Last update: 01 Aug 2026 00:32
First version: 30 January 2022
Initial notes, January 2022
An early attempt at using networked computers for economic management in Allende's Chile, with the involvement of the British cyberneticist Stafford Beer. This has something of a cult following among contemporary socialists, in no small part, I suspect, because of the period glamour of the photographs of the control room, and because of the aura of righteous martyrdom given the fate of Allende and his government. Considering my interests in the possibilities and limits of economic planning, however, what I want to get very clear on are:
- What the various participants (Beer, the various groups among the Chileans) hoped to achieve with Cybersyn;
- What the system as implemented actually achieved; and
- What a similar system might do with modern, or reasonably-foreseeable, technology.
The main source on all this is Medina's book, which I need to actually finish. (Honestly it's been so long since I started it that I should just re-read from scratch.) But I should also try to see what's been done, in terms of historical research, since her book.
Update, 18 October 2023
Attention conservation notice: 1500+ words of book report.
Having just finished Medina's book (which seems to have no successors), and the papers from the 1970s she cites as the technical sources, a few more notes. (All page numbers are references to her book.)
Cybersyn was to have four main components:
- "Cybernet": A network of telex machines linking factories to the government agency that ran the nationalized sector of the economy (Corporación de Fomento de la Producción, CORFO), and thence to the one (!) mainframe computer available to the project. (My reference to networked computer_s_, plural, in the opening paragraph to this notebook was thus dead wrong, though I think it's a common misunderstanding.) The idea was that factories would send regular (ideally, daily) measurements of what we'd now call key performance indicators or metrics to the central computer, which would process them and communicate back to each factory what it had learned.
- "Cyberstride": A central program running on that mainframe which was essentially doing anomaly / change-point detection on the time series coming in from the periphery . This was basically implementing the method of Harrison and Stevens (1971), per Medina (p. 267n33). As Dan Davies puts it, the content of the signals it output would have amounted to basically either "situation nominal" or "there's something up at the mill".
An important part of the design here was that when Cyberstride did raise a warning, it was supposed to go back to the relevant factory, which would get a chance to deal with the matter on its own, thus preserving a certain measure of firm-level autonomy.
- "CHECO": A simulation model of the Chilean macroeconomy, which was supposed to let policy-makers do what-if exercises.
- The fabulous control room or operations room, which was supposed to display information from Cyberstride and CHECO to decision-makers. Medina says (pp. 121, 123) that the designers of the room swear up and down they weren't influenced by 2001 or Star Trek or the like. If that's true, the Chileans and the Hollywood set-designers must've both drawn drawn on some common sources for their visual style of The Future.
(Beer was also very taken with his thoughts about "algedonic" [=pain-pleasure] meters which The People could twist back and forth to indicate how satisfied or dis-satisfied they were, with a central read-out, but that was, if not literally vaporware because a handful of prototypes were built, then very clearly never going to be a thing.)
I have listed the four parts in order of decreasing completion and utility.
- The telex network was the only piece that seems to have been actually useful to the Allende administration --- and that not in the way intended. In October 1972, the mostly-conservative, mostly-small-business-owner trucking industry staged a nation-wide strike against Allende. The administration used the telex network to coordinate the trucks they had control, to try to keep the economy from completely grinding to a halt. This coordination seems to have made no use at all of Cyberstride, or the control room, or any cybernetic principles. The main advantage of the telex over phone calls was creating written records without the need for note-taking. (Telegrams would have worked as well!) Some tertiary sources make it sound like the government broke the strike in this way. In fact, as Medina makes clear, the strike ended with a political compromise, viz., Allende brought top generals into his cabinet, and otherwise temporarily appeased the conservatives, though without fully abandoning his program. It does seem that the telex network bought the government more time and a better bargaining position than it would otherwise have had.
- Cyberstride was eventually brought up and running, but the lag time between taking measurements, running them through the mainframe, and getting them back to decision makers was so long that it seems to have been a complete flop in its intended purpose of detecting problems early before they became serious --- let alone anticipating them before they became problems. (Medina refers to efforts to get the telex machines to directly communicate with the mainframe running Cyberstride, but it's not clear to me if those ever succeeded; I/O in the early 1970s was hard!) This, of course, did not help persuade busy, not to say frantic, factory managers to devote time and energy to feeding the system measurements early and often. (I say "managers" because ordinary workers were uninvolved.) It was, of course, an entirely centralized system in terms of computation. (How could it not be, with only one computer?) Rhetoric to the contrary, it did nothing to de-centralize control, or to involve workers in participatory decision-making. It also was in no sense a planning system, or any kind of replacement for market coordination. When the system did warn of problems, it was up to managers and central bureaucrats to scramble to find solutions (e.g., alternative sources of supplies). Figuring out what variables to measure for each factory was complicated, and involved sending trained engineers out to each plant and understanding what was going on there; this was not something workers had much to do with. (Beer talked a good game to the contrary on that last point, but it was just talk.)
I mentioned above that part of the original design was that when Cyberstride identified a problem at a plant, it was supposed to get some time to address it on its own, in order to preserve the autonomy of the enterprise. As Medina explains, however, when problems were noticed, the staff running the system "alerted the affected enterprise, those in the central telex room in CORFO, and [Cybersyn project director Raul] Espejo in the CORFO informatics directorate --- all at the same time". "Dismantling one of the primary safeguards of [enterprise] autonomy might have been as easy as having someone from the telex room walk down the hall". (All these quotations are from pp. 183--184.)
- CHECO produced some models, but at no point does Medina refer to any decision-maker actually consulting them. From her description of the models, it would have been extremely hard to connect them to the kind of data coming in through the telex network.
- The operations room also doesn't seem to have been much use. The screens were not, in fact, hooked up to computers; they were for displaying slides. "[I]t required some of Chile's best graphic designers to draw by hand every graph and chart the room displayed" (p. 125). I'm sure it was nicer than a dingy conference room with an overhead transparency projector, but it wasn't actually any more capable. Towards the end of his administration, in September 1973, Allende did ask to have the room moved to the presidential palace, but that seems to have been because he wanted closer access to the central node of the telex network (p. 206).
My assessment, based on all this, was that if the Allende government had, by some miracle, survived (*), and Cybersyn had been built out as intended, what would have resulted would've been an pioneering example of what we'd now call a "dashboard", tracking time series of performance indicators and throwing alerts to possible change-points in the series. This can be a useful thing for decision-makers, if the right stuff is being measured and they have some ability to act on the information, but the politics of that of course depends entirely on who has access to the information, who gets to make decisions on the basis of the information, what kinds of decisions they get to make, who they are accountable to for the results of their decisions, etc., etc. For that matter it depends on the information being entered into the system honestly in the first place, and not fudged to conceal problems, to exaggerate distress, or to set easier goals for oneself. (Medina doesn't mention this issue at all, so it might not have occurred to anyone, but it would have mattered if Cybersyn had actually become important.) In any case, as a replacement for market coordination, Cybersyn was simply a non-starter. To describe it as a decentralized planning system is nonsense.
Nowadays, of course, the software would run easily on anyone's phone. It'd be easy to give each factory manager their own anomaly-detector, and the telex network would be subsumed into the ordinary phone network. Why you'd want to share the information rather than having anomaly detection done locally is, with modern technology, less clear --- presumably it'd be because someone with access to all the local information could do some useful aggregation, perhaps by assimilating it into a macroeconomic model. (For that to be useful you'd need a good macro model, which are thin on the ground; maybe an input-output model of inter-plant/inter-industry linkages would be useful enough.) You could run that central node out of a spiffy room, where the screens could even be connected to computers.
I do not want to end on a dismissive note. The people who worked on Project Cybersyn tried to do something new and hard and worthwhile under difficult conditions. What they achieved was remarkable enough to need no exaggeration.
*: And I don't see how it could have, with its policies; if the CIA and/or domestic reactionaries didn't overthrow them, the Communist Party would have. (Cf. Nove.)
Recommended (in order):* Daniel \( D^2 \) Davies
* "Computers with ashtrays", Back of Mind, 26 July 2023
* "The statistical heart of the system", Back of Mind, 30 August 2023
- Eden Medina, Cybernetic Revolutionaries: Technology and Politics in Allende's Chile [In addition to my mini-review, see Alex Harrowell]
- Raul Espejo, "Cybersyn, big data, variety engineering and governance", AI and Society (2022) [End-of-career reflections by one of the Chilean project leaders]
- P. J. Harrison and C. F. Stevens, "A Bayesian Approach to Short-term Forecasting", Operational Research Quarterly 22 (1971): 341--362
- Alec Nove, "The Political Economy of the Allende Regime", pp. 51--78 in P. O'Brien (ed.), Allende's Chile (New York: Praeger, 1976), reprinted on pp. 3--29 of Nove's Socialism, Economics and Development (London: Allen and Unwin, 1986)
- Gene Castellano, "The DuPont Company's Chart Room", Hagley Museum and Library weblog, 18 December 2015
To read:* Stafford Beer [Need to get these from the university library, in some cases from storage...]
* Brain of the Firm, 2nd edition
* Cybernetics and Management
* Designing Freedom
Instagram Engineering
Shopify Engineering