Typically, self-organisation (SO) is defined as the evolution of a system into an organised form in the absence of external pressures. SO within a system brings about several attractive properties, in particular, robustness, adaptability and scalability. In the face of perturbations caused by adverse external factors or internal component failures, a robust self-organising system continues to function. Moreover, an adaptive system may re-configure when required, degrading in performance “gracefully” rather than catastrophically. In certain circumstances, a system may need to be extended with new components and/or new connections among existing modules — without SO such scaling must be preoptimised in advance, overloading the traditional design process.
In general, SO is a not a force that can be applied very naturally during a design process. In fact, one may argue that the notions of design and SO are contradictory: the former approach often assumes a methodical step-by-step planning process with predictable outcomes, while the latter involves non-deterministic spontaneous dynamics with emergent features. Thus, the main challenge faced by designers of self-organising systems is how to achieve and control the desired dynamics. Erring on the one side may result in over-engineering the system, completely eliminating emergent patterns and suppressing an increase in internal organisation with outside influence. Strongly favouring the other side may leave too much non-determinism in the system’s behaviour, making its verification and validation almost impossible. The balance between design and SO is the main theme of guided self-organisation (GSO). In short, GSO combines both task-independent objectives (e.g., information-theoretic and graph-theoretic utility functions) with task-dependent constraints.
Recent GSO trends are discussed in
More information can be found at GSO web site.
Information-Driven Self-Organisation (IDSO)
Information-Driven Self-Organisation (IDSO) is a specific instance of GSO, where the guidance places constraints on information dynamics.
Many evolutionary and self-organisation pressures can be characterised information-theoretically not only because it's an approximation useful in designing biologically-inspired systems, but also because numerous optimal structures evolve/self-organise in nature when information transfer within certain channels is maximised - i.e., evolution operates at a certain error threshold. In other words, constraints, e.g. noise, reduce the channel's bandwidth, and the system evolves to preserve information by self-improving, re-structuring, and so on (here information is understood in Shannon sense - as a reduction in uncertainty).
Information as "lingua franca"
The weak IDSO view is that it is easier to compare different engineering designs by analysing information dynamics. For instance, imagine a completely centralised modular robot, controlled from a single module/segment that regularly receives data from other segments, computes the best actions for every segment, and sends the instructions back. How would one systematically compare this design with other, more modular, designs? Measuring instructions' size, number of packets, memory usage, etc. would be prone to ambiguities. On the other hand, carrying out the analysis information-theoretically has the advantage of employing "the lingua franca" for multiple approaches.
The strong IDSO view maintains that if such "lingua franca" is possible then it is likely that the evolution/self-organisation (and eventually human designers) discover it, and put to use. According to this view, maximization of information transfer through certain channels is one of the main evolutionary pressures.
Shannon information vs semantic information
Information can be understood both as Shannon information (“reduction in uncertainty”) and semantic information (“meaningful data”). The relation between these two viewpoints remains a subject of active research, promising to shed light on questions whether relevance, reliability, usefulness and semantics of data can be quantified as elegantly as classical concepts in information (Shannon) theory of communication, e.g. data encoding, channel capacity, transmission efficiency, etc. Recent advances in information theory and graph theory allow to consistently and systematically define some of these concepts – most, notably (i) information transfer and flow, and (ii) relevance of information.
IDSO e-mail list
- To join please please let me know (by email)
- The e-mail address for the list itself is idso-csiro at lists.csiro.au
- The list is archived
You are also welcome to join our local discussion group on entropy and self-organisation in complex multi-agent systems and networks.