HXA articles

What is decentralization ? (B)

Common meanings for ‘decentralization’ are lacking, because they tend look for something too concrete. Instead, an info-topological view, of the flow of information through causation networks, is better.

3056 words (15 minutes)

Inadequate meanings

There are some common candidate ideas for the meaning of de/centralization, but when you examine them they all dissatisfy.

Rand/Baran diagram ?

"Introduction to Distributed Communications Networks"; Baran; 1964 / RM-3420 / article.

http://www.rand.org/pubs/research_memoranda/RM3420.html (2016-06-29)

This technical report is the source of a diagram often reproduced to illustrate ‘decentralization’.

Immediately it prompts a question: is not the common implication of ‘decentralized’ a flat system, with no centers ‒ that is, the third shape in the diagram? Yet that one is labelled ‘distributed’, and the second shape ‒ with a hierarchical structure of sub-centers ‒ is the one labelled ‘decentralized’. So the first hitch is that the diagram conflicts with common usage, and we do not know which is to be taken as correct. Well, names can change ‒ that does not matter. We can attend to the meaning, and choose the one most distinct from ‘centralized’ ‒ the shape labelled as ‘distributed’.

But there is a deeper problem: confusion of the implementation with its usage. The diagram is very nice, but it was only supplementary ‒ what does the text say?

“Let us consider the synthesis of a communication network which will allow several hundred major communications stations to talk with one another after an enemy attack. As a criterion of survivability we elect to use the percentage of stations both surviving the physical attack and remaining in electrical connection with the largest single group of surviving stations. This criterion is chosen as a conservative measure of the ability of the surviving stations to operate together as a coherent entity after the attack. This means that small groups of stations isolated from the single largest group are considered to be ineffective.”

‒ section 1, paragaph 1.

That is the gist of the article, and it is decidedly not ‘decentralized’ in the colloquial implication. It is clearly stated: it is about preserving a single coherent operational structure. Decentralized in a superficial/subsidiary way, but centralized in purpose and substance.

This is like the common confusion about the character of the internet. The internet is implemented in a decentralized/flat/distributed/network/peer structure, but that does not mean its use will be too. Despite the implementation, the abstraction it presents for use is structureless: it merely enables any point to connect to any other. Consequently the internet does not induce a distributed/flat network structure in its uses, nor in fact any kind of structure. It equally supports any kind ‒ which might include even completely, singly, centralized. The use can perfectly well be the complete opposite of decentralized.

So the Rand/Baran diagram seems a bad definition.

Computer science ?

"The PACE Project, Decentralization definition"; Khare; 2006 / note.

http://isr.uci.edu/projects/pace/decentralization.html (2016-06-29)

This can perhaps serve as a typical Computer-Science academic perspective:

“A decentralized system is one which requires multiple parties to make their own independent decisions”

“In such a decentralized system, there is no single centralized authority that makes decisions on behalf of all the parties. Instead each party, also called a peer, makes local autonomous decisions towards its individual goals which may possibly conflict with those of other peers. Peers directly interact with each other and share information or provide service to other peers.”

One ought notice that this is not a system ‒ no structure has been mentioned. There is nothing that organises behaviour; every individual simply pursues their own intent. (They might align, at some region, for some time, but that would be purely chance ‒ absent some stated ordering principle ‒ and such randomness would be improperly called a system.). You might reply that it is a system simply in being a collection of things, in being the stuff under consideration. But that seems much too weak a meaning: ‘system’ must mean more than merely some arbitrary jumble.

Yet the wording of ‘no single centralized authority that makes decisions on behalf of all the parties’ could allow some organisation in. It just must not actively, at the time, participate in steering individuals. So setting the rules of the game, up front and immutably would qualify: it would not be ‘making decisions on behalf’ of individuals, yet it would certainly shape their behaviour.

But this should feel uncomfortable. It is not really the immanence, the close presence of the control, that is the core of its objectionability (though having something looking over your shoulder is irritating), it is the control itself. If something sets the rules of the game you play, you are just as much controlled. And the covertness ‒ is it not worse to be unaware of your constraint? And are not the rules of a game ‒ internalised by the player ‒ more immanent than ongoing explicit/plain commands?

So either this definition of ‘decentralization’ is not describing a system, and so is useless ‒ you may as well simply say ‘disorder’. Or what it denominates as decentralization is so in name only: it actually has all the weight and flavour of the control that is objectionable in its supposed opposite, centralization.

So the typical computer-science perspective seems a bad definition.

Political, physical ?

Does ‘centralized’ mean having a governing organisation? That seems to fit colloquial sentiment, and the language often used. So does that mean people? A committee etc ‒ humans being the source of command? Then let us imagine an automation of it: let us put together a software simulation, and roughly capture what is done in algorithmic form. It might be less effective, less complete, etc, but it will now pass the test: the system is not human-run, so no longer centralized. But that seems dissatisfactory.

Well then, what if one day a bank sends all its employees away, telling them to work from home, is it then decentralized? If they work in their own offices with their own computers, is that decentralized? Is decentralization a measure of distance? A length? Is it which pieces of equipment, or which building is being used? That all seems on the wrong track.

If you now concede that centralization is not about people or location or material, but structure more abstractly ‒ interrelations, then the response must be: so the internet, Bitcoin, markets ‒ they must then all qualify too: they are just as centralized as anything you would usually complain of. Bitcoin is no more decentralized than a bank where all its employees work from home. Blockchains are one of the most centralized things around: a big lump of data that everyone must work to. Yes, on separate CPUs, separate equipment, but as we have seen, so what? The arrangement of the hardware is unimportant.

So there is a problem. What substantive definition of ‘decentralization’ is there that discludes the liked from the disliked?

So political, people-based, or location-based, or anything ostensible like that seems a bad definition.

Amount of control ?

Maybe one could retreat a little to a weaker claim of distinction, one of decentralization being of relative degree.

Is not less micro-management better? To have every detail of movement controlled is objectionable, and functionally inefficient too. Surely this is a legitimate principle of assessment, to declare the less centralized the better choice. It gives no sharp division, but judgement works by comparison.

But this does not really work: it is an under-determined criterion. If you rate by this alone you must always be driven to the very end-point. If you complain of the incumbent and offer a ‘less centralized’ alternative, someone else can do exactly the same to you. Such comparison could depose any candidate, and then any replacement would also fall to the next, and so on ‒ it will always drive to the extreme point of no management/constraint at all. Evidently, a single relative criterion is insufficient on its own to decide ‒ you need other considerations as well.

(In real cases there are multiple variables of preference, and these must also count. Abandoning the relative priniciple, the adjacent reasonable question is the cooperative-system one: does this amount/kind of control bring sufficient benefits to outweigh it?).

So relative amount of control is a bad definition. (Though it does bring us closer to a better understanding …).

Info-topological meaning

Decentralizers physicalise centralization ‒ people talk about no central authority, or person, or object, they talk about centralization as if it were a physical thing. And this misdirects understanding.

But the concern of decentralization is not really about physical objects, it is about control. If you offered them a system that had no central object of any kind, but still could control everyone, by some magic means, they would still have their original complaint.

And there is just such ‘magic’ means ‒ information. Control is material causation steered in a particular direction, and the choice of that steering is what is set by information.

This informational perspective is not superficially apparent. If you are thinking in terms of material objects, examples like flocking, species, language, seem obvious defenses of decentralization: they are substantial coherent phenomena, yet where is the the central object? There seems to be none.

You have to ask how can these work to see that they, and all systems, are information-driven structures, and that pattern/order/coordination of multiple elements ultimately can only flow from single seed-points.


The first thing to notice is that there is no such thing as a purely centralised or decentralised system; a system is always a balance of both.

De/centralization is indeed about where the processing will be done: what processing will be done on which hardware in which places. But to work as a system, that processing must be connected together by some equally clear shared protocol, some pattern of congruent action. And that structure must have been set earlier and centrally.

So any appraisal of de/centralization always depends on a tacit definition of context, which sets the basic limits and interactions. The current processing has explicitly independent parts, but they inevitably work within strict bounds of coordination, which are not explicit.

A system is therefore never simply ‘decentralized’: it is always both decentralized and centralized: it must always rest upon a central presumption. Or, from the opposite direction: a system cannot comprise only a single individual, it must have multiple parts, which, to be so, must be different and independent in some way ‒ so a system's centrality must always have some decentralization too.


Why are these two sides are necessarily combined? Because system organisation is correlation (in the colloquial sense), correlation implies shared information, shared information implies common causation, and common causation is the meaning of centralization.

The basic components here are individuals ‒ individuals/agents/actors/nodes: not particularly people, just abstract elements ‒ and their actions, and relations between those individuals.

If two individuals are uncorrelated (in their actions), they are free with respect to each other. Neither's actions have any implication for the other's ‒ neither can be used to predict the other, neither can be affecting the other. On the other hand, when there is a correlation, they are no longer free, there must be some constraint present. Organisation is mutual, reflexive, because it is correlation. One cannot be organised with respect to another, without them being organised with respect to you. What one does implies something about what the other does. Either A is controlling B, or B is controlling A, or some hidden C is controlling both.

By acting in concert, something is shared, and the only thing that can be truly shared is information. Physical material might be divided-up, but that which can exist in multiple places simultaneously, is abstract: information. And the only way substantial information can be the same in multiple places is by issuing from a single source. The bit-string of any substantial information is so long that the chance of it being duplicated by accident are infinitesimal. Same/similar info must be copied info.

So order and centralization are indivisible: centralization is just a word for a single-source cause, which is the only way to produce the same information and hence patterns of action, which is to say organisation. Organisation = centralization. So also ‘decentralization’ and ‘organisation’ are contradictory: there cannot be any such thing as a decentralized organisation/system; decentralized means (if it means anything substantial) no such center, in which case there would be no shared structure, and it could not be a system. Decentralization = disorder.


The informed meaning of de/centralization is as a sliding scale of organisational strictness. The more tightly organised, the more control by centralization; the more decentralized, the more loosely organised, or disorganised. You can describe something as more or less de/centralized, compared to others. But to say ‘decentralized system’ is rather a contradiction: the degree to which it is decentralized is just the degree to which the system is fading away, and it is not a system.

Examples considered

Setting up a telephone system

Imagine setting up a telephone system ‒ what would that imply? What would a decentralized one imply?

Everyone invents their own? Then none fit any others, and no-one can talk to anyone ‒ hardly a system. A set of clusters (‘decentralized’ as in the Rand-Baran diagram)? Then each cluster is following one design ‒ there are the centralizations. But if you want to call your friend, nothing happens. The number is not recognised ‒ they are in another cluster! The problem has merely shifted up a level. You have some connection, but not complete coverage. Is that what ‘telephone system’ normally means? Is that what you want? No: you want to call anyone, globally. But for global connection, you need a global centralization.

A continually aggregative project

What about Stigmergy: “work performed by an actor stimulates subsequent work by the same or other actors”, “The actors know what to do next, based on what has already been done.”? Is that not some kind of decentralized system? No.

The distinctive element in stigmergy is the intermediary of environmental state: communication is not direct between agents, but information is transmitted indirectly via environmentally recorded messsages/signals. The interaction structures available depend on what info can be stored, and what ways the storage medium can propagate that info.

This does not say anything in general definitive about decentralization, because it does not say anything about the network/graph topology of the info flow. If one agent can only write and all other agents can only read, then you have a very centralized structure.

Stigmergy is merely a form of implementation, so it falls under the previous ‘physicalisation’ criticism, of mistaking centralization as requiring a single physical thing. De/centralization is about shape of info flow, and stigmergy does not in general constrain that to either extreme.

Ants and interference patterns

Does not ant cooperation (and similar) provide a neat example of ‘decentralized system’? But the centralization is in the DNA, predominantly (and also the environment). Ants cooperate in that way because they share those informational premisses.

What about interference patterns ‒ are not the two sources separate sources of info? Not entirely: underlying both is a single source of info: physics. The way in which they interfere with each other is the way in which they are related to each other, and that is the shared substrate of physical law.

Abstract is real

Even so, someone might still complain: ‘Where is the single point of control? There is no single ant, or even a single pattern of DNA. Answering that the centrality is ‘abstract’ seems weak ‒ there is no actual single thing!’.

But in all these the mistake is to still disregard the abstract centralization, and only see discrete solid objects. Yet the abstract here means something, it denotes real use. If you assert there is no single point of control, does it not follow that there is no single object of understanding, no single thing that can be understood? But then what are we talking about? That a pattern has parts, or no sharp boundary, does not mean it is not a pattern, that it is not understandable. And to be understandable is to be manipulable. If one could understand a phenomena, and so know that by manipulating quite separate pieces, gain an intended outcome, does that not substantially represent a single point of control? Merely because the adjustments are spatially distant seems itself a weak way to suggest control is not possible.

‒ “But the structure is ‘emergent’.”. So it is not controllable because it is not predictable? But then it is not a system, and that half of the original proposition falls. Calling something a system only after it happens to appear is fallacious. It means this structure is never any use as a whole because you never know when it will occur. You might opportunistically use any internal pattern it has, but that is only possible because that part has predictable forms. You cannot use a phenomenon that is unpredictable ‒ it happens to you, you do not do anything to it: it is outside intended action.

If something is usable, it is predictable, thus understandable, therefore controllable. So if decentralized means not controllable, ‘decentralized system’ remains a contradiction.

Consider language: groups of agents, it could be posited, in an iterative process ‘negotiate’ a form of communication, a language ‒ it was not pre-set, it was ‘emergent’. That is the talisman: ‘emergence’; that seems to capture some special phenomenon. But the reply is as follows: what if you wanted to write a message to be readable in 1000 years? You have a problem. And why? Because the language evolution/emergence is not predictable. But if it is not predictable, not understandable, it is not ordered, it is not a system. (Let us not confuse the internal structure of the language with the ‘outer’ whole: the emergence defense only applies to the whole.). So ‘emergence’ is no explanation of ‘decentralized system’, not because the ‘decentralized’ half fails, but because the ‘system’ claim does.

If we can make predictions or exert control, there is a thing we are interacting with, no matter how dispersed or airy it may seem. And if we cannot so interact, there is no thing there.



{ "DC": {
   "title": "What is decentralization ? (B)",
   "creator": "Harrison Ainsworth",

   "date": "2021-07-18",
   "date": "2016-06-19",

   "type": "article",
   "format": "text/html",

   "language": "en-GB",
   "subject": "philosophy, information, decentralization",
   "description": "Common meanings for ‘decentralization’ are lacking, because they tend look for something too concrete. Instead, an info-topological view, of the flow of information through causation networks, is better. (3056 words)",

   "identifier": "urn:uuid:2022334F-130F-4C02-B8CE-1C9AE5075C09",
   "relation": "http://www.hxa.name/articles/content/what-is-decentralization-b_hxa7241_2021.html",

   "rights": "Creative Commons BY-SA 4.0 License"
} }