What is a system?

Systems are everywhere: information systems, artificial intelligence systems, vehicles, space stations, electronic devices, animals, ant colonies, brains, the financial system, healthcare systems, the weather system, and the whole universe. Almost everything around us that moves or processes material or information is a system or a part of a system. (A bridge on the other hand, while structurally complex, is not a system as it doesn’t exhibit any behavior.)

Two common definitions of a system are:

  • An integrated set of elements, subsystems, or assemblies that accomplish a defined objective. These elements include products (hardware, software, firmware), processes, people, information, techniques, facilities, services, and other support elements [1].
  • A combination of interacting elements organized to achieve one or more stated purposes [2].

Engineered and evolved systems such as vehicles and animals exhibit goal-oriented behavior. Weather systems and the universe do not.

System behavior can be described by algorithms that may be deterministic (producing exactly the same output for the same input) or probabilistic (drawing an output from a probability distribution that is conditional on the input).

Since much or our lives today depend on systems and many of the biggest risks for individuals and for humanity as a whole emanate from systems, it is imperative that we have proper methods and tools to analyze and design useful and safe systems.

To be able to analyze something that we believe is a system, we must first determine if it in fact is a system, or if we are just looking at a heap of inert or separate things. The purpose of this post is to discuss how to identify a system.

Recognizing a system

A simple rule is that if the elements of the “thing” we are looking at don’t exhibit behavior then it is not a system. The beams of a bridge don’t move or process information so they don’t constitute a system.

Whether a collection of active elements is a system depends on several characteristics of the collection of elements:

  • Cohesion: the degree to which the elements interact.
  • Teleonomy: the degree to which the system has a well defined purpose or objective function.
  • Collaboration: the degree to which the elements contribute to a system-level objective function.

Below follows a few examples of systems (and one non-system) that possess different degrees of the three quantities above.

Artificial neural network

Artificial neural networks (ANNs) are software systems inspired by the biological brain. ANNs may for instance be designed to answer questions entered by users or to find cancer in x-ray images.

An ANN consists of interconnected layers of system elements, or “neurons,”. Each neuron takes in data, processes it, and passes the results to the next layer. The connections between neurons are assigned weights (parameters) which influence how much the signal from the output of a neuron affects the input of a connected neuron. The first layer receives the input data e.g., text or images and the last layer produces the output, e.g., an answer or a mask showing where there is cancer in the input image.

The system designers usually choose a base architecture for the ANN, i.e., how many layers, how many neurons etc. They then “train” the ANN with a number of input – output samples where the output is assumed to be correct in some sense. The ANN “learns” by adjusting the weights of each neuron so as to minimize the total error over the training data.

The individual system elements of the ANN, the neurons, are totally controlled by the system-level error signal during training and their contributions to the system function can be mathematically described. ANNs are therefore tight systems along all three dimensions of tightness given above.


In contrast to ANNs, there is no purely mathematical way to design an excavator, even if we know rather well what behavior we want the excavator to have. Instead we have to use an iterative systems engineering approach.

In systems engineering, the system is described in terms of system requirements, statements about how the system is supposed to behave. Based on the requirements, the systems engineers use any previous version of the excavator, their experience, intuition, and creativity to propose a system decomposition that could satisfy the requirements. The system decomposition is defined in terms of system elements like the engine, the man-machine interface, the hydraulic pump, the control system etc. and the patterns of interaction between the system elements like the signaling from the man-machine interface to the control system and from there to the hydraulic valves.

The engineers then proceed to analyze the characteristics of the proposed system to determine if it actually satisfies the requirements. If it doesn’t, a new decomposition is suggested and the process is repeated. This process can be seen as a kind of directed evolution where both the “mutations” and the “selection” are done by the engineers.

Once one level of system decomposition is completed, the identified system elements, some of which are systems in their own right, can be decomposed to even smaller elements if need be. The control system for instance needs to be decomposed to its modules and functions using a software development process.

The individual system elements of an excavator are to a large degree controlled by the system level requirements although some compromises usually need to be made. There may for instance only be a few engine options available, none of them optimal. The excavator is, like the ANN, a tight system.

Ant colony

An ant colony is a system of individual ants. It exhibits behavior that the individual ants cannot exhibit (and probably don’t know anything about). An ant colony is sometimes called a superorganism.

The system is defined by the organizational structure of the ant colony, including the types and roles of ants (queens, workers, soldiers, etc.) and their patterns of interaction. The system elements, the ants, are more loosely connected to each other than the parts of the excavator or a single animal but they still exhibit enough interaction and cohesion to support system level behavior.

Successful colonies are those that effectively perform the tasks necessary for survival and reproduction at the superorganism level. There is therefore teleonomy not only on ant level but also on the ant colony level.

The balance between kin selection and group selection (or system level selection) can vary depending on factors such as the degree of relatedness among colony members, the colony’s reproductive strategy, and the specific environmental conditions the colony faces. However, in many cases, both forms of selection play a role in shaping the evolution of ant colonies.

Because they consist of individual animals and may obey selection on several levels, ant colonies are more loosely connected than the previous two systems.

Large health care system

A large health care system (LHCS) here means a collection of caregivers from primary care, laboratories, specialist care, hospitals, insurers and other players on a (US) state level or European country or regional level.

In LHCSs there is often disagreement within the system both about what outcomes are desirable and how to reach the outcomes. Parties including patients, insurance companies, politicians, hospital administrators, and health care professionals may not agree on what defines a good outcome. There is no full unity of purpose.

Since there is no full unity of  purpose, there cannot be full top-down system-level control.

The interactions between the system elements (care facilities, laboratories, individual healthcare professionals) are not always well-defined but made up along the way. Local idiosyncratic ways of working are common.

It is not possible to design an LHCS top-down in a detailed manner due to the reasons mentioned above.

An LCHS is a loosely connected system. It is loose along all  the three dimensions. It is continuously adapting, nonlinear, creative, unpredictable, and self-organizing.

Digital camera sensor

A digital camera sensor consists of millions of elements. Each element functions independently of other elements so the total is in this case exactly the sum of its parts.

A digital camera sensor is therefore not a system, not even a loosely coupled system. (Each cell in the sensor may have a little system attached but the sensor itself is not a system in the sense defined in the introduction.)

What then?

When we know what type of system we are dealing with, or if it indeed is a system at all, we can select proper methods such as systems engineering [1] to analyze, understand, create, or modify the system. We can also start to unpack phenomena such as emergence and epiphenomena.

More about system design in later posts.


[1] INCOSE Systems Engineering Handbook.
[2] ISO 15288. Systems and software engineering – System life cycle processes.

Related Post

Leave a Reply

Your email address will not be published. Required fields are marked *