus on self-referential systems and the importance of “eigen behaviors” for the explanation of complex phenomena. This refers to the “cybernetics of observing systems” leading to his famous distinction between trivial and non-trivial, which is the inception of complex cognitive behavior:

• – trivial machine: its operations are not influenced by previous operations (no memory) and are predictable;
• – non-trivial machine: the structure cannot be deduced from its behavior.

The cybernetics of observed systems may be considered to be first-order cybernetics, while second-order cybernetics is the cybernetics of observing systems. Social cybernetics belongs to second-order cybernetics.

Figure 1.6 portrays the main features that contrast between first-order and second-order cybernetics.

Second-order cybernetics can be analyzed as radical constructivism. It is an unconventional approach to knowledge and knowing. Knowledge is in the heads of people and the thinking subject has no alternative but to construct what (s)he knows on the basis of his/her experience. According to Piaget, experience constitutes the only world we consciously live in. The seminal difference between the conventional psychologist’s approach and Piaget’s is that the former focuses on observable behavior and performance, and the latter focuses on the results of reflective abstractions (mental operations inferred from observations).

If cybernetics is the formalization and modeling of information processing in technical artifacts and living organisms, when the cybernetic nature of self and the world is studied, the observer needs to be made part of the picture within the framework of systems plus environments paradigm. This approach draws on the arguments brought about by Gregory Bateson in 1972. The observer is a complex system in itself. This idea is presented in von Förster’s book (von Förster 1982).

Heinz von Förster was influenced by Gregory Bateson. In his book Steps to an Ecology of Mind, Gregory Bateson applied cybernetics to the field of ecological anthropology and the concept of homeostasis. He saw the world as a series of systems fitting into each other, those of individuals, societies and ecosystems. Within each system, we can find competition and dependency. Each of these systems has adaptive changes which depend on feedback loops to control balance by changing multiple variables. Gregory Bateson believed that these self-correcting systems were conservative by controlling exponential slippage. He saw the natural ecological system as innately good, as long as it was allowed to maintain a homeostasis and the key unit of survival in evolution was an organism and its environment.

Gregory Bateson also viewed all three systems of the individual, society and ecosystem as being all together a part of one supreme cybernetic system that controls everything rather than interacting systems. This supreme cybernetic system is beyond the self of the individual and could be equated to what many people refer to as God, although G. Bateson referred to it as Mind. While Mind is a cybernetic system, it can only be distinguished as a whole and not parts. Gregory Bateson felt Mind was immanent in the messages and pathways of the supreme cybernetic system. He saw that the root of system collapses as a result of western epistemology. According to Gregory Bateson, consciousness is the bridge between the cybernetic networks of individual, society and ecology, and the mismatch between the systems due to improper understanding will result in the degradation of the entire supreme cybernetic system or Mind. Gregory Bateson thought that consciousness as developed through western epistemology was at direct odds with Mind.

“Cybernetics introduces for the first time and not only by saying it but methodologically the notion of circularity, a circular causal system” (Heinz von Förster 1999). The relationship between observed and observing systems as viewed by Heinz von Förster is shown in Figure 1.7.