Press "Enter" to skip to content

Tag: distributed networks

On energy loss in a system

Every system is in its essence a network of actors that perform it from moment to moment into existence. The participants in the system, or actors in the network, enact and perform it through their daily routine operations.

Some of these routine operations are beneficial to the system being performed, and some are not. Some add to the energy of the system and therefore reduce entropy, while others take away from that energy and increase entropy. If the former outweigh the latter, we can say the system is net positive in its energy balance because it generates more energy than it wastes. If the latter outweigh the former, we can say the system is net negative in its energy balance as it wastes more energy than it generates. How to distinguish between the two in practice?

The rule of thumb is that any action that increases complexity in a system is long term entropic for that system. In other words, it increases disorder and the energy costs needed to maintain the internal coherence of the system and is therefore irrational from the system’s perspective. For example, this includes all actions and system routines that increase friction within the system, such as adding steps needed to complete a task, adding reporting paperwork, adding bureaucratic levels a message must go through, etc. Every operation a piece of information needs to go through in order to travel between the periphery, where contact with external reality happens, and the center, where decision making occurs, comes at an energy cost and generates friction. Over time and at scale these stack up and increase entropy within the system.

Needless to say, the more hierarchical and centralized an organization is, the more entropy it generates internally.

In addition, what appears as a rational action at a certain level is irrational from the perspective of the system as a whole. For example, if a layer of management increases paperwork this is a perfectly rational action for that management layer, because it makes it more needed and important within the system’s internal information flow; however, this is a totally irrational action from the point of view of the system because it increases its internal operational costs.

Put differently, from the point of view of a system such as a large hierarchical organization or a  corporation, the only actions of the agents comprising it that can be considered rational are the ones that increase the net positive energy balance of the system – i.e. reduce internal friction and/or increase external energy intake.

Importantly, this should be viewed across a time axis.

For example, when it comes to a complex operation such as a merger between two departments, or two companies, it might be a good idea to compare the before and after energy net balance for the two systems and the new system that has emerged as a result of their merger. It is also important to look in high enough granularity in order to understand the specifics of each network within the system, and its operations in time.

Say you had two admin structures servicing two different departments, and, now that the departments have merged, senior management optimizes the two admin structures into one, and cuts 50% of the stuff due to ‘overlapping roles’. On the face of it this is logical and should reduce internal energy drag, as admin structures are net negative – they don’t bring in new energy and have no contact with external reality.

However, the new merged admin structure now must service a twice larger part of the system than before, and as a result ends up delegating 30% of that new work back to the front line staff it is nominally servicing. As a result, the front line staff now have to perform 30% more reporting paperwork, which is net energy negative, and that much less time to bring in new energy into the system. In effect, the long-term effects of this ‘optimization’ are net energy negative and result in increased friction within the entire system that was supposed to be ‘optimized’.

Management entropy and the Red Queen Trap

I had an interesting conversation about my essay on the Red Queen Trap with someone on LinkedIn, and it made me think about something I did not explain in the essay.

In an ideal environment each element of a system will be acting rationally and striving towards its own preservation and, by extension, the preservation of the system. Rational action here can be understood as the action resulting in optimal energy efficiency from a given number of viable options, where optimal energy efficiency is a function of the energy that must be spent on the action vs the energy that is gained from performing the action. The scenario I describe in the Red Queen Trap essay is set in such an ideal environment.

However, in the real world individual network actors do not often act rationally towards their own or the system’s preservation. This is not necessarily out of stupidity or malice but is often due to limited information – what Clausewitz called ‘the fog of war’ – or a host of other potential motivations which appear irrational from the perspective of the system’s survival. What is more, the closer an actor is to the system’s decision-making centers, the higher the impact of their irrational decisions on the overall state of the system. The irrational decisions of front-line staff [the periphery] are of an entirely different magnitude to the irrational decisions of senior management [the decision-making center].

In practice this means that in complex hierarchical systems decision-making centers will have much higher entropy than the periphery. In other words, they will be dissipating a lot of energy on internal battles over irrational decisions, in effect actively sabotaging the internal cohesion of the system. As a reminder, the lower the internal cohesion of a system, the more energy the system must spend on performing itself into existence. The higher entropy of decision-making centers may be harder to observe in the normal course of operations but becomes immediately visible during special cases such as organizational mergers or other types of system-wide restructuring.

Interestingly, it is in such special cases when senior management is often tempted to make the internal environment of the system even more competitive – through the layering of KPIs or other means – in order to ‘optimize the system’ and protect its own position in the hierarchy. While on the face of it this appears to be a rational decision, it invariably ends up lowering internal cohesion even further, thereby increasing energy costs and routing even more resources away from the periphery and contact with reality [market competition].

The Red Queen Trap

The Red Queen Trap is to be found in the famous Red Queen paradox from Lewis Carroll’s Through the Looking Glass. In this story, a sequel to Alice’s Adventures in Wonderland, Alice climbs into a mirror and enters a world in which everything is reversed. There, she encounters the Red Queen who explains to her the rules of the world resembling a game of chess. Among other things, the Red Queen tells Alice:

It takes all the running you can do, to keep in the same place.

On the face of it, this is an absurd paradox, but it reveals an important insight about a critical point in the life of every system. Let me explain.

Every system, be that a single entity or a large organization, must perform itself into existence from moment to moment. If it stops doing that it succumbs to entropy and falls apart. Spoiler alert, in the long run entropy always wins.

To perform itself into existence every system must expand a certain amount of energy, which is a function of the relationship between its internal state and the external conditions it operates in. In other words, it must expand some energy on keeping its internals working smoothly together, and then expand some energy on resisting and adapting to adverse external conditions.

The better adapted a system’s internal state is to its external conditions, the less energy it must dedicate to perform itself into existence, and the larger the potential energy surplus it can use to grow, expand, or replicate itself.

However, external reality is complicated [not to be confused with complex] and changes dynamically in ways which cannot be modeled over the long term and require constant adjustments by the systems [organisms, humans, organizations] operating within it. In other words, an external state observable at time A is no longer present at time B.

This is a problem for all systems, because it requires them to change how they operate.

It is a small problem for simple systems which are usually internally homogeneous and highly distributed. Their homogeneity means they don’t need to spend much energy to maintain their internal state, and their distributed topology means they make decisions and react very fast.  

It is a serious problem for complex systems [large organizations] which are usually rather centralized and heterogeneous. Their heterogeneity means they must expand a lot of energy to maintain a coherent internal state consisting of various qualitatively different elements, and their centralized topology means they react and make decisions rather slow.

It is a profound problem for complex hierarchical systems [large organizations with vertically integrated decision making] which consist of multiple heterogeneous elements stacked along one or more vertical axes. Vertical integration means that each successive layer going up is further removed from direct exposure to external conditions and is therefore slower in adjusting to them.

A system might be quite successful in adjusting its internal state to external conditions at time A, but a later time B might present a different configuration of conditions to which the internal state of the system at time A is profoundly inadequate. The more complex the system, the more energy it must expand in adjusting to changes in external conditions from time A to time B.

Complex hierarchical systems have the hardest time in making these adjustments because key strategic elements of their internal state [i.e. decision-making centers high in the hierarchy] are far removed from direct contact with external conditions. To orient themselves and perform the system’s OODA loop they rely on communication about external conditions reaching them from the periphery of the system, while orders on necessary adjustments must travel the other way, from center to periphery. This takes time, and the more layers the signal communicated from the periphery must pass through on its way to the center the more abstracted it becomes from external conditions. In other words, the center receives a highly imperfect version of the external conditions about which it must make adaptive decisions.

Over time, this generates a growing number of errors in the internal state of the system, requiring more and more energy to be routed to internal maintenance [i.e. bureaucratic paperwork], leaving less and less surplus energy for adaptation, growth and expansion. Eventually, and this stage can arrive very fast, the system reaches a state of pseudo-equilibrium in which all energy it can produce goes towards internal maintenance and there is zero surplus energy left. This is where the Red Queen Trap kicks in:

The system does all the running it can do, to keep in the same place.

How does the trap work? First, from the inside everything in the system still seems to be operating smoothly and things are humming along in accordance with external conditions at present time A. However, this is a false perception of equilibrium, because when external conditions invariably change in future time B the system will have no surplus energy reserves to adjust to the new conditions.

The more imperfect the version of external conditions reaching the center of decision-making, the more pronounced the system’s inertia in this state of pseudo-equilibrium, and the deeper it goes into the Red Queen Trap.

Second, having eventually discovered there are no more surplus energy reserves left, the system must now make a choice.  In the absence of surplus energy and provided there is no energy transfer from the outside, it must somehow free up energy from within its internal state in order to adapt. The question is, which internal elements should be sacrificed to free up that energy? This is where the Red Queen Trap’s simple elegance is fully revealed.

Essentially, there are two options – a seductively easy one and an unthinkable one. The seductively easy option is to sacrifice the periphery, or elements of it, and preserve the decision-making center. It is an easy choice for the center to make because it naturally sees itself as the key element of the system and this choice allows it to remain intact. It is a seductive choice because the center suddenly finds itself with a flush of spare energy which it can use to maintain the pseudo-equilibrium and often even to grow itself at the cost of the periphery. Alas, the elegance of the trap is in the fact that the seductively easy option removes the center even further from external conditions; less periphery equals less opportunities to observe and react quickly to external reality, thereby further magnifying the initial conditions that brought the system to this state in the first place. By making that choice the center sinks further into the trap.

By contrast, the unthinkable option is to sacrifice the center and preserve the periphery, thereby flattening the internal structure of the system into a less hierarchical form. It is an unthinkable option for the center to make because, as pointed out above, it naturally sees itself as the key element of the system and this choice forces it to sacrifice itself. It is also unthinkable because it involves a thorough rethinking of the internal structure of the system, which until that moment was organized entirely around vertically integrated decision making, with little to no autonomy in the periphery. The centre must not only sacrifice some of itself, but also reorganize the periphery in such a way so that it can now perform those functions in place of the center. This would allow the system to free itself from the trap.

Obviously, most systems choose the seductively easy option and the Red Queen Trap eventually grinds them into oblivion. Those few systems that go for the unthinkable option escape the trap and, if they remain persistent in their application of the unthinkable, learn how to go different places with running to spare.

Distributed swarms, OODA loops, and stigmergy

This is a third paper in a cycle on distributed swarms, OODA loops and stigmergy co-authored with a PhD student of mine. The paper is titled Distributed Swarming and Stigmergic Effects on ISIS Networks: OODA Loop Model, and was published in the Journal of Media and Information Warfare. This is probably the densest and most interesting paper in the series, as we analyse information warfare waged by distributed swarms in the context of network-centric warfare theory, stigmergic adaptation, and John Boyd’s work on the OODA loop concept. For me the most interesting elements of the paper involve our discussion of Von Moltke’s concept of auftragstactic in the context of maneuver warfare in the information domain.

Network architecture encounters

These are some loosely organized observations about the nature of network topologies in the wild.

In terms of both agency and information, all entities, be they singular [person], plural [clan/tribe/small company], or meta-plural [nation/empire/global corporation] are essentially stacks of various network topologies. To understand how the entities operate in space these topologies can be simplified to a set of basic characteristics. When networks are mapped and discussed, it is usually at this 2-dimensional level. However, in addition to operating in space, all entities have to perform themselves in time.

This performative aspect of networks is harder to grasp, as it involves a continuously looping process of encountering other networks and adapting to them. In the process of performative adaptation all networks experience dynamic changes to their topologies, which in turn challenge their internal coherence. This process is fractal, in that at any one moment there is a vast multiplicity of networks interacting with each other across the entire surface of their periphery [important qualification here – fully distributed networks are all periphery]. There are several important aspects to this process, which for simplicity’s sake can be reduced to an interaction of two networks and classified as follows:

1] the topology of the network we are observing [A];

2] the topology of network B, that A is in the process of encountering;

3] the nature of the encounter: positive [dynamic collaboration], negative [dynamic war], zero sum [dynamic equilibrium].

All encounters are dynamic, and can collapse into each other at any moment. All encounters are also expressed in terms of entropy – they increase or decrease it within the network. Centralized networks cannot manage entropy very well and are extremely fragile to it.

Positive encounters are self explanatory, in that they allow networks to operate in a quasi-symbiotic relationship strengthening each network. These encounters are dynamically negentropic for both networks, in that they enable both networks to increase coherence and reduce entropy.

Negative encounters can be offensive or defensive, whereby one or both [or multiple] networks attempt to undermine and/or disrupt the internal coherency of the other network/s. These encounters are by definition entropic for at least one of the networks involved [often for all], in that they dramatically increase entropy in at least one of the combatants. They can however be negentropic for some of the participants. For example, WW2 was arguably negentropic for the US and highly entropic for European states.

Zero sum encounters are interesting, in that they represent a dynamic cancelling out of networks. There is neither cooperation nor war, but a state of co-presence without an exchange of entropy in a dynamic time-space range. I believe this is a rare type of encounters, because the absence of entropy exchange can appear only if 1] there is no exchange of information or agency, or 2] the amount of agency/information exchanged is identical from both sides. Needless to say, this process cannot be easily stabilized over a long time period and either morphs into one of the other two states or the networks stop encountering each other.

 

Swarm networks and meme warfare

This is a draft of a paper titled Swarm networks and the design process of a distributed meme warfare campaign, which I co-authored with my PhD student Travis Wall, to be presented at the IAMCR 2017 conference in Cartagena, Colombia, July 16-20, 2017.

This paper aims to develop a systemic perspective of the mechanics of generation of targeted memes forming a meme warfare campaign, by analyzing the swarm-like topology of 4chan’s /pol/ forum, and the logistics of the swarm’s rapid prototyping, coordination, production, and dissemination of content. The paper uses as its case study the #DraftOurDaughters campaign, which is documented in its entirety from inception to completion. The main focus of the argument is in developing a coherent and systemic perspective on the logistics of distributed memetic production in online spaces potentiating swarm-like behavior in their user-base. We examine this process in its entirety, from the logistics of swarm formation to the rapid prototyping of ideas leveraging short feedback loops, and the collaborative creation of semantically targeted media. Anonymous online spaces such as 4chan are identified as environments fostering a powerful feedback loop of distributed ideation, content production and dissemination. Through examining these phenomena, the paper also provides perspective on the manifestation of collaborative design practice in online participatory media spaces.