Project Cybersyn and the Origins of the Big Data Nation (2014)
newyorker.com
newyorker.com
And that 7 people were going to be able to effectively manage the production of every sector of the economy?
It's not just a greedy algorithm like a market system to get the lowest costs for the first orders only.
To speak metaphorically, decentralized market regulation by supply and demand is the Bitcoin style of solving the problem. That is extremely inefficient, wasting lots of resources, being very slow but avoiding a single point of failure.
If you are surprised by the comparison, consider that you are avoiding the calculation problem with "the market regulating itself" phrase which falsely implies that this regulation would happen for free. It is equivalent to saying "Bitcoin approves transactions itself". That is true but there is still huge processing power employed to do so, as the huge mining farms show. In a free market calculations are performed by every individual actor in the market. In fact companies spent huge amounts of resources on market research to find out what and how much they should produce.
Central planning does naturally require far less processing power and can be more efficient (it has a much better best case but, to be fair, also a much worse worst case).
Even primitive, paper based, Soviet style planning has been proven to be quite adequate for early 20th century economies (don't be tempted to move the goalpost here, yes there were inefficiencies and it sometimes did perform worse that free markets, the point is it was already good enough to keep a country running for decades)
Now cybersyn was the right idea to go a step further. Make the central planning faster, more flexible and more democratic. To answer your question a little bit less simplistic, it depends. Obviously cybersyn was an early experiment and would surely not sufficed but would have Chile been better off with it than with the bloody CIA backed dictatorship that was put in power instead? Yeah, I am quite sure. We have now strong empirical evidence that neoliberalism does not work as promised.
Central planning works horribly and there's wide evidence of that.
> decentralized market regulation by supply and demand is the Bitcoin style of solving the problem.
This is word salad. The criticism of bitcoin - that I would absolutely agree with - is that it wastes resources to solve the same problem that a SQL database solves, just more inefficiently.
In order to solve the same problem a market solves you would need to take into account the preferences of millions of independent actors and mash them together into a Frankenstein solution, which is clearly beyond the capabilities of any government, with or without computers.
> Even primitive, paper based, Soviet style planning has been proven to be quite adequate for early 20th century economies
This claim is complete insanity. At no point in history the USSR has performed better than the first world in any sector, with the exception of a few propaganda stunts (like the space race) that came at costs so enormous that could only be borne by a brutal dictatorship, and that were promptly and easily matched and surpassed by the US.
Frankly, if your argument for central planning is to point at how well it worked out for the Soviets, I don't need to argue any further.
> it was already good enough to keep a country running for decades
Anything is good enough to keep a country running for decades, that's not a reasonable indicator of any kind of success whatsoever. The alternative to central planning is to stop doing it, not to go back to the 19th century.
> would have Chile been better off with it than with the bloody CIA backed dictatorship
A comment that contains a literal invitation to not move the goalposts ends with an irrelevant tangent. How ironic.
Yes, the OG 9/11 was bad, you get a gold star.
Cybersyn is interesting as an information system for the technology it employed, and is for sure a great display of ingenuity.
But come on, let's stop beating this dead horse. Central planning is a pipe dream.
That is like trying to model the exponential growth of laternfish[1] by emulating the behavior of each individual fish and then claiming it impossible to compute on consumer hardware. (The reference might sound like "word salad" if you have not participated in the last advent of code, it was just the first example that came to my mind. The point is, just because some naive approach at computing something is not realistic, does not prove the problem to be impossible to solve.)
For example here is a modern piece of software that can create a five years plan in reasonable time:
https://github.com/wc22m/NewHarmonywithJulia
But anyway, you missed the point of my "word salad". People have to obviously plan in a free market. I will not be successful if I just produce things blindly, or at least it will be pure luck. Would you say that success in the free market is purely based on luck? Maybe part of it, for sure, but obviously it is possible to plan ahead, to determine and model demand and so on. While overproduction is impossible to avoid in a free market economy, most companies at least hilt the general ballpark, they will produce a bit too much but not hundreds times too much or they wont stay in business. So obviously planning does happen by the individual actors, they do calculation. They do the work that individual miners in a bitcoin network do, hence the metaphor. The free market itself is a giant computation engine, that solves the problem of economic planning through the interactions of the individual actors of the economy. A very crude machine but it solves it. From this it follows that economic planning is obviously possible and can be solved.
> This claim is complete insanity. At no point in history the USSR has performed better than the first world in any sector, with the exception of a few propaganda stunts (like the space race) that came at costs so enormous that could only be borne by a brutal dictatorship, and that were promptly and easily matched and surpassed by the US.
The comparison is meaningless, as the Soviet Union and the US had vastly different starting conditions. If one country is economically stronger than another, it does not follow that the economic system is better or worse as many factors influence performance. If we run a 10k race against each other and you start from the beginning while I have a 5k lead before the run starts, it does not follow that I am the better runner.
The Soviet Union started as one of the least developed areas of this planet, literally still using the wooden plow and was devastated by two world wars and one civil war in a short period of time. It even being considered a competitor to the US is more that a miracle.
Could it be that your views on the efficiency of central planing are more based on prejudices that looking at actual empirical data? And again, we are looking at the most naive implementation of central planning. This is like judging the speed of garbage collected languages on some weekend project.
No. This is the issue. Markets solve a different problem. If I like A better than B and you like B better than A, this has to be taken into account for the eventual solution to be Pareto-optimal.
You would need to set prices for all of the hundreds of thousands of goods and services a modern economy produces, which requires access to an amount of information that no entity can possibly handle. It's not just computational power, it's a fundamentally intractable problem that is only solvable if you make wild assumptions.
All of that only to get a solution at most as good as the market equilibrium.
> The comparison is meaningless, as the Soviet Union and the US had vastly different starting conditions
Come on, this is a total charade. Don't want to compare it to the US? How about Japan? Or any other first-world country of your choice? Or how about West and East Berlin?
> Could it be that your views on the efficiency of central planing are more based on prejudices that looking at actual empirical data?
Yes, I have priors, as we all do. I believe my priors are correctly adjusted given the available data.
>You would need to set prices for all of the hundreds of thousands of goods and services a modern economy produces, which requires access to an amount of information that no entity can possibly handle. It's not just computational power, it's a fundamentally intractable problem that is only solvable if you make wild assumptions.
That just isn't how cybernetic planning was ever designed to work, so I agree with who you responded to that it's a case of only considering a naïve approach.
Almost all proposals have a consumer goods market - others have a functional equivalent. You let consumers compete to buy products, and in doing so you don't have to set the prices and predict what people like more - you just let them choose.
Markets under a capitalist system attempt to use prices on the other hand, by having the capitalist decide where to invest his capital for maximum return, to decide how things should be produced. This is a terribly inefficient solution computationally because you are going through hundreds of layers of indirection and noise, and actively hiding a lot of the relevant information. In doing so, most of the necessary information for production is constrained into low information price signals, which causes endless issues.
> Come on, this is a total charade. Don't want to compare it to the US? How about Japan? Or any other first-world country of your choice? Or how about West and East Berlin?
Firstly, you have to realize that the Soviets did not do cybernetic planning. They didn't continuously integrate realtime information or production and demand. They did their central planning in yearly increments using pen and paper. The USSR started off much worse than any first world country, even Japan
The fact that even though it was planned so poorly using such inefficient techniques and with such low responsitivity, and still managed to perform better than most of the world and compete with the best capitalist economies, is in my opinion a good signal that the advantages of central planning are considerable, even if with the horrible implementation of the Soviets, they were ultimately outbalanced by their issues.
Good luck with that! I am telling you, it was guaranteed to fail. We never saw it fail because it got dismantled.
People still fantasize about sitting on that control room and those unergonomic chairs, controlling the economy using 8 buttons at a time.
The programme was initially viewed with traditionalist scepticism by Allende’s Socialist Party. But Project Cybersyn’s hour arrived in October 1972 during a strike of 40,000 truck drivers led by the hard-right Confederación Nacional del Transporte. As Allende’s opponents sought to wreck the economy by preventing the transport of food and raw materials, Cybersyn was deployed to underwrite the resistance. Through the electronic network, the government was able to coordinate deliveries by active trucks and to evade blockades. “We felt that we were in the centre of the universe,” Espejo remarked.
After 24 days, the strike was defeated. Ministers then became “much more interested” in Cybersyn, Espejo told me. Allende even proposed transferring the operations room to La Moneda, the presidential palace.
On 10 September 1973, the government finally prepared to install an upgraded version of the system. But the following day, with the connivance of the CIA, Pinochet’s forces stormed the palace and bombed it from the air
From: https://www.newstatesman.com/culture/observations/2018/08/pr...
But if you could compare the performance of Cybersyn against a self-regulating free market economy, my instinct tellsm me that Cybersyn would underperform in comparison.
That being said, the free market "regulating itself" is pretty bad at it. First of all, it only has one cost function to optimize, which is never going to be what humans want to optimize. Then, it forces production information largely through low fidelity price signals. Finally, these signals are very much delayed, and by the time you answer the price signal, it may have changed wildly, which means that in some cases straightforward organisation that anyone with the relevant information could rapidly approximate can take years to come because the signal changes more rapidly and with considerable delay leading to dysfunction as other prices signals attempt to respond to it. There are many other issues with it.
Of course, an IBM System/360 and 7 people couldn't do it. It's just an attempt to see where the system had to improve. If it did work out it would have ended up with a staff of thousands of people, direct connections to much of the machinery and to people in the production lines, specially built supercomputers with linear and non-linear optimisations accelerators, etc..., while still allowing many markets where it made sense and use them as information.
As it was, a bunch of old crusty bureaucrats in the Soviet Union managed to get within the order of magnitude of the market regulating itself via supply and demand and free competition with data manually collated and corrupted every step of the way and refresh cycles on the order of months if not years, with the computation itself being in large part done manually. So it was an interesting and worthwhile experiment.
Directly as a result of it and the work Stafford Beer did on it, we got the whole field of system dynamics.
Also, Cybersyn was a pioneer in the idea of complex control systems with multiple humans in the loop.
Perhaps it would be more accurate to say, operational research applied to organizations?
https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...
[1] https://cidadeinseguranca.files.wordpress.com/2012/02/deleuz...
The CIA set up a special branch to study the Soviet cybernetics menace. It issued numerous reports, pointing out, among other strategic threats, the Soviet plans to build a 'Unified Information Net.' Based on CIA reports, in October 1962 President Kennedy's top aid wrote in an internal memo that the 'all-out Soviet commitment to cybernetics' would give the Soviets 'a tremendous advantage.' He warned that 'by 1970 the USSR may have a radically new production technology, involving total enterprises or complexes of industries, managed by closed loop, feedback control employing self-teaching computers.' If the American negligence of cybernetics continues, he concluded, 'we are finished.'
[1] https://balticworlds.com/the-cybernetics-scare-and-the-origi...
But this plan was not implemented, because the stalinist bureaucracy was afraid they can lose his privileges:
Glushkov's proposal faced opposition on two sides. Industrial managers and government bureaucrats opposed the computerization of economic planning and management because it exposed their inefficiency, reduced their power and control of information, and ultimately threatened to make them redundant. On the other hand, liberal economic reformers viewed Glushkov's proposal as a conservative attempt to further centralize the control of the economy and to suppress the autonomy of small economic units. A controversy erupted.
[2] https://web.mit.edu/slava/homepage/articles/Gerovitch-InterN...
There are modern thinkers working on this same line:
[3] https://en.wikipedia.org/wiki/Towards_a_New_Socialism
[4] https://www.amazon.com/Peoples-Republic-Walmart-Corporations...
As soon as a measure becomes a metric is ceases to become a good measurement of success.
The Soviet system failed both because the planners couldn't deal with modeling a complex economy, but also because they had no idea how much a given good cost to make, produce, or deliver to its point of sale.
It failed because all decision making was centralized, but no services were, so every enterprise had its own mix kindergartens, schools, recreation facilities, clinics, commissaries and dining facilities.
Any system that measured productivity of railways for example simply in ton-miles, with a quota of ton-miles to be met, I'd say is basically doomed to failure, particularly one with an inflexible plan governing it.
When the system worked, it was from local managers calling in favors and doing trades to keep production rolling.
A neat statistic, the planned economy of the SU, used as much or more oil per capita (I think twice as much) than as west, and produced 30% or more less in GDP, per capita. This is true of other consumable raw materials, all while spewing comparatively massive amounts of pollution out.
Cybernetics can improve on that somewhat by changing static plans into dynamic sets of rules for how to react, but the need to respond dynamically is pretty much inescapable and then it would seem better to lean into that.
I don't quite understand your point. Could you elaborate further? What do you mean by leaning into the dynamism?
E.g. I get the desire to control prices - it's a way of feeling that one has control over important social goals. If you genuinely want to address poverty, for example, it may feel very reassuring to be able to say that food for one person should be possible to buy for X amount. If you dictate that and you're able to enforce it, then ensuring everyone earns or get at least X for food plus whatever they need for other things means you've solved their access. In theory. As long as production actually meets sufficient levels. The problem is that you're providing no effective signal into the system of how important sub-goals of this is. On the contrary, if you set prices of products across the board, you're suppressing such signals. E.g. if the agricultural sector is in dire need of new machinery and fertiliser to meet production goals for food at a certain price but can't actually bid more to get it, they depend on convincing a whole lot of people that they actually need it more. And in practice we've seen that fail time and time again.
An alternative is to focus on attempting to control the final outputs you actually care about by bidding for the outcomes you want in an open market, and leaving your control problem to one of controlling your bids to get the best outcomes possible. You can "bid" either directly - by buying contracts for e.g. delivery of food at certain prices if that is what you want - or indirectly by e.g. adjusting welfare payments to compensate for prices outside of your desired levels, or a whole lot of similar mechanisms.
In that context cybernetic thinking can still have a place in terms of control loops to try to keep competing outputs within reasonable parameters based on as up-to-date information of factors that may affect prices as possible.
The point is that you get the best out of dynamic systems when you focus on controlling only the bits you actually care about well rather than imagining you know how it all hangs together and trying to control all the intermediate steps in detail because the latter requires you to keep up to date with every aspect of the most cost effective ways of meeting your targets. Not least, by focusing on controlling outputs you allow for redundancy in the methods used to achieve them that has a good shot at mitigating failures, as well as to avoid catastrophic results of conflicting priorities.
(EDIT: I'm involved in another discussion about typing in programming languages, and the paragraphs above actually fit into much of my thinking of static vs. dynamic typing as well as in much of life in general: A lot of problems are tempting to solve by exerting as much control as possible, but in dynamic, fast changing system expending your effort on exerting the control only over the things you actually care about allows for a more flexible overall system)
Even for a socialist system there's absolutely no ideological reasons for requiring central planning. The notion of the need to do away with market mechanisms as planning signals in favour of a central, state controlled plan, was largely a Bolshevik conceit that a lot of other socialist ideologies strongly disagreed with (not least because many of them want to dismantle the state, leaving no organ with power to enforce central planning). But at the time there was also a severely lacking understanding of the inherent complexity of large scale planning mixed with a rather unjustified optimism about the ability of humanity to overcome them.
EDIT: Allende represented a mid-point - nowhere near going as far as the Bolsheviks, and Cybersyn addressed a lot of problems in introducing feedback loops rather than strict central planning, to be clear.
The post-Stalin dogma as well as the calcification of the Soviet economy was an impediment to the trials of such planning, but some Soviet economists like Kantorovich in fact proposed essentially the same system. It was refused for two reasons - large computational cost at a time where computers were quite weak, and cynical Soviet bureaucrats not wanting to lose their power, hence many opposed it. But quite a few were indeed supportive.
Later in the 70s, the Soviets did adopt some market economic implements in the form of cooperatives, so I don't think they were unconditionally opposed to it.
Later, there would be the issue that any planning system would be opposed not only by pro-planning bureaucrats, but also by the rising anti-communist faction of the party.
Interestingly, I think it is Lenin that most acutely understood the difficulty of central planning, hence why he proposed the NEP that Stalin later dismantled.
On the rest however, I think we are both in agreement :)
I don't think I'd agree Lenin was all that favourable to markets initially though. It seemed to me to be more of a pragmatic attitude and acceptance that things couldn't continue the way they did prior to NEP, and it being short enough after the civil war created a very convenient way of drawing a line under the old policies as driven by the war.
Stalins role in dismantling it is certainly critical.
So instead of having a quota of ton-miles, you will simply have a bunch of goods on sale. If a lot of people are buying them, you can increase their prices and at the same time increase production.
Once you compute how best to minimize the difference between demand and supply, you compute what goods have to be produced, and work backwards from there until you can calculate what inputs they need.
There are a few proposed metric. Essentially the idea is to completely do away with any quotas outside of exceptional circumstances, and respond to demand dynamically in a way to reduce the amounts of hours worked and increase the consumer satisfaction, as measured by their comparative subjective value for various goods.
We both agree that the quota system and the inefficiency of the Soviet system was considerable. However, another interesting statistic you may not be aware of is that, outside of times of war and for the last few months, Soviet GDP per capita never decreased, while still providing fairly acceptable living standards, and good ones by global standards.
The fact that it was actually not really doomed to failure at all and it's comparative performance compared to the rest of the world, as well as that it lasted quite a while shows that there is enough value in the good sides of the historical centrally planned economies to partly make up for the atrocious and seemingly obviously debilitating issues.
The idea of cybernetic central planning is to reduce stress, offer more social safety, reduce economic inequalities, all while doing away with quotas, inflexible plans, precomputed prices, inflexibility to demand, and sensitivity to fraudulent information.
Basically, give up on trying to guess what to make and in what quantities, and let the consumer decide what they prefer, but then intelligently use that information to determine how to organize production as well as possible, and then measure the results of these organizations in real-time to provide feedback and adjust various processes.
An interesting way to look at it is that, in capitalism, Capital has a mind of it's own and basically acts as a control system. In essence, in capitalism, capital wants to grow, and it does so by estimating all of the returns it could be generating, and if it is credible to increase returns, re-allocate itself to the most profitable return. So, through a few dozens of layers of price signals, it can control the economy to increase it's own profit, and in doing itself it also increases productions and allows capital pledged to consumption to have goods and service. Cybernetic planning intends to directly measure the price signals as close to the source as possible, and when possible even underlying signals before that, and use those to optimize some function that will allow people to work as little as possible but have as many goods and services as possible. The idea is that, as with the control systems humans program, we deal with the lower complexity of the model (at least for the foreseeable future) with more direct access and dynamic access to the data and to the levers of action.
The market is a constant failure at the only thing it should do, properly distributing resources. It's failing clamorously to distribute covid vaccines globally, for example. Additionally, it is easily exploitable everytime there is any unbalance, see bargainers.
The lack of control you seem to present as an advantage is exactly what makes it controllable by a very small minority of powerful entities.
Actually I start to see some parallels with the theoretical "lack of control" that is claimed to be the advantage of cryptocurrencies by enthusiasts.
A market is ultimately nothing more than the iterative application of a function that takes a list of buy and sell orders and outputs matches. It can be entirely unregulated or however regulated you want. There can be no constraints on who accesses it, or lots. There can be synthetic manipulation on one or both sides to adjust the functioning of the market, or not.
I'd argue for all kinds of regulations - I don't think unregulated markets are a good thing at all. But a well regulated market is simply a consensus mechanism that reduces what is a fundamentally computationally intractable control problem if you try to handle it in a centralised manner by dividing it into a range of smaller problems and distribution the effort of figuring out the best way of meetings goals. It's never going to give an optimal outcome, but you can mitigate the risk that it will produce really bad outcomes, and get closer to decent results.
I'm also not arguing for one market, but for many, I favour using market mechanisms to price externalities whenever possible.
The current global market is not set up to "properly distributing resources" because that is not the goal of its participants, so you have a good argument against current global markets for products. If you were to create a market actually intended to fairly distribute scarce healthcare supplies and it would look very different, with pricing based on e.g. predicted outcomes and benefits for different bidders.
You can still plan in the face of well regulated markets, but the planning process would involve packaging up incentives and regulating the pricing of externalities without trying to guess what will work best in ways that forces the planners to become experts in everything.
> The lack of control you seem to present as an advantage is exactly what makes it controllable by a very small minority of powerful entities.
Even if we were to agree this was unavoidable (I'd agree it's unavoidable in an unregulated market with vastly different access to resources, but that is just one subset of possible markets), a planned system is potentially controlled by an even smaller entity of even more powerful entities. We've seen that fail time and time again, because of the hubris of people with power thinking they knew best. Decentralising power is if nothing else a risk mitigation, and the failure of current markets to do so well enough is not an argument for a mechanism with a distinctly worse track record.
The so called 'cybernetic factory' at the heart of the project was about making real time information exchange between several levels of hierarchy possible, with much delegated to the factories themselves. Centralized decision-making was reserved for high-level planning akin to Auftragstaktik in the German army (I believe it's called mission command in the UK and US).
Friedrich Hayek of all people, who met Beer at a conference in 1960 in Illinois was actually very sympathetic to Cybersyn. The frequent comparisons between Cybersyn and Soviet Planning are pretty much the result of a political campaign against Allende, it doesn't have much to do with the scientific ideas behind the project.
If you have heard people talk about 'smart factories', 'industry 4.0', IoT devices gathering data in real time and relaying them to big computers where they can be analysed and quickly sent back to the workers you have in reality stumbled upon people who are trying to build CyberSyn 2.0, probably without knowing it and less socialist branding.
Cybernetic Revolutionaries: Technology and Politics in Allende's Chile by Eden Medina is a good, in depth book on the topic.
Ironically, it's the tankies (neostalinists) who seem to be the loudest defenders of it. If it's propaganda it seemingly worked on the wrong audience.
On the other hand, cybersyn had no quotas, and no long term rigid plans. Informational exchange did not have to go through bureaucrats, and could be done directly from the factory, eventually automatically by the equipment, and update a digital model that would do a large part of what the bureaucrats at GOSPLAN would do, but automatically and verifiably.
So it seems very, very different to me.
Not uncommon though, right? The tankies have also adopted the narrative of China as a Communist threat, or in Latin America you end up with the paradoxical situation of most opposition parties in Venezuela belonging to the Socialist International, yet you find feverish and uncritical Maduro support among young radicalized left-wing Americans.
We have the same phenomenon in Germany where ironically the formerly West-German members of Die Linke have a much more cartoonish, ideological view compared to their actually ex-socialist former East-German counterparts.