The Loop That Stayed an Intention

Project Cybersyn promised a line running from worker participation to government action. Beer’s own account documents that line as an intention and a procedure. The channels actually built ran one way, which is why the missing loop looks designed out rather than accidentally omitted.

The promise and the documented design

Chile is our founding case, and we use it for something specific. When the Viable System Model is asked to show that it is more than a management diagram, that it belongs to a politics as well as to a control theory, we point to Santiago: a national economy, a government that asked for the work, a designer who described what he was building as an instrument of the workers’ own self-management. The case carries a great deal of argumentative weight for us. Most of us have never gone back to check what the design specified, or what Beer himself reported about how it ran.

The channels that were specified run in one direction. Indices are drawn from the enterprises and filtered as they rise. Forecasts are computed centrally, although Beer’s design promised each level its own tools and its own computer.^[1] Exceptions are raised to a level above the one that produced them, and the government, at the top of the structure, can act on what arrives.^[2] Each of those channels is described with care. The channel that would run the other way is described differently, as a procedure and an intention.^[3] The teams modeling each plant were to build the model with the people who worked there and agree the performance measures with them, and Beer records that “this was not always being done in the intended spirit.”^[4] The algedonic meter, the one channel from the people to the government, had a wall reserved for it in the operations room: “there remained a blank wall in the Operations Room: it was intended to house an algedonic meter.” We read Beer’s sentence as saying that the meter was never installed.^[5] What never existed is a route with standing, by which the workers the design names as its client could contest an index, revise the model of their own enterprise against the modelers’ view, or limit what the government does with the reports.

We cannot treat that as an omission awaiting a later draft. Everything adjacent to it was built. A design in which the client’s return channel is the one element left as a procedure and an intention tells us what it takes the client to be: a source of measurements and a partner in modeling, without a part in the decision. That is the missing loop, and it is our problem before it is anyone else’s, because we have taught the case for fifty years as though the intention had been the channel.

Twenty-seven pages, and a tape

In 1981 Werner Ulrich publishes twenty-seven pages in the Journal of Applied Systems Analysis. They put to Project Cybersyn five questions C. West Churchman had formulated for any systems study: who is the client, what is the purpose, who is the decision maker, who is the expert, and who is the guarantor.

He is careful about what he is not doing. “It is not my intent,” he writes, “to question the social and democratic ideals that both Salvador Allende and Stafford Beer associated with the project.” He revises the paper after comments from Fernando Flores, Allende’s Secretary of Economics. An anonymous reviewer who worked on the project confirms its accuracy.

His conclusion is one sentence: “Cybersyn’s alleged goal of devolving power to workers is not confirmed by its design.”

Beer sent his comments on tape. He could not agree with the critique. Nothing from the tape was published, and the tape, if it survives, has not been found in the catalogue of Beer’s papers. Two years later Beer published a short reply in the same journal, which we have not yet read.^[6]

Apart from Beer’s own five pages, that is the whole of this field’s recorded reply to Ulrich’s structural argument.^[7]

Beer did answer, first in the medium he chose, as the acknowledgements record, and then briefly in print. What is missing is anything the field can read. A critique built with the cooperation of the people who ran the project, published in a systems journal, addressed to the founder of the model and to the case we cite most often, has gone forty-five years without a reply in print from any of us. The omission belongs to us, not to Beer.

Beer did engage the political content of the charge in print, in the same year and without naming Ulrich. The second edition of Brain of the Firm concedes that the lapse from participatory modeling “seemed likely” to have “aided detractors who later wished to call Cybersyn ‘technocratic’,” describes a current inside the project that was “technocratically minded,” and calls some of the hopes for handing the tools to workers’ groups “manifestly starry-eyed.”^[8] That is a concession. It does not answer Ulrich’s argument about what the channels were wired to do.

What the critique actually says

Ulrich did not write from a distance. “A Critique of Pure Cybernetic Reason: The Chilean Experience with Cybernetics” is a case study of Project Cybersyn built on Beer’s own reports, on Fernando Flores’s account, and on an insider reviewer who confirmed its accuracy.^[9] The finding is not that Cybersyn failed. It is that the design does not do what its designer said it did. The client Beer declared was the workers. The client the design serves is the government, which alone can act on the system’s reports without a homeostat above it, change the measure of performance, and control what goes in. Beer’s own recursion supports the reading: it names the government as the viable system in focus, “System Five being the President of the Republic.” His principle that “System Five within a people’s government cannot be an élite ruling class: it is somehow the embodiment of the mass of the people themselves” names no mechanism.^[10] Autonomy in that architecture holds only while things run smoothly, because an algedonic↗ signal is a transfer of control upward as well as a cry of pain. Hence the sentence in the paper that is hardest to answer, that “what distinguishes Cybersyn from a traditional state bureaucracy is not a less hierarchical structure but the greater speed at which vertical communication (and control) takes place.” Part Four is built to answer that objection, made in 1981 and aimed at the founder.

Two things follow for the rest of this book. The objection is structural rather than historical: it would stand even if the project had run for a decade, because it concerns what the channels are wired to do. And it lands on the part of the model we have specified least. Chapter 4, Channels Without Owners, sets out how little the algedonic channel has been given, beyond Beer’s statistical thresholds, in the way of owners, addressees or rights of reply; Ulrich’s paper is the earliest place we know of where that gap is treated as a political fact rather than an engineering detail. Chapter 16 states his argument in full, together with Hayek’s and Cooper’s.

The strongest defense available

The best defense is that Cybersyn was never built to compute a plan. It attenuated. It filtered indices, forecast statistically, reported exceptions, and was meant to keep an alert with the unit for a defined interval so that the unit could absorb its own disturbance.^[11] That is a Hayekian architecture:^[12] leave the decision to the person on the spot and escalate only the residue. Beer called his public lectures Designing Freedom, and meant it.^[13]

We take that defense seriously, and quoting the phrase “cybernetic management” back at it does not dispose of it. On this reading the design is a subsidiarity machine. The interval before escalation is a right to be left alone for a stated period, and the filters protect the enterprise from the center’s attention instead of delivering the enterprise to it. Chapter 26 takes up what Beer meant by freedom and whether it can be stated as a quantity. If the reading is right, the missing loop does less damage than it appears to, because the architecture would concede to each unit the decisions that matter most to it.

What the record cannot settle

The historical record neither confirms nor refutes the defense, because the system never ran at a scale that could have tested it. The most damaging criticism is empirical. Cosma Shalizi’s verdict is severe, and we accept its force where the evidence supports it. To describe Cybersyn as a decentralized planning system “is nonsense”, since it was computationally centralized on a single mainframe. Time lags made its alerting close to useless for early detection. Its safeguards for enterprise autonomy could, on Medina’s account, have been undone as easily as by someone from the telex room walking down the hall. The only part that demonstrably worked was the telex network, which proved its worth coordinating trucks during the October 1972 strike, as logistics rather than as cybernetic management. At best it would have been a dashboard.^[14]

Beer’s account says otherwise on both points. He describes the strike as a test of filtration, in which “some signals were algedonic, requiring instant decisions, while others could be attenuated,” and he puts the coverage at “something like seventy percent of the socio-industrial economy,” about four hundred enterprises.^[15] These are the founder’s claims, made in an account he wrote in the first person and as a partisan. Medina’s figures for the scale at which the system ran differ.

What that licenses is limited. It does not show that attenuation is the wrong architecture, or that the Hayekian reading of the design is mistaken. It shows that the founding case supplies no evidence for either, and that the safeguard on which the whole defense rests, the interval Beer’s design gave a unit before CORFO intervened, which in practice ran while the alert had already reached CORFO, was on the available record a convention rather than a constraint. If a staff member could remove a protection by walking down a corridor, the design does not enforce it. Ulrich reached the same finding from the other side: the client’s position in the architecture depends on the goodwill of the party above it.

We have kept the case in our teaching for fifty years without either finding attached to it. Chapter 9 asks why that was so easy, and its answer does not flatter us: nothing in our practice made the omission expensive.

The boundary condition

Ostrom’s seventh principle supplies the boundary condition.^[16] No amount of internal viability survives the withdrawal of the right to self-organize. That right is a precondition, prior to any design choice. The model places external authority in the containing recursion and in its command and legal channels, and Beer knew that such authority could dispose of a unit: his example is the division chairman “secretly moving to sell his division.”^[17] The model does not treat the revocation of that embedding as a variable with its own regulation. September 1973 does not test the VSM’s performance; it demonstrates that an externally revocable right to organize decides whether any internal design gets to operate. The Shalizi material, likewise, bears on what the founding case documents and not on a performance test the project never reached.

We state this with the restraint the events require. Nothing in the model is answerable for what happened in Chile, and we make no claim about it here beyond the structural one: a system whose right to exist is held by another party has a variable outside its own regulation, and our diagrams do not represent it. Chapter 14 sets out what recognizing external authority as a design variable would require of us, and what evidence we would have to gather to say anything about it beyond assertion.

What would settle it

Receipt — find the tape. Read Beer’s 1983 reply first. Then search Stafford Beer’s papers at Liverpool John Moores University and Werner Ulrich’s papers for the oral reply, recording catalog scope, search terms and date. Price: archive fees and one researcher-week. If the tape cannot be found, publish the reply the field owed Ulrich: reconstruct Beer’s strongest answer from the contemporaneous record and submit it with Ulrich’s unanswered claims stated in full. Forty-five years after 1981, the silence is itself part of the evidence, and it is ours to account for. We return to what it has cost us in the Coda.