Six Things Nothing Else Does — and What Each Costs

What Part Two leaves us owing

Chapter 14 set our case record beside Ostrom’s, and we did not come out ahead. It was the last of Part Two’s comparisons. Chapter 10 had listed the components, Chapter 11 followed nesting through Simon, Koestler and Jaques, and Chapter 12 followed variety logic through Galbraith. Six neighboring literatures, set beside the model, had arrived at most of its anatomy independently. We cannot answer that survey with six claims of discovery. Nearly every part we hold was found by someone else, working from other premises, in several cases earlier and with a better evidential record than ours.

What is left to us is a claim about assembly. The claim to invention did not survive Part Two, and Chapter 23 has already narrowed what remains as far as we think it will go. Here are its six parts in full. Each is a credit we think the model has earned, stated together with what it costs, what it has not been tested against, or what it cannot do. That pairing is the convention of this part, and the reason for it is practical: a credit released without its concession travels further than the evidence behind it, and we have watched that happen to every one of these six.

Each part also carries a maintenance cost, and we mean three specific things by the phrase. Practitioners must learn a vocabulary before the framework does any work at all. The framework makes a narrower set of questions easy to ask, which is the other face of its precision. And a small community must keep spending attention to stop the distinctions from eroding into loose usage. These six cost claims are argued from our own practice and are not grounded in documented cases.

One: a closed set, asserted as necessary

The first is that Beer makes a closed-set necessity claim. Galbraith, Mintzberg and Jaques describe what organizations have;^[1] Beer asserts what any viable system must have, and puts a number on it. Five kinds of work, no fewer, or the thing stops holding itself together. He argued the closure in logic: “There could not be a ‘System Six’ — because the five-fold system is a closed system, in logic.” He was less sure of the count when he first set it out. Introducing the five systems in Brain of the Firm as a hierarchy modelled on the somatic nervous system, he wrote that it “seems to be” five, “because the choice of five (rather than some other number) is somewhat arbitrary.”^[2]

The shape of that claim matters more than its content. A descriptive typology absorbs a new case by growing a category. A necessity claim absorbs nothing: a sixth indispensable function would end the model rather than extend it. That makes the model refusable as well as serviceable, and refusability is the quality Part Five is written to exercise.

The concession is that a claim this strong ought to be testable and has barely been tested. We have not specified the observation that would show a viable system running without one of the five, and we have not agreed among ourselves what such an observation would look like. Beer’s texts make it hard to specify, because he holds that a persisting organization always performs the functions in some form; Chapter 28 sets out the coding that would let such an observation count. Chapter 9 is our account of why that absence never cost us anything. The comparative claim needs the same caution. We say that almost nothing else in organizational thought makes an invariance claim of this shape, but that rests on our reading and not on a survey we have conducted, and we offer it as an impression.

The maintenance cost goes beyond remembering five labels. The community has to keep necessity distinct from a useful descriptive checklist, which is the form the model most often takes in teaching. And a closed set narrows attention to failures that can be expressed as a missing or pathological function. An organization can fail from a cause with no home among the five, and a fluent practitioner will still find a name for it inside the model.

Two: recursion with a criterion

The second is recursion↗^[3] with a criterion. Levels are common property, as Chapter 11 shows at length. What the others lack is the claim that the same anatomy appears at every level, joined to a test for whether a candidate level qualifies at all. In our paraphrase of Beer, the candidate must be viable in its own right and one of the units that do what the containing system does.^[4] Koestler had the intuition and supplied no test,^[5] and the test is what separates a criterion from a resemblance.

The criterion earns its keep by embarrassing work done in the model’s name. The ascending order in our companion book Wild Futures is, at most of its joints, a ladder of scale and not of recursion, and it was Beer’s own test that showed it. A model able to correct a book written in its name is doing more work than a metaphor does.

The concession is that the criterion, applied honestly, excludes a great deal of what people want to do with it, including work of ours that reads better before the test than after. Nor have we applied it systematically across our own case record; the Wild Futures result is one application, not a survey. Beer did not always apply it either. His recursions “can be extended upward to the terrestrial globe (within the Universe) and downward to the cell,”^[6] a chain his own test would have to check link by link. The maintenance cost is the repeated boundary work of telling a viable recursion from a convenient level, and the opportunity cost of refusing attractive cross-scale comparisons that fail the test. Refusal is unrewarded work, which is why the criterion tends to be quoted more often than applied.

Three: a diagnosis, not a description

The third is that the model yields a diagnosis rather than a description. Given a named system, it says which function is missing or pathological and what will follow from the absence. Soft systems methodology yields a learning process; Cynefin yields a sense-making device; Luhmann yields an account; actor-network theory yields a description of unusual richness.^[7] None of them walks into a particular firm on a particular Tuesday and comes back with: your System 2↗ is done by nobody designated to do it, too slowly for these two units, and they will oscillate within the quarter.

That specificity is what exposes the model to correction. A framework that only reframes cannot be caught being wrong. One that names a function and dates a consequence can be, and the practitioner who offers such a diagnosis takes a risk the reframing traditions never have to take.

The concession is that the diagnosis is only as good as a naming the model cannot supply: which system, bounded how, in whose interest. Chapter 5 makes the point for measurement and the Coda makes it for boundaries. Change the naming and the same procedure returns a different diagnosis, with no internal signal that anything has moved. The example above also shows where the model protects itself. For a persisting organization it cannot return “absent” for System Two, which Beer says “is always present, or the organization would shake itself to pieces.”^[8] It returns degraded presence, unrecognized, slow or overloaded, and degraded presence is harder to show wrong than an absence. The coding in Chapter 28 is meant to tell the two apart. The maintenance cost is disciplined naming before diagnosis, and sustained attention to the evidence that would separate a prediction generated in advance from a label applied after the event. Chapter 15 is candid about how little of that evidence we hold.

Four: a pathology catalog that is generated

The fourth is a pathology catalog that is generated rather than collected. System 3–4 imbalance, the autonomy–cohesion oscillation, the algedonic↗ channel that is present but unheeded, a metasystem that has collapsed into System Three: these are derived from the structure and, in Beer’s presentation, confirmed in practice.^[9] Most rival frameworks we know have a case library where this has a theory, and a theory predicts failures nobody has recorded yet, which a library cannot.

The concession is that generativity without testing is also the recipe for a vocabulary that cannot be wrong. A catalog derived from structure always has a name ready for whatever happened, and the fluency with which we supply the name is no evidence for the derivation. Maintaining the catalog therefore costs prospective coding and failed predictions, where we have so far offered ever more ingenious retrospective diagnoses: the pathology written down before the quarter, the prediction that did not come true recorded next to the ones that did. Chapters 28 and 29 set out what that would take.^[10] The one quantitative instrument built on the model has the same gap. Michael D. Pfiffner’s viability construct (Chapter 18) was never tested against a baseline or a rival predictor of crisis; that comparison could be run on his 135-case dataset if it is released (Chapter 29). Without it, the framework narrows every failure into one of its own names and reports the narrowing as a confirmation.

Five: one currency across scales and substrates

The fifth is one currency across scales and substrates. Variety lets a person, a firm, an economy, a species and a machine collective be discussed in a single vocabulary, and that property is what allows this book to compare claims made at unlike scales at all. Ostrom does not scale down to a person. Active inference does not, at present, scale up to a polity in any usable form. Luhmann does not descend below the social.^[11] The convertibility is not a stylistic convenience; it is the condition of the argument we are making.

The concession is that a currency which converts everything may be a metaphor wearing the clothes of a measure. Beer saw it: “Could we ever say: this is the variety of the chemical industry, say? Of course we could not.”^[12] Chapter 3 shows how far the law we quote has drifted from the law Ashby wrote, and Chapter 5 shows what happens when the ladder is used without a ruler. The maintenance cost follows directly: every comparison needs an explicit statement of what counts as a distinguishable state on both sides of it. Where that statement is missing, the shared word hides quantities that do not commute. It also steers attention away from whatever will not convert, which is often the part of the situation that matters most to the people inside it.

Six: the algedonic channel

The sixth is the algedonic channel, and it comes close to being singular. The other models we know have no explicit architecture for pain signals that bypass the hierarchy, with a theory of why the bypass must exist rather than merely be tolerated.^[13] Control theory, as we read it, has escalation without legitimacy; organizational design mostly has a suggestion box.

The concession is made in Chapter 4, and we do not re-argue it here. Beer says how a signal qualifies, by a statistical filter on points of stability whose criteria the metasystem sets. He did prototype one channel for people outside management, a citizens’ algedonic meter on which the holder signals anywhere between “total disquiet and total satisfaction” and “does not have to explain anything”; a prototype of ten linked meters was taken to Chile “for experimental purposes.”^[14] A citizen inside the national recursion is still not an affected non-member. The architecture provides no seat for the affected non-member, and no answer to who chooses the points of stability, who may contest the filter, or who audits it. That chapter was the prosecution. Here the credit comes with the concession attached, and both are true at once.

The maintenance cost is protected attention outside the reporting line. Someone has to specify thresholds, keep the bypass open when the line has every reason to close it, and check whether the people harmed by the system can reach the channel at all. None of that sustains itself, and a channel that exists on a diagram while going unheeded in practice is one of the pathologies the model itself predicts.

The composite, and what it does not settle

The composite is the argument. Each part is attested somewhere else. Nothing else we have found assembles all six. And the assembly is what a book that wants to move between a person and a planet actually requires. That is the whole of the claim, and it is narrower than the claims usually made on the model’s behalf.

Uniqueness buys less than it seems to. An assembly can be unique and still wrong. Six credits, each carrying a concession about evidence we have not gathered, add up to a distinctive model and fall short of a validated one. Distinctiveness is a claim about the space of available frameworks; it says nothing about the world. Nothing in this chapter shows that the five functions are necessary, that the recursion criterion is the right criterion, or that a generated pathology outperforms a collected one in a live organization.

Pricing that work is what Part Five exists for, and before we get there the next two chapters take up the transducer and the measurement of freedom. What we are entitled to say here is smaller and, we think, defensible: if the model went, anyone who wanted what the assembly does would have to reassemble these six from six different literatures, and we have not seen anyone attempt it.