Michel Bauwens asked me to discuss bow-tie structures in relation to John Robb’s ongoing use of them on his “Global Guerillas” blog. There is so much to say about bow-tie structures and the ways in which the concept is deployed. This article will give a brief background, take a closer look, and end with a note on the importance of bow-ties for the future.
From the Wikipedia entry “Bow tie (biology)” (http://en.wikipedia.org/wiki/Bow_tie_%28biology%29):
The term bow tie (so called for its shape) refers in science to a recent concept that tries to grasp the essence of some operational and functional structures observed in biological organisms and other kinds of complex and self-organizing systems. In general, bow tie architectures refer to ordered and recurrent structures that underlie complex technological or biological networks, and that are capable of conferring them a balance among efficiency, robustness and evolvability. Bow ties seem to be capable of managing a wide range of inputs through a core (or knot) constituted by a limited number of elements. In such structures, inputs are conveyed into a sort of funnel, towards a “synthesis” core, where they can be duly organized, processed and managed by means of protocols, and from where, in turn, a variety of outputs, or responses, is propagated….
The bow-tie metaphor and diagram has been used to describe everything from biological and technological networks (http://en.wikipedia.org/wiki/Bow_tie_%28biology%29) to economics (http://globalguerrillas.typepad.com/globalguerrillas/2010/03/journal-new-takes-on-the-2008-meltdown.html).
A Closer Look
So, why do bow-tie structures exist? The bow-tie structure is a naturally evolved structure that solves a crucial problem in many-to-many systems.
Complex systems which rely on connections between many different elements, such as from many inputs to many outputs, inevitably face a crucial problem of scale. As these systems evolve, the number of inputs and outputs generally increases. Each time a new node is added to the network, the number of potential connections required scales exponentially (n^2 to be precise).
The solutions to this problem exist on a continuum from 1 to n^2 (the number of nodes in the network, squared). At “1”, everything in the system uses a single component for transformations. At the opposite end of the spectrum, the system uses a unique connection between every node in the network. There are benefits and vulnerabilities to both extremes. The problem with both solutions is one of structural coupling.
If the “1” element is a single node, then it becomes an overloaded bottleneck. If the “1” element is a single standard, then inefficiencies in that standard are imposed on the entire network. Furthermore, because there is only one standard, there is no incentive for innovation, which means that the system cannot evolve. Complex systems theorists say that the system is too tightly structurally coupled. Single standards are notoriously difficult to overcome or dislodge, even when they become ludicrously inefficient, as is the case with the Western “QWERTY” keyboard layout. Tightly structurally coupled systems achieve high-efficiencies at the cost of flexibiliy, but only in the environment in which those efficiencies evolve. When the environment changes, the system has great difficulty overcoming its own internal structure and adapting to the change.
If the “many” elements are nodes, then it is impossible to maintain the network, as the number of connections increases so much faster than the number of nodes. Complex systems of this type, that are too loosely structurally coupled, maximize their openness to innovation but do so entirely at the cost of being able to exploit those innovations when they are useful. Because the system is too diverse, advances cannot easily propagate to other parts of the network. Efficiencies achieved in one part of the system are statistically offset by inefficencies in another part of the system. Again the system as a whole cannot evolve.
“Fortunately, there exists the possibility of optimal network connectivity, i.e. a balance between control and autonomy, a panarchy.” (Hartzog, Paul B. “Panarchy – Governance in the Network Age,” Master’s Essay, University of Utah, 2005, p28, http://www.panarchy.com/Members/PaulBHartzog/Papers/). The bow-tie structure manages these tensions by occupying an “edge of chaos” zone in between too much rigidity and too much flexibility, between too little diversity, and too much.
The periphery remains open to innovation and exploration, while the core takes advantage of the efficiencies available to more tightly coupled processes.
Variations of the bow-tie diagram have been used to describe many kinds of systems, not always accurately.
A diagram of ingoing and outgoing links, as well as the World Wide Web’s “core,” is shown in the “Nature” article “The web is a bow tie” (http://www.nature.com/nature/journal/v405/n6783/full/405113a0.html)
It should come as no surprise that the network scholar who popularized the bow-tie diagram of the World Wide Web (Barabasi, Albert-Laszlo, “Linked,” Plume Press, 2003) also has spent time studying metabolic networks (E. Ravasz, et al. , “Hierarchical Organization of Modularity in Metabolic Networks,” Science, 1551 (2002); p297).
However, to be strictly accurate the center of the bow-tie has to be smaller than the bows, but the WWW bow-tie has 56 million pages in the core, and only 44 million in the sides. Moreover, the diagram is not a process diagram at all. A more likely candidate for a bow-tie diagram of the Internet would show the numbers of servers in the bows, contrasted with a small number of communication protocols in the core. While there are plenty of protocols, each machine does not have to have a special protocol to communicate with each other machine.
It is crucial to understand that the number of highly specialized components in the core is not an absolute, although it might be a constant relative to the network as a whole. This has yet to be determined. It is also possible that complex systems in different environments exhibit a wide range of stable core properties.
Nevertheless, again from the Wikipedia entry “Bow tie (biology)”:
Bow ties seem to be able to mediate trade-offs among robustness and efficiency, at the same time assuring to the system the capability to evolve. Conversely, the same efficient architecture may be prone and vulnerable to fragilities due to specific changes, perturbations, and focused attacks directed against the core set of modules and protocols.
It is for these reasons that one must be careful of referring to complex systems as either “robust/resilient” or “fragile.” Because systems couple to their environment and to their internal organization schema, they are only robust or fragile in relation to perturbations of specific types. For example, some network topologies are robust to random failures but vulnerable to targeted attacks, whereas other network topologies exhibit exactly the opposite characteristics. The “small-world” network (a spoke-and-hub topology) popularized by Duncan Watts is highly vulnerable to the loss of its hubs. Paul Baran’s original network topology design for the Internet is fairly robust to both kinds of failures.
As Gregory Bateson once pointed out, these systems are not “maximizing” anything, but, rather, they are “optimizing.” There is a need to capitalize on potential efficiencies in one’s current environment while at the same time remaining flexible enough to adapt if the environment changes (as environment tend to do).
Finally, it is not precisely accurate to refer to these as “control” mechanisms, as John Robb and others do. While early system theorists like Norbert Wiener often characterized feedback as a kind of control mechanism, contemporary complex systems discourse tends not eschew the language of “control.” However, Kevin Kelly’s book “Out of Control” does a good job of explaining why we need to give up the use of “control” as a metaphor for understanding (Kelly, Kevin. “Out of Control,” Reading, Mass.: Addison-Wesley, 1994, p375-6):
We are all steering….
Direct feedback models… can achieve stabilization — one attribute of living systems—but they can’t learn, grow, or diversify — three essential complexities for a model of changing culture or life….. But we cannot import evolution and learning without exporting control…. There is no control outside a self-making system.
MIT’s Thomas Malone goes even further, suggesting that the hierarchical mode of “command and control” is being replaced by the network mode of “coordinate and cultivate.” (Malone, Thomas W. “The Future of Work” (Boston, Mass.: Harvard Business School Press, 2004).
Consequently, it is difficult to know what to make, for example, of John Robb’s 2008 post about the bow-tie (http://globalguerrillas.typepad.com/globalguerrillas/2008/06/bow-tie-contr-1.html), in which he clearly is capable of explaining the concepts concisely and informatively, with his more recent 2010 post (http://globalguerrillas.typepad.com/globalguerrillas/2010/06/the-resilient-energy-bowtie.html) in which he says:
resilient communities need to be able to input all available energy sources and convert them into a standardized format.
confusing the necessary cluster of evolving core elements with a “standard.”
The notion that we need a universal standard is entirely a 20th century notion. It appears in early works on the internet economy — usually associated with a “standards war” — when traditional economists were struggling to apply their existing, and possibly outmoded, concepts, likely because there was a shortage of new concepts available (Shapiro, Carl and Hal R. Varian, “Information Rules,” Boston, Mass.: Harvard Business School Press, 1999). Future networks operate on multiple standards in the core — optimal levels of infrastructure arrived at by open innovation in the periphery that makes its way into the core as adoption and usage increase. For example, data interchange formats like XML or JSON were proposed methods, but could not be said to have taken up residence in the “core” until adoption became widespread.
Moreover, network culture — political, social, technological — is characterized not by monolithic single standards, the moniker of 20th Century industrial consumer capitalism, but by committments to openness, diversity, and multi-cultural pluralism, that establishes a field in which play and sharing explore and innovate, providing benefits that can be utilized by a highly efficient core of widely agreed upon cultural understandings and practices.
I hope to see you there!
Bow-Tie Control Systems
The Resilient Energy Bow-Tie
JOURNAL: New Takes on the 2008 Meltdown
Bow tie (biology)
Bow Tie Architecture